University of Southern California Dissertations and Theses

Reasoning with uncertainty and epistemic modals

(USC Thesis Other)

Reasoning with uncertainty and epistemic modals

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Reasoning with Uncertainty and Epistemic Modals
by
Benjamin Lennertz
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PHILOSOPHY)
May 2014
Acknowledgments
The financial, administrative, academic, social, and emotional support of many people
and organizations have been integral in the writing of this dissertation. Thanks to the USC
Provost’s Office for many years of financial support in the form of a Provost’s Fellowship and
to the Graduate School for the Grayson and Judith Manning Fellowship and the Dissertation
Completion Fellowship. Thanks also to Ralph and Francine Flewelling for multiple years of
summer support. And thanks to the USC Philosophy Department and the Graduate Student
Government for support for travel to many academic conferences.
Thanks to the former Philosophy Department administrative assistants Tomiko
Higuchi and Valerie Hunt. And special thanks to current administrative assistants Cynthia
Lugo, Corey Resnick, and Barrington Smith-Seetachitt for all of their help (also, Cynthia, go
Red Sox!).
I owe a debt to the Philosophy and Mathematics Departments at Vassar College.
Many of my professors there helped start me on the path that led to this dissertation. Thanks,
in particular, to my undergraduate philosophy advisor, Barry Lam, for his help at Vassar and
his continued interest in and advice on my work.
The entire USC faculty deserves my thanks. Besides my committee members, thank
you are owed to Andrew Bacon, Kenny Easwaran, John Hawthorne, Robin Jeshion, Shieva
Kleinschmidt, Janet Levin, Jacob Ross, Gabriel Uzquiano-Cruz, James van Cleve, Kadri
Vivhelin, and Ralph Wedgwood. Whether through comments on my work, private
discussions, or classes, these folks have had a positive impact on this dissertation.
Many specialists from the epistemic modal literature (and from institutions other than
USC) have given their time to comment on my papers and/or chat with me about these topics.
Thanks to Gunnar Björnsson, David Braun, Andy Egan, Sarah Moss, Eric Swanson, Tamina
Stephenson, Malte Willer, and Seth Yalcin. Particular thanks are due to Kai von Fintel and
Anthony Gillies, who commented on my paper at the 2011 Pacific APA, and to Janice Dowell,
who commented on my paper at the 2013 Pacific APA. Both papers were criticisms of work
by these authors, and I appreciated how they responded with grace and were generous with
their time.
My fellow graduate students at USC deserve recognition for their academic help, their
emotional support, and their friendship. In particular, I want to thank the following people
i
who have conversed with me about issues related to this dissertation: Greg Ackerman, Matt
Babb, Brian Blackwell, Justin Dallmann, Keith Hall, Matt Lutz, Kenny Pearce, Caleb Perl,
Johannes “the J-man” Schmitt, Julia Staffel, Ryan Walsh, Aness Webster, and Jonathan
Wright (Eliot Michaelson, who was a grad student at UCLA also belongs on this list). Lewis
Powell was my roommate during my first year in Los Angeles and was always a great model of
an engaged grad student and a top notch philosopher. I appreciate the interest Shyam Nair
has taken in much of my work. He has helped it improve by leaps and bounds. It has been a
pleasure being around Indrek Reiland for the past six years and growing into the philosophy
of language community with him (especially during those summers in Budapest and Cerisy).
I’ve probably spent more time talking about philosophy with Justin Snedegar than anyone
else. Thanks to him for being a constant sounding board for ideas both philosophical and
other (And thanks to Emmy for enduring our blathering and for feeding us).
I have had great support from my committee. Barry Schein gave me my introduction
to intensional semantics from a linguistics perspective and has offered valuable feedback on
my project. Karen Lewis was part of my qualifying committee, but, unfortunately for USC,
moved to Barnard a year later. During her year here, she made a big impact on this
dissertation, and I thank her for her continued interest (Karen, we will finish that paper we’ve
been writing one day!). Steve Finlay’s views on modals have been an inspiration for my own.
Steve has a nice way of giving like a hundred really helpful, but critical comments on a paper
while still conveying that he is excited about the project and its direction. My dissertation co-
chairs are both excellent advisors, and I am incredibly lucky to work with both. They must be
the two favorites in the competition for quickest average reply time to a student email. They
have been supportive both academically – in my scholarly pursuits – and emotionally – being
sensitive to my trials as a graduate student and beginning scholar. Scott Soames’s
philosophical impact on me started right away at USC. I wrote a paper for Scott every week of
my first year of grad school. He has continued mentor me throughout my graduate career
and, especially, in thinking about these topics. His guidance changed me from being a
philosophically inclined novice to a serious professional philosopher. Mark Schroeder is a
crazy person, spending hours and hours each week meeting with students. I want to thank
him for allowing me to be one of those students he spends tons of time with and for leading
me toward the topic of epistemic modals. Mark has taught me many things – among them the
fruitfulness of thinking big-picture and the importance of framing a piece of writing to allow
ii
my arguments and ideas shine forth (though my mastery of these insights is still a work in
progress). Sweet.
A big thanks goes out to my friends – especially those who lived in the LA area during
the time I was at USC. They have been supportive and have understood the times I was busy
with school when I’d rather be hanging out with them. Many people think their family is the
best, but they are wrong. My family is. It has been great getting to live near my siblings for
the past six years and witness their blossoming families. Thanks to my brother, Chris, for
being a model of how talent and hard work can lead to success. My sister, Ange, has been one
of my best friends and a constant supporter throughout my life. Thanks to my dad for always
being interested in my work (we’ll co-write that paper about metaphor in the law one day!).
Though I’ve ended up on a path very much like his, he always let me know he’d be proud of
me whatever I did, as long as I was happy. My mom is the most caring and selfless person I
know, and I'm fortunate that so much of her care has been for me. I don’t know how to
express just how important her support has been. Actually, my mom might only be tied for
the most caring person I know. There is also Hannah, who has been my companion
throughout my work on this dissertation. Her support has helped me through many tough
times, and her presence has made my life, while writing this dissertation, so much better.
iii
Table of Contents
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .i
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Chapter One – Uses of ‘Might’ and Theories of Direct Speech Acts . . . . . . . . . . . . . . . 1
§1 Uses of ‘Might’ and Responses to Uses of ‘Might’ . . . . . . . . . . . . . . . . . . . . . .1
§2 Theoretical Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
§3 Contextualism and The Direct Approach . . . . . . . . . . . . . . . . . . . . . . . . . . 7
§4 Relativism and The Direct Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
§5 The Conceptual Problem for Relativism . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
§6 Expressivism and The Direct Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
§7 The Frege-Geach Problem for Expressivism . . . . . . . . . . . . . . . . . . . . . . . . 21
§8 Taking Stock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Chapter Two – More Liberal Theories of At-Issue Speech Acts . . . . . . . . . . . . . . . . . . 30
§1 Re-evaluating Contextualism and The Direct Approach . . . . . . . . . . . . . . 30
§2 The Implicature Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
§3 The Hearer’s Context Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
§4 The Prejacent Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
§5 The Multiple MP Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .45
§6 Lessons Learned . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Chapter Three – Reasoning with Uncertainty: Floats and Sinks . . . . . . . . . . . . . . . . . . .51
§1 Reasoning with Uncertainty and Floats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
§2 Sinks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
§3 Norms on Reasoning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
§4 Conditional Floats and Sinks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
§5 Quantificational Floats and Sinks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .61
§6 Floats of Different Strengths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
§7 Epistemic Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .69
§8 Disagreement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
§9 The Plan Going Forward . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
iv
Chapter Four – How Simple ‘Might’-Communication Works . . . . . . . . . . . . . . . . . . . . 81
§1 Expression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
§2 Expression of a Float as an Indirect Speech Act . . . . . . . . . . . . . . . . . . . . . . . 85
§3 Expression of a Float as a Standardized Indirect Speech Act . . . . . . . . . . . . .87
§4 Applying the Theory to the Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Chapter Five – Complex ‘Might’-Sentences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
§1 Methodology and Compositionality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
§2 Conditional ‘Might’-Sentences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
§3 Disjunctive ‘Might’-Sentences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
§4 Quantificational ‘Might’-Sentences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .105
§5 Negated ‘Might’-Sentences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
§6 Attitude Verbs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
§7 Antecedents of Conditionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
§8 Tense . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
Chapter Six – Other Strengths of Modality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .130
§1 ‘Must’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
§2 Probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
§3 How ‘Should’ Differs from ‘Likely’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
§4 Embedding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .143
Chapter Seven – Floats, Credences, and Acceptance in an Inquiry . . . . . . . . . . . . . . . . 146
§1 Reduction to Credences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .146
§2 An Analogy to Acceptance and Belief . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
§3 Floats as Partial Versions of Acceptance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
§4 Floats as Practical Attitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
§5 Other Float-like Attitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
§6 Epistemic Evaluation and Disagreement Revisited . . . . . . . . . . . . . . . . . . . . .168
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .174
v
Introduction
This dissertation is an exploration into how agents reason and communicate when they
are uncertain. In it, I develop a descriptive and normative theory of explicit reasoning in
situations of uncertainty, which includes an account of the attitudes involved in such
reasoning. I use this theory in creating an integrated picture of epistemic language – language
involving words like ‘might’, ‘must’, and ‘likely’ – which goes beyond claims about semantics to
characterize its general communicative impact.
The picture that I develop looks, in broad outline, as follows: In explicit reasoning in a
situation of uncertainty, an agent takes some privileged propositions as serious options. So, to
understand such reasoning, we can’t focus merely on the propositions that she believes to be
consistent with her information or those in which she has some positive credence. Instead, we
need a theory of reasoning that acknowledges attitudes of taking propositions as options –
what I call floats – as well as attitudes of ruling out propositions as options – sinks.
Agents use epistemic language to coordinate collective reasoning in situations in which
they are uncertain. To this end, they pragmatically express floats and sinks by using epistemic
language. For example a speaker who utters, “Angelica might be the thief” expresses a float
that Angelica is the thief. This pragmatic fact fits nicely with the traditional, contextualist
semantics for epistemic language. Seeing things from this integrated semantic and pragmatic
point of view allows us to explain data about appropriate uses of and responses to uses of
epistemic language – data that have been used in objecting to the traditional semantic theory.
The dissertation begins with a survey of some problems for the traditional,
contextualist view of epistemic language. I show, however, that all of the dominant theories of
‘might’ – traditional contextualism, relativism, and expressivism – face similar problems.
None can straightforwardly explain the range of uses of and responses to uses of ‘might’-
sentences by way of the direct speech act (the speech act that is closely correlated with the
sentence’s semantic value). I also show that relativism and expressivism have further
problems that incline us to investigate the resources available to traditional contextualism.
The key to this investigation is the realization that speakers who use ‘might’-sentences
typically do more than perform the direct speech act. The question is what? In Chapter Two,
I argue that existing theories have answered this question incorrectly because they have failed
to recognize how the indirect speech acts (the speech acts that are not closely correlated with a
vi
sentence’s semantic value) performed by a speaker in uttering a ‘might’-sentence play a vital
role in coordinating collective reasoning under uncertainty.
So, in Chapter Three, I investigate how agents should and do reason in situations of
uncertainty, paying particular attention to the attitudes involved in explicit reasoning. An
agent reasons starting from a privileged set of the open possibilities – the propositions she
floats, or takes as options in reasoning. An agent can float propositions to varying degrees,
and she can rule out propositions as options in reasoning, or sink them. Since agents engage
in contingency reasoning, they must also have conditional versions of floats and sinks. And I
argue that agents also have what I call quantificational versions of floats and sinks. I identify
the rational relations between these attitudes – the various types of floats and sinks – and show
how agents can disagree with each other in virtue of having them.
At this point, we have the materials in the philosophy of mind needed to understand
‘might’-communication. In Chapter Four, I develop the central idea – that a speaker who
utters a ‘might’-sentence typically expresses a float, as an indirect speech act. Although hearers
could figure out which indirect speech act is performed by explicitly thinking about the direct
speech act performed and the purpose of the conversation, such reasoning is typically
bypassed. This fact, along with others, suggests that the indirect speech act is what Kent Bach
and Robert Harnish (1979) call standardized, which explains how it can be responsible for our
judgments about the appropriateness of an utterance and responses to it. With this, we arrive
at a satisfying explanation of the problematic data from Chapter One.
In Chapter Three I explore complex attitudes used in reasoning with uncertainty. In
Chapter Five, I argue that complex ‘might’-sentences are sometimes used to indirectly express
these attitudes. For example, conditional ‘might’-sentences are used to indirectly express
conditional floats. Though some other complex ‘might’-sentences are not used to express
attitudes, they are used to aid in reasoning with uncertainty. For instance, attitude ascriptions
with ‘might’-sentences in the content clause are used to indirectly convey information about
agents floating propositions. I employ this observation in solving Seth Yalcin’s (2007) puzzle
about why it is infelicitous to say, “Suppose it is raining and it might not be.” It is infelicitous
because, according my account, a speaker who says this requests that the addressee be in an
irrational suppositional state. Once this is seen, Yalcin’s puzzle for contextualist semantics
collapses.
Chapter Six extends my account of epistemic communication to epistemic ‘must’-
vii
sentences, which are used to indirectly express sinks, and to epistemic ‘likely’-sentences, which
are used to indirectly express degreed floats. I also deal with some intricacies surrounding the
use of the epistemic ’should’.
In Chapter Seven, I investigate the issue of whether floats are reducible to some more
familiar sort of mental state. A natural thought is that floats are just partial beliefs or
credences. I show that although floats are not credences, they are related to partial credences
in essentially the same way that attitudes of taking propositions for granted (or accepting
propositions) in inquiry are related to attitudes of full credence. This position implies that
floats, like attitudes of acceptance, are properly responsive to some practical reasons (and not
just evidential ones). I close by re-situating the issue of disagreement in virtue of float-like
attitudes in the context of this discovery that they are practical attitudes.
viii
Chapter One: Uses of ‘Might’ and Theories of Direct Speech Acts
There are a variety of ways in which ‘might’-sentences, like “Angelica might be the
thief”, can be used and those uses can be responded to. One central project of this
dissertation is to give a theory of communication involving ‘might’-sentences. In this chapter,
I’ll argue that the three most popular theories of the communicative impact of uses of ‘might’-
sentences – contextualism, relativism, and expressivism – fail to straightforwardly explain the
range of ways in which ‘might’-sentences can be used and responded to. I will suggest an
independent reason to doubt relativism – based on doubts about the coherence of relative
propositional truth – and an independent reason to doubt expressivism – based on hurdles for
explaining how the meanings of complex ‘might’-sentences arise systematically from the
meanings of simple ‘might’-sentences and other linguistic expressions. This will set the stage
for much of the rest of the dissertation, in which I develop a sophisticated pragmatics of
‘might’-communication. This pragmatics, when paired with contextualism, can explain the
range of ways that ‘might’-sentences are used and those uses are responded to.
1. Uses of ‘Might’ and Responses to Uses of ‘Might’
In this section, I’ll present some different sorts of conversations where epistemic
‘might’-sentences are used. These will display a characteristic flexibility in how ‘might’-
sentences are used and how those uses are responded to. One goal of this dissertation is to
develop a plausible theory which can explain the uses of and responses to uses of ‘might’-
sentences in all such conversations. I’ll start with a very popular idea from the literature on
modals (see, for example, Kratzer 1977; 1981; 1991) – that what someone says when they utter a
simple ‘might’-sentence is that the prejacent is consistent with some information.
1
(A quick
definition: when a ‘might’-sentence is in the form,
┌
It might be that S
┐
, the sentence that is
1
The most natural construal of consistency for comparing propositions (or sets thereof) in the epistemic case
is a priori consistency. The idea is that a set of propositions is inconsistent if and only if a contradiction can
be derived from it a priori. But this leads to the following problem (Huemer 2007). Suppose we are
mathematicians considering some large number, x, for which it is unknown whether it is prime or composite.
It seems that I’ve said something true when I say, “x might be prime and it might not be.” But, the a priori
account of consistency predicts what I said to be false. This is because I accept the axioms of arithmetic. And
from these axioms, it either follows a priori that x is prime or it follows a priori that x is not prime. Either
way, what I said in uttering the disjunction was false.
Here are two ways to deal with this problem which are consistent with the view to be developed in
this dissertation. We could weaken the relevant notion of consistency – so that a set of propositions is
epistemically possible if a contradiction cannot be deduced a priori from the set, in some small number of
steps given normal cognitive abilities (this idea comes from Huemer (2007)). Or we could use the tools
developed in Chapters Three and Four to explain why what I say with my utterance in the above example
seems true even though it is, strictly speaking, false. See Finlay and Lennertz MS for this approach.
1
substituted for S is called the ‘prejacent’. The prejacent is, strictly speaking, a sentence. But,
for readability, I often use the word ‘prejacent’ to talk about what is really the content of the
prejacent in a context – a proposition.)
I said that what matters is consistency with some information. But what information,
exactly, enters into what is communicated? I will give a number of examples that show that
the relevant body of information is different in different scenarios. So, I will suggest that we
should be flexible about what information can be relevant for a use of a ‘might’-sentence. For
each example, I will suggest what the relevant information is for the use of ‘might’. I take my
suggestions to be natural first starting points, though I question some of these suggestions in
the rest of this chapter and the next. In these conversations, I will also include a response to
the use of a ‘might’-sentence. The response will be crucial in forming a complete account of
how ‘might’-sentences function in conversations, and I will discuss responses throughout this
dissertation.
Consider the following scenario. There were 10 cookies in the cookie jar this morning.
In the afternoon the cookies are gone. Lil and Susie are investigating the cookie theft. They
know that Angelica isn’t the thief. But Phil doesn’t know that Angelica isn’t the thief. Phil
walks in:
(Cookies) Susie and Lil: Phil, we’re interviewing people to get information about
who stole the cookies. Do you know who stole them?
Phil: No, but Angelica might be the thief.
Lil: No. She can’t be. I was with her all afternoon.
In (Cookies), Phil seems to convey that it is consistent with his information that Angelica is
the thief (though I will discuss objections to this position in §3). So, his utterance is what I
will call a speaker-centric use of ‘might’.
This makes it clear that Phil can use a ‘might’-sentence to talk about what is consistent
with his own information. But Phil is not peculiar here. Any speaker can, in at least some
situations, use a ‘might’-sentence to talk about what is consistent with her information. So,
our account of ‘might’ must respect this fact – that the information relevant to what is
communicated by a ‘might’-sentence can be the information of the speaker.
But this is not the only way to use ‘might’. A speaker can also use ‘might’ to talk about
what is consistent with the information distributed across the members of a group to which
she belongs. I will, from here on, call this the group’s information.
2
Consider the following
2
Though I focus here on a group’s distributed information, there is also the notion of the group’s shared
information – the information possessed by each member of the group. The distributed information will
2
example. Holmes and Watson are investigating a murder. They list all of the evidence they
have, which straightforwardly rules out that Moriarty is the murderer. Watson, however, does
not do a good job of deducing what follows from the evidence. They have the following
conversation:
(Murder) Holmes: What does our evidence show? Who are the remaining
suspects?
Watson: Well, Moriarty might be the murderer.
Holmes: No. He can’t be. Our evidence rules it out.
We can call Watson’s a group-centric use of ‘might’. This is because his use of ‘might’ is
based on the information of the group consisting of he and Holmes. So, our theory must
allow that speakers can use ‘might’ to talk about what is consistent with their own information
or with the group’s information.
We can also use ‘might’ to talk about what is consistent with a body of information
that is neither the information of the speaker nor of a group that includes the speaker. The
following is an adaptation of von Fintel and Gillies’ (2007; 2008; 2011) Mastermind case.
Suppose that Phil and Lil are playing battleship and that Lil, in fact, has a PT boat, not a
battleship, on C4 and knows she does:
(Battleship) Phil: There might be a battleship on C4.
Lil: Yeah. There might be.
Let’s call Lil’s use of ‘might’ an exo-centric use. This is because it is not made relative to her
information nor is it made relative to the information of some group of which she is a
member. Rather, it seems that Lil’s use of ‘might’ takes into account Phil’s information. She
says that it is consistent with his information that there is a battleship on C4. So, our theory
of ‘might’ must allow for all of these sorts of uses – speaker-centric, group-centric, and exo-
centric.
I have focused on simple, unembedded ‘might’-sentences. But ‘might’-sentences can
embed in a variety of linguistic contexts (under connectives, quantifiers, attitude verbs, etc.).
For much of this chapter (excepting §7), I will be concerned primarily with simple ‘might’-
sentences. Chapter Five focuses on complex ‘might’-sentences.
So, what are we supposed to take from these conversations? First, ‘might’ can be used
to talk about different sorts of information – eg. the speaker’s information, the conversational
play a much larger role in our discussion, so I focus on it in the text, but the theory given later in this section,
Contextualism, allows that, in some scenarios, a speaker’s use of ‘might’ is about what is consistent with the
shared information of the group. See Cook 2013 for more on group information.
3
group’s information, and even a body of information that does not include the information of
the speaker. We want a theory that is flexible enough to explain all the ways that ‘might’ can
be used.
But it is not just that there is flexibility in how ‘might’-sentences are used. There is
flexibility in how these uses are responded to. As I will discuss in what follows, it sometimes
appears that the hearer responds to an utterance of a ‘might’-sentence by responding to the
proposition that the speaker seems, at first blush, to have asserted (eg. Holmes’s negative
response in (Murder) and Lil’s positive response in (Battleship)). But at other times it
appears that the hearer responds to something else, since her response wouldn’t be
appropriate if it were a response to the proposition the speaker seems to have asserted (eg.
Lil’s response in (Cookies)). I will spend much more time discussing these specific examples
throughout the dissertation. What is important to understand now is that we need a theory of
‘might’-communication that is flexible enough to explain not just all the ways that ‘might’-
sentences can be used, but how to appropriately respond to these different sorts of uses of
‘might’-sentences.
2. Theoretical Outlook
My project contrasts with one that has sometimes been pursued in the literature. The
structure of some arguments on the topic of epistemic modals proceeds as follows. We
suppose the truth of some semantic theory of ‘might’. Then we consider a particular use of a
‘might’-sentence in a context. We are told what the semantic theory predicts is the truth-value
of the sentence in the context. We are asked to compare the predicted truth-value with our
judgment of the truth-value. Often the prediction and our judgment will come apart. The
conclusion is that the semantic theory that yielded the prediction is false.
Something like this structure of argument appears in the epistemic modal literature in
Hacking 1967, DeRose 1991, and most relevantly in Egan, Hawthorne, and Weatherson 2005.
I’ll quickly relay a succinct version of this sort of argument – one that appears in MacFarlane
2011a (though it is important to note that MacFarlane is also interested in data about speakers’
retracting previous utterances and about disputes between groups with different information).
He calls this sort of argument one from third-person assessments: “You overhear George and
Sally talking in the coffee line. . . Sally says, ‘Joe might be in Boston right now.’ You think to
yourself: Joe can’t be in Boston; I just saw him an hour ago here in Berkeley. Question: Did
Sally speak falsely?” MacFarlane assumes our judgment is that she did, though the theory he
4
is arguing against predicts that Sally spoke truly. So, we should take that as evidence against
the theory. I don’t mean to either endorse this sort argument across the board or to say that it
is never useful. I just want to contrast it with my method.
The way that I will proceed diverges from third-person assessment arguments in at
least two ways. First, I’m not going to attempt to give a semantics for epistemic modals.
Rather, I aim to give a pragmatic theory. Whatever pragmatic theory is correct will place
constraints on the semantic theory one can adopt while endorsing plausible semantic-
pragmatic bridge principles. Nonetheless, there are no unassailable conclusions that can be
drawn about the semantics of ‘might’ from a theory of how ‘might’-sentences are used and
how uses of them are responded to.
The second difference between my method and the one that involves third-person
assessments is that I don’t ask the reader to assess the truth of what someone said in using a
‘might’-sentence. In fact, I’m not particularly concerned with truth. One reason for my lack
of concern is that theorists disagree about the truth-values of ‘might’-sentences. Knobe and
Yalcin (MS) relay experimental results that suggest that ordinary speakers do as well. They
find that judgments about truth-values in some cases are mixed (I don’t mean to endorse these
results; I only mean to stay away from the data they appear to neutralize). Furthermore,
focusing on truth-values rules out (or at least seriously disadvantages) some of the possible
theories of ‘might’ from the outset – those that say what it at stake is not truth-evaluable.
Judgments about appropriate uses of and responses to uses of ‘might’-sentences in
conversations don’t rule out these theories out of hand. Furthermore, these judgments allow
us to easily take into account pragmatic factors that may be playing a role in our judgments
about conversations involving ‘might’-sentences. For example, even if a ‘might’-sentence in a
context is true according to some semantic theory, we may judge a negative response
appropriate based on something else that is conveyed. So, by not focusing on judgments
about truth-values, I can (i) avoid shoehorning our truth-value judgments into a neat package,
(ii) allow all theories (including non-truth-conditional ones) to compete on a level playing
field, and (iii) avoid bias in favor of semantic rather than pragmatic explanations of the
relevant data.
3
So what, more specifically, is the method I’ll be using to judge between different
theories of ‘might’-communication? As I’ve said, I want a theory that incorporates an
3
I am not the first to see the virtue of looking at conversations in order to see how pragmatics play a large role.
See, in particular, Björnsson and Finlay 2010, von Fintel and Gillies 2011, Dowell 2011, Montminy 2012, and
Bach 2011.
5
explanation of both uses of and responses to uses of ‘might’-sentences. Later in this chapter,
I’ll investigate a number of theories of the direct speech act performed by a speaker who uses a
‘might’-sentence. These theories are candidates for serving as at least part of an explanation of
communication involving ‘might’-sentences. But, by themselves, the theories say nothing
about why an utterance would be appropriate or about the sorts of ways a hearer can respond
to a use of a ‘might’-sentence. For that, we need some more theoretical tools.
We need a theory of what speech acts performed by a speaker who uses a ‘might’-
sentence can be at-issue – in the sense that hearers can respond to them in a straightforward
manner and that they are responsible for our judgments of appropriateness. For example, an
(perhaps the) at-issue speech act performed by a speaker who utters “Angelica is a
philosopher” in some context is an assertion of the sentence’s semantic value – the proposition
that Angelica is a philosopher. However, an at-issue speech act need not always be an
assertion of the sentence’s semantic value. For example, it can sometimes be a proffering of
the embedded clause:
(Philosopher) Betty: Phil thinks that Angelica is a philosopher.
Lil: No. She’s not.
4
Lil must be responding to some act of putting forward the embedded clause, not to an
assertion of the proposition that is the semantic value of the whole sentence that Betty utters.
And we will see in Chapters Two and Four that there are more options for at-issue speech acts
than just these. Nonetheless, it will be instructive to start with the simplest candidate theory of
at-issue speech acts:
The Direct Approach: The direct speech act performed by a speaker who assertively
utters a ‘might’-sentence is the (only) at-issue speech act.
I take the direct speech act performed by a speaker who utters a sentence to be the one that is
correlated with the form of the sentence and the meaning of the words of the sentence.
5
For
instance, the direct speech act performed by a speaker who uses the sentence, “Grass is green”
is an assertion of the proposition that grass is green while the direct speech act performed by a
speaker who uses the sentence, “Is grass green?” is an asking of whether grass is green. The
4
This point is made, among other places, in Björnsson and Finlay 2010 and Simons 2010.
5
Particular features of a sentence determine particular aspects of the direct speech act. The type of speech act
is determined largely by the form of the sentence. The meanings of the words are generally only in play in
determining the object of the speech act – not the sort of speech act. This is important when dealing with
performatives. It is plausible (pace Austin 1979; 1962) that the direct speech act performed by a speaker who
utters “I promise to be there by eight” is an assertion that the speaker promises to be there by eight – not a
promise to be there by eight. Typically, such an utterance will also count as a promise, but not as the direct
speech act. See Bach 1975 and Bach and Harnish 1979; 1992. Some have claimed that the notion of the direct
speech act is a notion located at the interface of syntax and pragmatics. See, in particular, Morgan 1975.
6
Direct Approach says that the direct speech act performed by a speaker who uses a ‘might’-
sentence is responsible for our judgments about appropriateness and is what a hearer can
appropriately respond to.
I’ve yet to say what direct speech act is performed by a speaker who uses a ‘might’-
sentence. In the remainder this chapter I will survey three theories which purport to answer
this question – contextualism, relativism, and expressivism. These are typically discussed in
the literature as semantic theories. I will discuss the corresponding theories of direct speech
acts. In most cases, a view that takes a certain position about semantics takes the same
position about direct speech acts. One reason I frame the discussion in terms of the direct
speech act rather than the semantic content has already been said – that I am concerned
primarily with communication. The second reason for framing the discussion in this way is to
make room in the classification for semantic views that are somewhat less traditional, though
make similar predictions about communication – like Bach 2011. Though he doesn’t adopt a
contextualist semantics, his view has many of the same advantages and encounters many of the
same challenges as a contextualist semantic view – since the views agree on the direct speech
act performed.
In §3, §4, and §6, I show that none of these views of the direct speech act performed by
an assertive use of a ‘might’-sentence, in tandem with The Direct Approach, provide a simple
explanation of the various ways that we can use and respond to uses of ‘might’-sentences. In
§5 and §7, I give some further evidence to doubt relativist and expressivist theories of direct
speech acts – independent of The Direct Approach. In §8, I take this as a sufficient reason to
re-examine the traditional, contextualist theory and pair it with more liberal theories of at-
issue speech acts (such a re-examination is the topic of Chapters Two and Four).
3. Contextualism and The Direct Approach
In this section, I investigate how the traditional, contextualist theory of epistemic
modals, together with The Direct Approach, fares in explaining the conversations that we’ve
considered. I’ll start by explaining the basis for contextualism and stating the theory. Then,
I’ll show why it, together with The Direct Approach, cannot explain conversations like
(Cookies). My conclusion is that the conjunction of the contextualist theory and The Direct
Approach is incorrect.
The traditional, contextualist semantic theory of epistemic modals is based on a unified
treatment of all modals – like ‘might’, ‘must’, and ‘should’. In the philosophy literature, it is
7
often presented as a simplified version of Angelika Kratzer’s (1977; 1981; 1991) semantics for
modals. On such an account, the semantic value of a modal can be represented as a quantifier
over possible worlds. Different flavors of modality (metaphysical, circumstantial, deontic)
correspond to differences in the possible worlds quantified over. For epistemic modals, the
possible worlds quantified over are those in which the propositions that make up the relevant
body of information are all true. So, for different bodies of information, epistemic modals
quantify over different possible worlds.
I’m going to give an account of epistemic modals that is in this spirit. But two things
are important to note. First, I talk about a proposition being consistent with information
rather than there being a possible world in which the proposition is true. Given a notion of a
priori consistency, this comes out to (nearly enough) the same thing.
6
Second, remember that
I am not giving an account of the semantics of ‘might’-sentences. Rather, I am giving an
account of the direct speech act performed by an assertive utterance of a ‘might’-sentence.
Here is the account:
Contextualism: The direct speech act performed by a speaker who assertively utters
┌
It might be that S
┐
, where p is the content of S in the context, is an assertion of the
proposition that p is consistent with the information of the context.
As we’ve seen, the relevant body of information can change from context to context. Simple
‘might’-sentences are sometimes used to assert propositions about the speaker’s information,
other times used to assert propositions about the information of some group including the
speaker, and other times used to assert propositions about a set of information that does not
include the information of the speaker. A major virtue of Contextualism is that it allows for
this sort of flexibility.
Contextualism, as a theory of direct speech acts, fits nicely with two views of the
semantics of the epistemic ‘might’. The first is the view that the semantic content of a ‘might’-
sentence, at a context, is what we can call a ‘might’-proposition (or MP) – a proposition that
predicates of the prejacent consistency with some body of information. For example, such a
view would say that the semantic content of “Angelica might be the thief” at a context is the
proposition that it is consistent with the information of that context that Angelica is the thief.
We would then pair this view with a simple principle linking semantic content and direct
6
Differences would arise involving the necessary a posteriori and contingent a priori. Our examples will be
free of these complications. Also, it may be helpful to start with a background not of metaphysically possible
worlds, but rather of what Soames (2006) calls epistemically possible worlds (or world-states) – which are
worlds that can't be ruled out a priori (from no premises whatsoever).
8
speech acts – that the direct speech act performed by an assertive utterance of a sentence in a
context is an assertion of that sentence’s semantic content in that context. I take this to be
roughly the standard view about the semantics of epistemic modals. Something like this is
endorsed by von Fintel and Gillies (2011), Dowell (2011), Schaffer (2011), Montminy (2012), and
Stanley (2005).
The second semantic view that fits nicely with Contextualism is one according to which
the semantic content of a ‘might’-sentence at a context is something less than fully
propositional. We might think of it as a proposition with a gap – the gap is where the body of
information goes. So, the semantic content of “Angelica might be the thief” at a context is not
a proposition. It is some object with a gap where the information is supposed to go. When a
‘might’-sentence is assertively uttered, the gap gets plugged in with the contextually relevant
body of information. And, so, a complete proposition is asserted. This proposition is an MP,
like the proposition that it is consistent with the information of that context that Angelica is
the thief. This sort of view is found in Bach 2011.
We can give a view of epistemic communication without committing to either of the
semantic accounts just sketched. What is at stake between these views stems largely from
differences in theoretical systems and in views about the syntax/semantics and
semantics/pragmatics interfaces – differences that need not concern us here.
One important feature of both of these semantic pictures of ‘might’ is that they allow
for a unified semantics for modals in general, rather than a specific account for only epistemic
modals. This is a semantics which applies regardless of the flavor of the modal term. For
example, these accounts allow that a word like ‘may’ has a single meaning even though it can
be used epistemically or deontically. We are not forced to posit ambiguity between ‘may’ in
sentences like the following: “Angelica may be the thief” and “You may go to the bathroom.”
‘May’ always quantifies over the relevant set of possible worlds, though what worlds these are
changes for different flavors of modality (and even inside a particular flavor – as is the case
with the epistemic ‘may’). But we would be required to posit ambiguity according to the other
accounts of epistemic modals to be discussed in this chapter – relativism and expressivism.
That Contextualism fits nicely with the unified standard semantic account is a reason in its
favor – one that will provide part of the impetus for exploring the theoretical options for
rescuing it after we see problems with simple versions of all three theories.
But the combination of Contextualism and The Direct Approach is not explanatorily
adequate. It cannot explain the conversation, (Cookies). I’ll repeat the scenario surrounding
9
(Cookies) here. There were 10 cookies in the cookie jar this morning. In the afternoon the
cookies are gone. Lil and Susie are investigating the cookie theft. They know that Angelica
isn’t the thief. But Phil doesn’t know that Angelica isn’t the thief. Phil walks in:
(Cookies) Susie and Lil: Phil, we’re interviewing people to get information about
who stole the cookies. Do you know who stole them?
Phil: No, but Angelica might be the thief.
Lil: No. She can’t be. I was with her all afternoon.
In (Cookies) Phil is asked for his information about who stole the cookies. He responds by
uttering a ‘might’-sentence. It seems that the information of the context is Phil’s. So, given
Contextualism, Phil asserts the proposition that it is consistent with his information that
Angelica is the thief. Given The Direct Approach, his assertion is the only at-issue speech act.
It is what Lil can appropriately respond to. Since, for all Lil knows, it is consistent with Phil’s
information that Angelica is the thief, a negative reply should be inappropriate. But Lil gives
a negative reply. And it is appropriate. Thus, it seems that the conjunction of Contextualism
and The Direct Approach is incorrect.
This is taken by many to be an important result. Before agreeing, we should consider
one response. It says that we might want to revise our judgments about what information is
relevant in (Cookies). We have assumed that the relevant information is Phil’s. But, perhaps,
the information that is relevant is the group’s (distributed) information. According to
Contextualism, together with the hypothesis that the group’s information is relevant, Phil
asserts the proposition that it is consistent with the group’s information that Angelica is the
thief. By The Direct Approach, his assertion is the at-issue speech act. Since Lil knows that
Angelica is not the thief, the proposition that Angelica is the thief is not consistent with the
group’s information. And so, it makes sense why Lil can respond negatively. What Phil
asserted is false.
However, this combination of theses – Contextualism, The Direct Approach, and the
hypothesis that the information of the context is the group’s information – is incorrect.
Though it seems to make sense of Lil’s response in (Cookies), it cannot make sense of the
appropriateness of Phil’s assertion (von Fintel and Gillies 2011; MacFarlane 2011a). Given
The Direct Approach, Phil’s assertion is the only speech act that is at-issue – the only one that
could explain why his utterance is appropriate. But Phil is not warranted in asserting that it is
consistent with the group’s information that Angelica is the thief, since he doesn’t know what
is consistent with the group’s information. It would be unwarranted for Phil to say instead, “It
10
is consistent with all of our information that Angelica is the thief.” Since this sentence is used
to assert the same proposition as the group reading of the ‘might’-sentence, according to the
hypothesis, Phil’s original utterance in (Cookies) should be unwarranted, too. However, his
utterance is, in fact, perfectly appropriate. So, this hypothesis about what is the information of
the context must be incorrect.
7
MacFarlane (2011a) uses this sort of example to create a general dilemma for the
combination of Contextualism and The Direct Approach. For a conversation like (Cookies),
the theorist who accepts this combination must say what the relevant information is. This
body of information either includes Lil’s information or it does not. If it does not, then it is
impossible for the theory to explain the appropriateness of her response in (Cookies). If it
does, then it is impossible to explain why Phil’s utterance is warranted. Either way, the
combination of Contextualism and The Direct Approach cannot give a complete explanation
of why (Cookies) is a perfectly natural and appropriate conversation.
Some have taken this to show that Contextualism is false. Really, however, all this
shows is that the combination of Contextualism and The Direct Approach is false. For now, I
will investigate two alternate theories of direct speech acts – relativism and expressivism. I will
return to evaluate The Direct Approach at the end of the chapter.
4. Relativism and The Direct Approach
Relativism about epistemic modals has, in the last decade or so, become a popular
position (see Egan, Hawthorne, and Weatherson 2005; MacFarlane 2011a; Stephenson 2007).
Relativists are motivated, at least in part, by the failure of Contextualism and The Direct
Approach to explain conversations like (Cookies). They generally deny the central claim of
Contextualism – that the proposition a ‘might’-sentence is used to assert (the content of the
direct speech act) varies based on the information of the context of utterance. Rather, they
think that the proposition it is used to assert is the same in every context. But they think that
that proposition has a truth-value not absolutely but only relative to a further parameter,
which is provided by the context in which the proposition is assessed. So, the truth-value of
the proposition that is asserted depends on who is assessing it:
Relativism: The direct speech act performed by a speaker who assertively utters
┌
It
might be that S
┐
, where p is the content of S in the context of utterance, is an assertion
of a proposition which is true at a context of assessment, just in case p is consistent
7
This move will be discussed further in Chapter Two. The reader can also consult Lennertz MS c.
11
with the information of that context of assessment.
We can think of a context of assessment as a situation in which someone considers a
proposition. Relativism says that the very same proposition can be true at one context of
assessment and false at another. So, a proposition can have different truth values (in the same
world) depending on the information of the person who assesses it. In this section, I will
assume that this idea of truth relative to the information of a context of assessment makes
sense, in order to evaluate the conversational predictions of Relativism. In the next section I
will question the comprehensibility of this notion of relative truth.
One reason for Relativism’s rise in popularity is that it, together with The Direct
Approach, appears to give a simple, straightforward explanation of (Cookies). According to
Relativism, Phil asserts a proposition that is true at a context of assessment just in case it is
consistent with the information of that context of assessment that Angelica is the thief. The
information of the context of assessment we need to consider for thinking about the
appropriateness of Phil’s utterance is his information. And the proposition is true relative to
this information. So, his utterance is appropriate since the assertion of this proposition is,
according to The Direct Approach, the at-issue speech act. The context of assessment we
need to consider in thinking about Lil’s reply is one where Lil is the assessor. So the
information of this context is Lil’s. And the proposition Phil asserts (the very same one that is
true relative to his information) is not true relative to Lil’s information – it is not consistent
with her information that Angelica is the thief. So, Lil appropriately denies this proposition.
Thus, Relativism and The Direct Approach yield the desired prediction – that Phil
appropriately asserts a proposition which Lil appropriately denies.
One supposed virtue of this account is that it is extremely straightforward. On the
Contextualist account, we had the complications of figuring out what information is relevant
in a context – a complication that leads to some epicycles, though no solution, to the problem
presented using the case of (Cookies). The Relativist, in contrast, seems to have a simple
answer to what information is relevant – the information of the agent assessing the
proposition. However, in the rest of this section, I will suggest that Relativism needs to
introduce complications, some of which seem implausible, to account for the whole range of
data.
Let’s revisit our example of an exo-centric use. Remember that Phil and Lil are
playing battleship and that Lil, in fact, has a PT boat, not a battleship, on C4 and knows she
does:
12
(Battleship) Phil: There might be a battleship on C4.
Lil: Yeah. There might be.
According to Relativism, Phil asserts a proposition that is true at a context of assessment just
in case it is consistent with the information of that context of assessment that there is a
battleship on C4. Given The Direct Approach, his assertion is the at-issue speech act. The
context of assessment of Lil’s reply is the one in which she assesses the claim. And it is not
consistent with Lil’s information that there is a battleship on C4. Lil is, after all, looking at the
PT boat on that space. So it should be inappropriate for her to respond positively. But it is
appropriate for her to respond in this way. This makes it look like the conjunction of
Relativism and The Direct Approach cannot explain (Battleship).
8
There are a number of ways that Relativists can complicate their theory to try to
explain this sort of case. They could adopt a hybrid view – where sometimes ‘might’-sentences
are used to assert contextualist propositions and sometimes they are used to assert relativist
ones.
9
It is easy to see how such a view can explain (Battleship). In that case, Phil asserts a
contextualist proposition – the proposition that it is consistent with his information that there
is a battleship on C4 – and Lil responds to that proposition. However, this account lacks an
explanation of the flexibility that hearers have in responding. Consider two different responses
that Lil can give in a conversation that starts like (Battleship):
(Battleship') Phil: There might be a battleship on C4.
Lil: Yeah. There might be. / No. There can’t be.
Lil’s ability to respond in either way suggests that Phil asserts neither just a contextualist
proposition nor just a relativist proposition. That is, this case seems to be evidence against
any account where there is a single at-issue speech act (more on this in Chapter Two).
Relativists like Egan, Hawthorne, and Weatherson (2005), Stephenson (2007), and
Egan (2007) take a different approach. They claim that (Battleship) is not a standard
8
Ross and Schroeder (2013) present a more complex challenge for Relativism. They charge that it can’t
explain what they call reversibility cases. These are cases where Phil utters a sentence like “The battleship
might be on C4 and it might not be”, knowing that he will later find out where the ship is and will be in a
position to utter that sentence’s negation: “It’s not the case that the battleship might be on C4 and it might
not be.” Given that he knows at the time of the original utterance that he will make this later utterance, it
would seem irrational for him to make his utterance in the first place. This is because, given the principle of
reflection, if he thinks he will later come to disbelieve the proposition based on rational evolution of his
views, then he should now disbelieve it. The problem, then, seems to be Relativism’s claim that Phil asserts
a proposition and later the negation of that proposition – and that these assertions land him in disagreement
with his former self.
9
I use lowercase ordinary type here because I am not talking about the theories of direct speech acts,
Contextualism and Relativism, but the theories of propositions that they presuppose. A contextualist
proposition is an MP – one that is true or false absolutely. A relativist proposition is one that is true or false
only relative to the information of the person who is assessing it.
13
conversation. And so, the conversational mood is not one of genuine inquiry. Rather it is
one of pretense. Lil replies pretending that her information is the same as Phil’s. She is
projecting herself into Phil’s epistemic position. This move is implausible. This is because, in
(Battleship), the word ‘might’ is essential to the appropriateness of Lil’s response. She can’t,
for instance, appropriately say “Yeah. It is consistent with my information that there is a
battleship on C4.” But if the explanatory work were done by a general feature of the
conversation – that Lil is speaking as if she had the same information as Phil – we should
expect such an utterance to be exactly as appropriate as her actual utterance. But it clearly is
not. So, it looks as if an explanation in terms of the general mood of the conversation is
incorrect. The features of ‘might’ play an essential role.
10
Finally, one could accept a flexible version of Relativism. So far, I have been assuming
that the information of the context of assessment is just the information of the assessor. But,
just as Contextualists can claim that the information of the context of utterance need not be
the speaker’s, Relativists can claim that the information of the context of assessment need not
be the assessor’s.
11
Here’s how this could help explain (Battleship). The Relativist could say
that the information of the context of assessment is not Lil’s (the assessor’s) information but
rather Phil’s (the speaker’s). So, since the proposition that Phil asserts is true relative to the
context of assessment, it makes sense for Lil to respond positively.
Questions arise for this view about how a context of assessment makes a body of
information – other than the assessor’s – relevant. Nonetheless, I think this final attempt is
the most promising version of Relativism in response to data like (Battleship). It is important
to realize that, even if it is successful in explaining this case, the resulting view of
communication will be much more complicated than the Relativist originally led us to believe
– when we were drawn in by a straightforward explanation of the data. I won’t try to develop
this sort of view here, but it is worth reminding the reader that Relativism also requires
theorists to treat modals, like ‘may’ as ambiguous – as opposed to the standard theory of
modality that is deployed in Contextualism. The complexity and ambiguity required by
Relativism are two reasons against accepting it. The final reason – that it requires a suspect
10
Furthermore, if this move worked, the Contextualist could use it as well. In (Cookies) Lil could interpret
Phil’s utterance pretending that he has the same information as her. So, if this move were a good one, it
seems that the Relativist would lose her advantage over the Contextualist.
11
This is related to a move MacFarlane (2011a) makes in response to a worry posed by Dietz (2008) (a similar
move is made on behalf of the relativist about ‘ought’ in Björnsson and Finlay 2010). There MacFarlane
suggests that the information of the context of assessment can include the information of the speaker in
addition to the information of the assessor. This is to account for cases where the assessor knows less than
the speaker. But that isn’t the case here. Here we want the information of the context of assessment to
include only the information of the speaker.
14
notion of relative truth – is the topic of the next section.
5. The Conceptual Problem for Relativism
Many have charged that relative truth – of the sort employed in Relativism – does not
make conceptual sense. In this section, I will reject the most sophisticated attempt at making
sense of relative truth. I will conclude that Relativism is implausible since it requires that this
notion of truth makes sense, but, as well will see, we don’t have reason to believe that it does.
I will focus on the following question: what is it for a proposition to be true relative to
the body of information of a context of assessment? We have a pretty good grasp of what it is
for a proposition to be true simpliciter. The proposition that Angelica is a philosopher is true
just in case the universe is a certain way – such that Angelica is a philosopher. But it is not
easy to say what it is for a proposition to be true relative to the body of information of a
context of assessment. Suppose there is some proposition that Angelica might be the thief.
How does the universe have to be in order for this proposition to be true relative to the body
of information of a context of assessment? This is an ill-formed question for the relativist
because truth relative to the body of information of a context of assessment places conditions
on more than just the universe being a certain way. The idea is that the same proposition can
be true relative to the body of information of one context of assessment and false relative to
the body of information of another, even if the universe itself is the same – i.e. even in the
same possible world. Propositions about what might be the case are not true or false
simpliciter, but only relative to a body of information.
The most thoroughly worked out version of relativism is John MacFarlane’s (2005a;
2008a; 2011b; 2014), and it is clear from his writing that there are really two issues in making
sense of relative truth. First, one has to understand what it is for a proposition to be true or
false not simpliciter, but only relative to some further parameter (like a body of information).
Second one has to understand what it is for the value of this parameter, and thus the truth of
the proposition, to be determined by a context of assessment. MacFarlane thinks that the first
issue is unproblematic and the real work needs to be done in figuring out the second issue –
the issue of what he calls assessment-sensitivity. However, I think that there is a conceptual
problem at each of these steps. I will conclude that we haven’t been given good reason to
think that the notion of truth relative to the body of information of a context of assessment
makes sense.
I said that MacFarlane does not think that understanding truth relative to some
15
parameter is all that philosophically problematic. After all, theorists talk all the time about a
proposition being true relative to a possible world. For the case of epistemic modals, we just
need one more parameter. Instead of thinking that propositions are true relative to possible
worlds, we should think that they are true relative to pairs of possible worlds and bodies of
information (of course, the truth-value of non-epistemic modal propositions will not vary
across bodies of information).
12
So, we might think that relativizing propositional truth to
bodies of information should be no more problematic than relativizing it to possible worlds.
Proponents of relativism are correct that there is no technical obstacle to implementing
this idea. But we don’t yet understand what it is for a proposition to be true relative to a body
of information. A natural hypothesis is that we will be able to make sense of what it is for a
proposition to be true relative to a body of information in the same way that we understand
what it is for a proposition to be true relative to a possible world.
Unfortunately, this isn’t so. We can give a natural and informative explanation of
truth relative to a possible world in terms of our ordinary notion of monadic truth.
13
Suppose
that possible worlds are just maximal properties that the universe could have had. So talk
about a possible world where p is the case is just talk about a maximal property the universe
could have had, part of which includes p’s being the case. This allows us to give the following
analysis of truth relative to a possible world in terms of monadic truth. A proposition is true
relative to a possible world just in case, had the universe instantiated that possible world, the
proposition would have been true. For example, Angelica is a podiatrist is true relative to
possible world, w, just in case, had the universe instantiated w, Angelica would be a podiatrist.
In this way, we can make sense of the notion of truth relative to a possible world in terms of
our ordinary notion of truth.
Can we do something similar with the notion of truth relative to a body of
information? It’s not obvious how to do so. If we try to draw a strict parallel with the case of
truth relative to worlds, we run into a problem. We can treat possible worlds as maximal
properties that the universe could instantiate. The reason that this works is that the (ordinary)
truth of a proposition depends on the state of the universe. However, if we try to treat bodies
of information, likewise, as properties, there is no satisfying answer to what they are properties
of. Because, according to our ordinary notion of truth, the truth of a proposition doesn’t
12
One could try to model this using centered worlds. The idea of centered worlds comes from Quine (1969)
and Lewis (1979). It has been applied to epistemic modals in Egan, Hawthorne, and Weatherson 2005 and
Egan 2007.
13
In doing so, I follow Soames (2010). The same sort of strategy is pursued in Cappelen and Hawthorne 2009.
16
depend on the properties of anything beyond the state of the universe, there seems to be no
satisfying answer to this question. So a strict analogy to the case of worlds fails.
MacFarlane (2011b) suggests that things are not this difficult. He thinks the relativist
can explain truth relative to some parameter – for instance, a body of information – in terms
of our ordinary notion of truth as follows. He makes a suggestion for standards of taste – “a
proposition p is true at a standard of taste s if and only if by standard of taste s, p is true”
(446). We can amend it for the epistemic case as follows: p is true at a body of information, i,
if and only if by body of information, i, p is true. Unfortunately, MacFarlane is quite terse
about this point, so it is difficult to see exactly why he takes this biconditional to do the
required work.
MacFarlane is making a suspect assumption here. He is assuming that what appears
on the right side of the biconditional is not simply a restatement of what appears on the left
side. In particular, while the statement on the left side uses a relativized truth predicate, he
thinks the the one on the right uses our ordinary truth predicate together with an operator, “by
body of information, i”. But it isn’t clear why we should think that such an operator exists
without already assuming that propositional truth can be relative to bodies of information
(and thus can be shifted by operators which change the relevant body of information). To
simply stipulate the existence of such an operator will not illuminate the notion of truth
relative to a body of information to those who don’t already understand it. MacFarlane does
suggest that we could use a counterfactual construction as an alternative – something like “If i
were the correct body of information, p would be true.” But, it doesn’t seem that there is any
sense to be made of a body of information being correct when that information can be
incomplete (as it often is in the target cases of truth relative to a body of information).
14
So,
because we don’t understand such a counterfactual, it won’t help us understand the notion of
truth relative to a body of information.
I take it that the preceding has shown, at least, that it is not at all obvious how to make
sense of truth relative to a body of information. I suspect that this is because there is no
14
We could try different amendments to the antecedent of the counterfactual. For instance, “If I had
possessed body of information, i, p would be true.” But this gets strange results. For example, suppose I
know you are not in Boston. Then it seems false so say that if I had lacked the information that you are not
in Boston, it would have been the case that you might be in Boston. Only changes in your whereabouts
would seem to reopen the possibility of the truth of the proposition that you might be in Boston. But
plugging this into the biconditional yields that the proposition that you might have been in Boston is false at
a body of information that leaves it open that you are in Boston – not MacFarlane’s desired result. The
judgment about the counterfactual here seems similar to judgments about counterfactuals used to argue
against simple versions of meta-ethical expressivism – if I hadn’t disapproved of murder, it wouldn’t be
wrong. Thanks to Keith Hall for this point.
17
making sense of such a notion. In the remainder of this section, I am going to put this issue
to the side and evaluate the second portion of MacFarlane’s explanation of relative truth – the
portion that he takes to be more contentious and interesting. This is the project of making
sense of assessment-sensitivity – or what it is for a proposition to be true at a context of
assessment.
In MacFarlane 2005a, he tries to make sense of the notion of truth at a context of
assessment. His idea is that we can get a grip on a notion of truth by seeing how it figures in
the commitments that we take on in making assertions. He notes three sorts of commitments,
but focusing on one (2005a, 318) suffices for our purposes:
(W) Commitment to withdraw the assertion if and when it is shown to have been
untrue.
His (2005a, 320) idea, then, is to try to understand commitment to a relativist proposition by
relativizing (W):
(W*) In asserting that p at C1, one commits oneself to withdrawing the assertion (in any
future context C2) if p is shown to be untrue relative to context of use C1 and context of
assessment C2.
15
MacFarlane’s idea is that our understanding of the commitment stated in (W*) provides us
with insight into what truth relative to a context of assessment is.
It is difficult to see what traction this is supposed to give us on understanding the
notion of truth at a context of assessment. We understand the commitment in (W) because
we already understand what it is for a proposition to be shown to be untrue. But the
commitment in (W*) is difficult to understand precisely because we don’t know what it is to
show a proposition to be untrue relative to a context of assessment. Knowing what it is to
show this seems to require knowing what it is for a proposition to be true relative to a context
of assessment. But this is exactly the notion we are trying to get at.
All MacFarlane (2005a, 320-21) says about this is, “There should be no worries about
the intelligibility of (W*). Logically it is no more complex than a commitment to refill the
pitcher (at any future time t) if it is shown to be empty (at t).” But, rather than helping us
understand the notion of being empty at t, this commitment is something we can understand
because we already understand what it is for the pitcher to be empty at t. In both cases,
understanding the commitment to do some action on being shown that some condition holds
(that the proposition is untrue at a context of assessment, that the pitcher is empty at t)
15
In MacFarlane 2014, he gives a similar rule: “Retraction Rule: An agent in context c2 is required to retract an
(unretracted) assertion of p made at c1 if p is not true as used at c1 and assessed from c2.”
18
requires understanding what it takes for the condition to hold – not the other way around.
We don’t get a grasp on what it is for the condition to hold by, before that, understanding the
commitment. So, we don’t get a grasp on what it is for a proposition to be true relative to a
context of assessment by understanding a commitment to doing something when shown that a
proposition is true relative to a context of assessment. The order is backward. So,
MacFarlane’s suggestion does not succeed.
16
In this section, we’ve seen that MacFarlane’s attempt – the most sophisticated in the
literature – to make sense of truth relative to the body of information of a context of
assessment fails. We haven’t seen good reason to think that we can make sense of either truth
relative to a body of information or truth at a context of assessment. Perhaps there are
successful ways to understand truth relative to the body of information of a context of
assessment, but the burden clearly is on the Relativist. I won’t further investigate this position
here. Rather, I will leave Relativism to the side for the rest of the dissertation and explore
other options.
6. Expressivism and The Direct Approach
In this section, I’m going to consider a different theory of ‘might’-communication –
expressivism. Despite being different from Relativism, expressivism is, at first blush,
confronted with the same sort of challenge as Relativism and The Direct Approach. The view
must be complicated to account for conversations like (Battleship).
The semantic theory, expressivism, is a non-factualist theory of meaning, and the
associated expressivist theory of direct speech acts is likewise non-factualist. That is, it does
not say that the direct speech act performed by a forceful utterance of a ‘might’-sentence is an
assertion that attempts to describe some fact about the world. Rather, expressivism says that
such an utterance expresses an attitude other than belief.
17,18
Expressivists about ‘might’ say
16
MacFarlane may think that we understand the commitments prior to understanding their triggering
conditions in virtue of being competent practitioners of making assertions. But this would be puzzling in the
context of his example involving the pitcher. Surely we don’t understand that commitment merely in virtue
of being competent practitioners in some practice. Furthermore, this reading of MacFarlane’s aims requires
a significant departure from his dialectical strategy. For taking this line requires that our practice of assertion
actually involves relativist commitments – i.e. we are competent at asserting relativist propositions. While
MacFarlane does in many places (2003; 2005b; 2008b; 2011a, 2014) try to make the case that various
fragments of natural language have relativist semantics, he wants to show in MacFarlane 2005a that the
notion of truth relative to a context of assessment makes sense independent of whether or not one is a
relativist about any actual fragments of language.
17
See Chapter Three for a full discussion of the speech act expression. For now the reader can just think of
expression in the standard way that we talk about a person expressing her pain, gratitude, condolences, etc.
18
I will not be discussing so-called hybrid expressivist views – views on which an utterance results in both an
assertion of a proposition and an expression of another mental state. For hybrid views about normative
19
that uses of ‘might’-sentences express some mental state not of belief, like ordinary assertions,
but of uncertainty:
Expressivism: The direct speech act performed by a speaker who assertively utters
┌
It might be that S
┐
, where p is the content of S in the context, is an expression of the
speaker’s mental state: leaving p open.
19
I’m taking this as the paradigm Expressivist view. It is possible to have something very much
like the above view which says that some other non-belief attitude is expressed.
20
Since they don’t think that propositions are asserted when speakers forcefully utter
‘might’-sentences, Expressivists can’t give an account of disagreement in terms of believing
contradictory propositions. Instead, they can say that two people disagree if one leaves p open
and the other rules out p. I will assume this without argument here.
21
In what follows we’ll
see that Expressivism, together with The Direct Approach, achieves mixed results – indeed the
same mixed results that Relativism and The Direct Approach achieved – in explaining the
different sorts of uses of ‘might’.
Together, Expressivism and The Direct Approach seem to be able to explain
(Cookies). According to Expressivism, Phil expresses his mental state of leaving it open that
Angelica is the thief. Lil believes that Angelica is not the thief – she rules it out that Angelica
is the thief. Given this, she disagrees with Phil. According to The Direct Approach, Phil’s
expression is the at-issue speech act. So, Lil can respond negatively to it.
But (Battleship) presents exactly the same challenge for Expressivism as for Relativism.
Remember that Phil and Lil are playing battleship and that Lil, in fact, has a PT boat, not a
battleship, on C4 and knows she does:
(Battleship) Phil: There might be a battleship on C4.
Lil: Yeah. There might be.
We need an explanation of why Lil’s response is appropriate. Expressivism says that Phil
expresses his mental state of leaving it open that there is a battleship on C4. The Direct
Approach says that this expression is the only at-issue speech act. Lil believes that there is not
a battleship on C4. She rules out that there is a battleship on C4. So, she should offer a
negative response. Nonetheless, her positive response is appropriate.
language, see Ridge 2006 and Boisvert 2008.
19
Compare with Yalcin 2007; 2011.
20
For example, the attitude expressed might be a credence as in Swanson forthcoming or an attitude of taking p
as a serious option in reasoning and deliberation (see Chapter Three for more on this attitude).
21
I will discuss disagreement of this general sort, which is related to what Stevenson (1937) calls disagreement in
attitude, in Chapter Three.
20
Expressivists who endorse The Direct Approach may try to give analogous replies to
the ones Relativists can give. One could have a mixed account where some uses of ‘might’ are
given an Expressivist interpretation and some are given a Contextualist one. And
Contextualism could be the operative explanation in (Battleship). Or, the Expressivist could
claim, just as in the second Relativist suggestion, that (Battleship) is not a standard
conversation. Really, Lil replies pretending that she has the same mental state as Phil. Under
this pretense, she doesn’t disagree in attitude with Phil and, so, can respond positively.
Above, I also noted that a Relativist might try to deal with (Battleship) by claiming that the
information of the context of assessment need not be the assessor’s, but could be the speaker’s.
Expressivists could, in a similar vein, say that Lil, the assessor, can express Phil’s mental state
and not her own. Such a move is, like the corresponding move for the Relativist, the most
promising, but also in need of serious development – a project that I will leave to others. In
general, the same reasons for doubting the analogous Relativist responses apply to these
Expressivist moves.
Instead of exploring these issues further, I will spend the next section sketching a
family of problems of compositionality for Expressivism itself (independent of The Direct
Approach). These will show that Expressivists not only need to do work to explain the data,
as argued in this section, they also need to do work to achieve a viable compositional
semantics.
7. The Frege-Geach Problem for Expressivism
Relativism and Expressivism offer identical predictions in the cases we’ve considered.
22

But Expressivism doesn’t share the same philosophical debt as Relativism, since it doesn’t use
the notion of truth relative to the body of information of a context of assessment. Instead, it
matches utterances of sentences with the mental states they are used to express. However,
Expressivism is traditionally thought to have a problem of its own – one that the Relativist
does not have. In what follows I’ll sketch two ways of thinking of this problem, corresponding
to the two popular ways that expressivist views are often developed. The problem – really a
family of obstacles having to do with compositionality – is traditionally known as the Frege-
Geach problem.
In what follows, I’ll be talking about the semantic value of a ‘might’-sentence. This is
22
However, it is plausible that they differ in their predictions about complex ‘might’-sentences (though John
MacFarlane (in a talk called “Relativism vs. Expressivism: the Case of Epistemic Modals”) has argued that his
Relativist view and Seth Yalcin’s (2007, 2011) Expressivist view offer the same results in all such cases).
21
in contrast to the discussion above, where I characterized Expressivism as a thesis about direct
speech acts, not semantics. The Frege-Geach problem is a problem for an expressivist
semantic theory, but we can also see it as a problem for the theory of direct speech acts,
Expressivism, as long as we accept a principle that says that the object of the direct speech act
of a forceful utterance of a ‘might’-sentence is the semantic value of that sentence (a plausible
general, though perhaps not universal, assumption).
In my sketch of the Frege-Geach problem, I follow (and owe a great debt to) Mark
Schroeder (2010; MS; pc). Suppose our expressivist account says that the semantic value of
“Angelica might be the thief” is the mental state of leaving it open that Angelica is the thief.
What should we say about the semantic values of larger sentences in which that sentence is
embedded? For example, consider the sentence, “Either Angelica was out of town or she
might be the thief.” What is the semantic value of this sentence? One reason this is a difficult
question is that ‘or’ is usually thought of as denoting an operator on pairs of propositions.
But the expressivist isn’t starting with propositions. Instead, the ingredients are attitudes. So
there is a question of how to give a general account of disjunction that can make sense of such
sentences. What sort of attitude do disjunctive ‘might’-sentences express? I’m going to
explore the two popular sorts of answers from the literature.
The first is what we can call a higher-order account – where a disjunctive ‘might’-
sentence expresses a higher-order attitude about the attitudes expressed by the constituent
simple sentences. We could follow Simon Blackburn (1988) in saying that “Either Angelica
was out of town or she might be the thief” has as its semantic value the mental state of being
tied to a tree between believing Angelica was out of town and leaving it open that she is the
thief.
23
Blackburn (1988, 512) says that being tied to a tree between two states is a
“commitment . . . to accepting the one branch should the other prove untenable.”
Now that we have a hypothesis about the semantic value of “Either Angelica was out of
town or she might be the thief”, we need to make sure that we have assigned the right semantic
value to the sentence. This value should allow us to predict the right inferential and
inconsistency properties. For example, the expressivist account needs to explain why the
following argument is valid:
(1) Either Angelica was out of town or she might be the thief.
23
This approach requires accepting the contentious position that all declarative sentences of English have
mental states as their primary semantic values (otherwise, connectives, like ‘or’ would have to be multiply
ambiguous). For example, it must say that the semantic value of “Angelica was out of town” is the belief that
Angelica was out of town.
22
(2) Angelica was not out of town.
(3) So, she might be the thief.
Given that the semantic values of these sentences are mental states and not something truth-
conditional, expressivists need a novel explanation of validity (in this example, as well as with
the other connectives, attitude verbs, and quantifiers). They can’t explain the validity of an
argument in terms of the truth of the semantic values of the premises guaranteeing the truth of
the semantic value of the conclusion. Instead, they must explain the validity of an argument
in terms of the rational relations between the attitudes that are the semantic values of the
sentences of the argument.
For our purposes we can simply consider one feature (and not the whole story) of the
expressivist’s account of validity. If an agent is tied to a tree between mental state P and
mental state Q and the agent is in a mental state that conflicts with P, then the agent is
rationally committed to coming to have mental state Q. We can apply this to the argument
from above:
(1) Either Angelica was out of town or she might be the thief.
(2) Angelica was not out of town.
(3) So, she might be the thief.
(1) has as its semantic value the mental state of being tied to a tree between believing that
Angelica was out of town and leaving it open that she is the thief, (2) has as its semantic value
the belief that she was not out of town, and (3) has as its semantic value the mental state of
leaving it open that Angelica is the thief. Because these states stand in the rational relation
just noted, the expressivist predicts that the argument is valid – as we desire.
But, as Bob Hale (1993) and Mark van Roojen (1996) show (and is brought out clearly
in Schroeder 2010, which I follow), this account makes too many inferences valid. It seems
that we can introduce a predicate, “are TTed”, which, when applied to a pair of attitudes
expresses the attitude of being tied to a tree between those attitudes. If we do this, then (1) and
(1') express the same attitude – being tied to a tree between believing that Angelica was out of
town and leaving it open that she is the thief:
(1') Angelica was out of town and Angelica might be the thief are TTed.
But then, the expressivist predicts that the following argument is valid for exactly the reason
(1)-(3) was:
(1') Angelica was out of town and Angelica might be the thief are TTed.
(2) Angelica was not out of town.
23
(3) So, Angelica might be the thief.
But this argument doesn’t seem valid. It doesn’t have the right sort of structure to be valid.
So, we should follow van Roojen and Hale in concluding that this expressivist account
overgenerates predictions of validity (and other logical relations). Since the expressivist’s
project was to give an explanation of our ordinary judgments about validity, it is incorrect.
24
It seems that we can do better if we pursue a different way of developing expressivism –
one taken by Seth Yalcin (2007; 2011) (who draws heavily from Alan Gibbard’s (1990; 2003)
metaethical expressivism). My discussion of and objection to Yalcin’s view largely follow
Schroeder MS. Yalcin’s aim is to recover the virtues of the truth-conditional account of
complex sentences in an expressivist system. He starts with the traditional possible world
model of truth conditional semantics. In such a model we can characterize the meaning of a
sentence in terms of the conditions under which it would be true – i.e. by the set of possible
worlds in which it is true. Yalcin attempts to enrich the framework so that it admits of an
expressivist interpretation, while preserving the compositional account. Instead of using a set
of worlds to represent the meaning of a sentence (the set in which the sentence is true), he
uses a set of pairs, the first member of which is a possible world and the second member of
which is an information-state – represented as a set of worlds.
25
Roughly, the first element of
each pair models the descriptive aspect of meaning while the second models the epistemic
modal (non-descriptive) aspect.
For instance, Yalcin represents the semantic value of the sentence “Angelica was out of
town” by the set of all and only the world/information-state pairs where the first element is a
world in which Angelica was out of town and the second is any information-state whatsoever.
Since “Angelica was out of town” is a non-modal sentence, there is a constraint on worlds but
not on information-states. On the other hand, a sentence like “Angelica might be the thief”
places no constraints on the world element of the pair. This accords well with the expressivist
claim that epistemic modal sentences don’t describe or state facts. The sentence’s semantic
value includes all and only pairs in which the second element – the information-state –
contains a world in which Angelica is the thief.
24
One might wonder why the expressivist is committed to the claim that (1) and (1') have the same semantic
value. All the argument requires, however, is that we can introduce a predicate that serves the function
mentioned above. It’s hard to see how the expressivist could resist that this is possible. The most salient way
to resist involves saying that only logical connectives could express states of being tied to a tree. But the
expressivist is trying to construct an expressivist logic. And it doesn’t seem that she could rely on the notion
of a logical connective in doing this without assuming the very notion she is trying to explain.
25
Strictly speaking, he represents semantic values as functions from world/information-state pairs to truth-
values. But this is equivalent to the formulation in terms of sets of such pairs – a way of stating it which
avoids the strange notion of truth that Yalcin employs. See note 26.
24
The purpose of this sort of semantic value is not to characterize the conditions under
which the sentence is true – in the sense of correctly describing some fact about the world.
26

Rather, such a semantic value constrains what an agent’s belief-state must be like in order to
accept the sentence. This notion of acceptance connects to our our ordinary linguistic
practices as follows. If an agent’s belief-state accepts a sentence, then it is semantically
appropriate for the agent to utter the sentence and assent to its use.
So, we need to understand how to represent, in the system, a belief-state accepting a
sentence. Let’s put Yalcin to the side briefly, and start with the standard semantic picture,
where the meanings of sentences are characterized by sets of possible worlds. An agent’s
descriptive belief-state can also be characterized by a set of possible worlds – those in which all
of her beliefs are true. In this system, for a belief-state to accept a sentence is for the worlds
that characterize the belief-state to be a subset of the worlds that characterize the semantic
value of a sentence. In such a case, it is appropriate for an agent who is in that belief-state to
utter the sentence or assent to its use.
What happens in Yalcin’s system is simply that the formal objects are enriched from
sets of worlds to sets of world/information-state pairs. So, the meaning of a sentence is
characterized by a set of world/information-state pairs. And a belief-state is, likewise,
characterized by a set of world/information-state pairs.
27
For a belief-state to accept a sentence
is for that belief-state to be characterized by a set of worlds that is a subset of the set of worlds
that characterize the semantic value of the sentence. In such a case, it is appropriate for an
agent who is in that belief-state to utter the sentence or assent to another’s utterance of it. In
this way, the sentence’s meaning is cashed out in terms of the belief-states that accept the
sentence. This is the sense in which the sentence can be said to express that belief-state.
Now that we have seen how Yalcin’s expressivist account models meaning without
giving factual truth-conditions, we can proceed to see how it handles embedding. In the
traditional system, we can make sense of the negation of a sentence as taking the complement
of the set of worlds that is its semantic value. Disjunctions are unions and conjunctions are
intersections. Yalcin carries this over to his system – the difference being that the elements of
the sets we are complementing, joining, and meeting are world/information-state pairs.
26
Yalcin (2011) discusses how his system does characterize meaning in terms of truth in a minimal sense, which
does not imply any facts about the world. Since my claims can be made without taking a stand on what
exactly this non-standard sense of truth is, I will ignore it in what follows.
27
It is natural to think that for any non-defective belief-state, these pairs are generated in the following way.
Start with an agent’s descriptive beliefs, which can be represented by the set of worlds in which each of those
beliefs are true. Take a world in the state and pair it with the state as a whole. Do this for each world in the
set characterizing the agent’s descriptive beliefs. The set of all such pairs characterizes the agent’s belief-state.
25
On this account, the semantic value of “Either Angelica was out of town or she might
be the thief” is a set of world/information-state pairs each of which either contains a world in
which Angelica was out of town or an information-state which leaves it open that she is the
thief. Consider, now, the sentence “Angelica was not out of town.” Its semantic value
contains all world/information-state pairs in which the world is one where Angelica is not out
of town. Any belief-state that accepts this sentence, in addition to the above disjunction
contains only pairs having an information-state that contains a world in which Angelica is the
thief. That is, we are left with belief-states that leave it open that Angelica is the thief. These
are states that accept “Angelica might be the thief.” So, we have an explanation of the
inference from above (from (1) and (2) to (3)). Furthermore, since we didn’t greatly alter the
semantics of the connectives, we can hope that this picture doesn’t overgenerate inferences in
the way the higher-order account did.
Yalcin’s account does not overgenerate inferences in the same way the the higher-order
account did, but it still has an overgeneration problem.
28
To see this, let’s work with a
sentence involving ‘must’ rather than ‘might’.
29
Before getting to the example, we need to
appreciate two things. First, on Yalcin’s account a ‘must’-sentence, like “Angelica must be the
thief” is like a ‘might’-sentence in that its semantic value is a constraint on an information-
state. But it is unlike a ‘might’-sentence in that it places a different constraint on an
information-state. It requires an information-state to contain only worlds in which Angelica is
the thief (while, of course, being neutral about the world parameter of the semantic value).
The second important point is not about semantics, but about representing belief. We are
representing an agent’s belief-state as a set of world/information-state pairs. An agent must
have the same information-state element in each of these pairs. This is because an
information-state is the set of worlds left open by her beliefs, and her beliefs determine a
single set of information.
With these two facts in hand, we are in a position to see the problem for Yalcin’s
28
There are problems in addition to the one that I discuss in the text. Given natural accounts of ‘believes’ and
‘not’, it turns out that for any proposition, an agent either believes that it might be so or believes that it’s not
the case that it might be so. However, one can surely fail to believe either of these simply by never having
considered the proposition in question or, perhaps, by being irrational in some way. Yalcin (2011) sees this
problem and complicates his system in order to avoid this worry – relativizing belief-states to questions. I’m
unsure whether this is a motivated solution or ad hoc fix. It is worth noting, in line with Schroeder MS, that
Yalcin’s complication offers a way of trying to skirt the problem I discuss in the text – however, only at the
cost of saying all agents who believe a mixed descriptive/modal disjunction, like “Either someone else did it
or Angelica must be the thief”, without believing either disjunct are not fully rational or reflective. But, this is
wildly implausible.
29
The issue of whether the problem obtains for ‘might’ is not straightforward, and I’ll avoid it here. The
phenomenon does obtain for ‘likely’. See Chapter Six for more on ‘likely’ and ‘must’.
26
system. Consider the sentence, “Either someone else did it or Angelica must be the thief.” An
agent can appropriately utter or assent to an utterance of this sentence without being in a
position to appropriately utter or assent to an utterance of either “Someone else did it” or
“Angelica must be the thief” (indeed this is the natural case where we would give voice to the
disjunction). But Yalcin’s account precludes this. On his account, each pair in the semantic
value of this disjunction contains either a world in the first argument in which someone else
did it or an information-state in the second argument which contains only worlds in which
Angelica is the thief. In order for an agent’s belief-state to accept the disjunction, it must
consist only of pairs of that form. As we saw, an agent will have the same information-state in
every pair. Either this state contains only worlds in which Angelica is the thief or it doesn’t. If
it does, then we merely have a case of accepting a disjunction in virtue of accepting one of the
disjuncts (“Angelica must be the thief”). If it doesn’t, then, in order to accept the disjunction,
the agent’s belief-state must consist only of pairs in which the world argument is one where
someone else did it. But this means that the agent’s belief-state accepts that someone else did
it – another case of accepting the disjunction in virtue of accepting one of the disjuncts
(“Someone else did it”). So, it seems that on Yalcin’s account, one can only appropriately
utter or assent to this sort of mixed factual/modal disjunction if one can appropriately utter or
assent to one of the disjuncts. This is a bad result, since it seems that we often do utter and
assent to such disjunctions without being in a position to utter or assent to either disjunct.
This is, in fact, the standard scenario in which a speaker appropriately utters a disjunction.
So, it looks like Yalcin’s theory cannot, at least without further complications, explain all of the
ways in which epistemic sentences embed. I worry that any addition to the theory that may
get around this problem will be ad hoc, though I won’t spend more time discussing this here.
30
The various sorts of hurdles that we’ve seen in this section, together, show that the
expressivist has a difficult task in front of her – to explain, in a non-ad hoc way, how the
meanings of simple epistemic sentences combine to yield the meanings of complex epistemic
sentences. I take this to be sufficient motivation to leave Expressivism to the side and
investigate what other resources Contextualists have.
8. Taking Stock
30
For more on this issue and other problems of this sort (especially involving quantifiers) see Schroeder MS.
As Schroeder notes, a kindred theory to expressivism – dynamic semantics – can likely explain this sort of
problem. The reader can also consult Marushak and Shaw MS for arguments against expressivism based on
embeddings under non-acceptance attitudes.
27
We started this chapter with some conversations in which ‘might’-sentences are used
and these uses are responded to. We assumed a theory – The Direct Approach – about what
features of an utterance of a ‘might’-sentence are at-issue, which says that the direct speech act
that the speaker performed in uttering the ‘might’-sentence is the only at-issue speech act. In
§3, we saw that Contextualism paired with The Direct Approach could not explain (Cookies).
But, in §4 and §6, we saw that the two main alternatives to Contextualism – Relativism and
Expressivism – paired with The Direct Approach require serious complications to have a
chance at explaining all conversations involving uses of ‘might’-sentences. This is reason to
doubt The Direct Approach itself. In the rest of the dissertation, I will look beyond The
Direct Approach for a unified explanation of the pragmatics of ‘might’-communication.
31
In
Chapter Two, I am going to survey a number of theories of at-issue speech acts that we can
choose to put in its place. These will loosen the requirements on what it takes for something
to be an at-issue speech act.
We’ve also seen independent reasons to doubt Relativism and Expressivism.
Relativism has the conceptual obstacle of making sense of relative truth. Expressivism has the
obstacle of giving an adequate compositional theory for a language. I have not decisively
refuted either theory, but I have shown that there is little reason to think that either can
satisfactorily overcome their obstacles in a principled manner. It is worth noting that these
obstacles for Relativism and Expressivism don’t arise for Contextualism. The sort of context-
dependence it ascribes to ‘might’-sentences, unlike Relativism, is perfectly ordinary and
philosophically unproblematic. For the Contextualist, ‘might’ works very much like other
context-sensitive terms. Just like quantifiers can be used to quantify over different domains in
different contexts, ‘might’ can be used to relate a proposition to different bodies of information
in different contexts. And the resulting propositions about consistency are true or false
absolutely, not relative to any further parameter. Contextualism also doesn’t run into any of
the technical obstacles that Expressivism does. Since ‘might’-sentences have ordinary
propositions as their semantic values, we can use our best compositional theory to explain
31
I haven’t shown that every possible theory of ‘might’-communication, paired with The Direct Approach is
unsatisfactory. Indeed, there are at least two extant theories that I haven’t dealt with. The first is a dynamic
theory of ‘might’-communication, exemplified by Veltman 1996 and Willer 2011; 2013 and explored in von
Fintel and Gillies 2007. On this theory, a use of a ‘might’-sentence is a test on the information-state of the
context. I find this view attractive in many ways, but it faces the same sort of inflexibility problems that
Relativism and Expressivism do in cases like (Battleship). The second sort of theory that I haven’t discussed
is a speech act modifier theory, like the ones espoused by Price (1983) and Schneider (2010). According to
this theory, the direct speech act performed by a speaker who uses a ‘might’-sentence is a partial speech act
with the prejacent as it’s content. This view faces the Frege-Geach problem in a much more pressing way
than the versions of expressivism that I surveyed above.
28
how the meanings of complex ‘might’-sentences are composed out of the meanings of simple
ones together with the meanings of other linguistic expressions and general compositional
rules. So, Contextualism avoids the problems of the other two theories.
Because of this, we should return to Contextualism, exploring whether it, free of the
constraints of The Direct Approach, can serve as part of an adequate explanation of ‘might’-
communication. In the next chapter, I’ll pair Contextualism with progressively more liberal
theories of at-issue speech acts. But, I’ll find all the theories surveyed there to be wanting. In
Chapter Four, I will present my own view, which retains Contextualism but pairs it with some
of the resources – though not the letter – of Expressivism in order to explain conversations like
(Cookies).
29
Chapter Two: More Liberal Theories of At-Issue Speech Acts
In Chapter One, I introduced a framework for understanding conversations involving
uses of and responses to uses of ‘might’-sentences. The idea is to pair a theory about what
speech acts can be at-issue – available for response and responsible for our judgments of
appropriateness – with a theory about the speech acts performed by speakers who use ‘might’-
sentences. In Chapter One, we saw that two candidate theories of the direct speech act
performed by an assertive utterance of a ‘might’-sentence – Relativism and Expressivism –
confront serious obstacles. Because of this I’ve decided to put Relativism and Expressivism to
the side for the remainder of this dissertation, focusing on Contextualism. The other main
result of Chapter One was that the theory of at-issue speech acts that we surveyed – The
Direct Approach – is too restrictive.
Let’s review the two main theses from which we will start our investigation. The first is
a theory about the direct speech act performed by an assertive utterance of a ‘might’-sentence:
Contextualism: The direct speech act performed by a speaker who assertively utters
┌
It might be that S
┐
, where p is the content of S in the context, is an assertion of the
proposition that p is consistent with the information of the context.
Remember that we called the sort of proposition that is asserted on this view – a proposition
that the prejacent is consistent with some body of information – a ‘might’-proposition, or
MP.
32
We’ll start from The Direct Approach :
The Direct Approach: The direct speech act performed by a speaker who assertively
utters a ‘might’-sentence is the (only) at-issue speech act.
In this chapter, I will entertain a number of alternative theories of at-issue speech acts.
I will start with an attempt to revive the project of explaining conversations involving ‘might’-
sentences with the combination of Contextualism and The Direct Approach. Then I will
draw on the literature in forming more liberal theories of at-issue speech acts. I will argue that
these theories are incorrect. The upshot is that we need to look in a somewhat different place
to find the correct theory of at-issue speech acts – we need to allow at-issue speech acts that
don’t have propositional content. In Chapter Four I will develop such a theory.
1. Re-evaluating Contextualism and The Direct Approach
In this section I’ll re-evaluate whether the combination of Contextualism and The
32
Here and in what follows, I use ‘proposition’ to talk about traditional propositions – ones that are true and
false simpliciter. I do not use it to talk about the sorts of objects that relativists call propositions – objects
that are true or false only relative to some further parameter.
30
Direct Approach can explain the range of conversations in which ‘might’-sentences are used
and those uses are responded to. I’ll briefly rehearse the argument I gave in Chapter One.
Then I’ll relay a challenge to the argument, made by Janice Dowell (2011). I’ll conclude that
Dowell is wrong to reject the argument.
Let’s revisit (Cookies) – the conversation that was a problem for the combination of
Contextualism and The Direct Approach. There were 10 cookies in the cookie jar this
morning. In the afternoon the cookies are gone. Lil and Susie are investigating the cookie
theft. They know that Angelica isn’t the thief. But Phil doesn’t know that Angelica isn’t the
thief. Phil walks in:
(Cookies) Susie and Lil: Phil, we’re interviewing people to get information about
who stole the cookies. Do you know who stole them?
Phil: No, but Angelica might be the thief.
Lil: No. She can’t be. I was with her all afternoon.
In Chapter One, I considered two bodies of information that could be the information of the
context – Phil’s and the group’s. However, on either reading, we don’t get the result that both
Phil’s utterance and Lil’s reply are appropriate. If Phil’s information is relevant, then
Contextualism says he asserts the MP that it is consistent with his information that Angelica is
the thief – something he is warranted in asserting. But, according to The Direct Approach,
this proposition is the only at-issue speech act. And it doesn’t license Lil’s negative reply since
she knows it is true. On the other hand, if the group’s information – that is, the information
distributed across members of the group – is relevant, then Contextualism says Phil asserts the
MP that it is consistent with the group’s information that Angelica is the thief. But Phil
doesn’t know, and may not even have any beliefs about, what is consistent with the
information of the group. So, according to The Direct Approach, his original use of the
‘might’-sentence shouldn’t be warranted. Thus, whatever we say about what information is
relevant, the combination of Contextualism and The Direct Approach can’t explain why
(Cookies) is appropriate.
This diagnosis of (Cookies) has been challenged by Janice Dowell (2011). She thinks
that what information is relevant in a context depends on the intentions of the speaker of the
context. So, the MP that the speaker asserts in uttering a ‘might’-sentence depends on what
information the speaker intends to be talking about. For us to have robust judgments about
the appropriateness of conversations like (Cookies), the speaker’s intention must be manifest.
But, Dowell thinks that this is often not the case and that our judgments really aren’t that
strong. In cases where the speaker’s intentions are manifest, she hypothesizes that we will
31
have judgments that track the predictions of the combination of Contextualism and The
Direct Approach.
33
Let’s illustrate her point about manifest intentions with some examples. Suppose that
Phil is alone in the playpen mulling over his own evidence about who stole the cookies and
mutters to himself, “What do I think? Well, Angelica might be the thief.” And suppose that
Lil is in the sandbox mulling over her own evidence about who stole the cookies and mutters
to herself, “What do I think? Well, Angelica can’t be the thief.” Let’s assume that in this
case, both of their intentions are manifest – they both intend to talk about their own
information. Then Dowell will think that we don’t have the judgment that Phil and Lil
disagree (I’ll grant this here, though I’m not sure about this judgment). This is as
Contextualism predicts – since it says that Phil and Lil don’t assert contradictory propositions.
Thus, Dowell will say, the judgment matches the prediction of the theory.
We can contrast this with an example that we have already encountered. Holmes and
Watson are investigating a murder. They list all of the evidence they have, which
straightforwardly rules out that Moriarty is the murderer. Watson, however, does not do a
good job of deducing what obviously follows from this evidence. They have the following
conversation:
(Murder) Holmes: What does our evidence show? Who are the remaining
suspects?
Watson: Well, Moriarty might be the murderer.
Holmes: No. He can’t be. Our evidence rules it out.
Here, Dowell will say that Watson’s intention to say something about the group’s information
is manifest. And we have the judgment that Holmes’s response is appropriate (and that they
do disagree). According to Contextualism, he asserts that it is consistent with the group’s
information that Moriarty is the murderer. Given The Direct Approach, this proposition is
the sole at-issue speech act. Holmes appropriately responds to it – negatively. In this case,
again, our judgment lines up with the theory.
Dowell thinks that our judgment will line up with the predictions of Contextualism and
The Direct Approach in all cases. And in some of the most problematic cases, she claims that
we don’t have strong judgments because the speaker’s intentions are not manifest.
What will Dowell say in response to the argument I gave above? In general, she’ll
either say that our judgments about (Cookies) are not strong or she’ll say they are strong and
33
Sometimes it seems like Dowell (2011) is willing to accept a more liberal theory of at-issue speech acts than
The Direct Approach. Whether or not she accepts The Direct Approach, many of her arguments can be
seen as a defense of the combination of Contextualism and The Direct Approach.
32
this is because Phil’s (the speaker’s) intention is manifest. This latter option splits into two
possibilities. Either Phil’s intention is to say something about his own information – in which
case we should find Lil’s reply inappropriate. Or Phil’s intention is to say something about
the group’s information – in which case we should find Phil’s assertion inappropriate. But
none of these options provide the correct diagnosis of the case. Let’s go through each in turn.
First, are our judgments of the appropriateness of (Cookies) weak? I doubt it.
(Cookies) seems as natural as can be. For those who waver, make sure that you are judging
the appropriateness of the conversation independent of semantic theorizing. One should not
be trying to give a semantic paraphrase for every sentence and then see what the result sounds
like. One should simply read the scenario with no theory attached. In that case, I assume
that the reader will – without hesitation – judge (Cookies) to be appropriate.
Second, is it manifest that in (Cookies) Phil intends to make a claim relative to his
information and that Lil’s response is inappropriate? No. I do think that Phil intends his
own information to be relevant. After all, he is asked about what he knows, not about the
group’s information. But, it also seems that Lil’s response is perfectly appropriate. As many
Relativists have pointed out, it would be bizarre for Lil, instead, to say “You’re right. But,
Angelica isn’t the thief.” But, if Contextualism and The Direct Approach were correct, a
positive response would be the only one that Lil could give.
Third, is it manifest that in (Cookies) Phil intends to make a claim relative to the
group’s information and that his assertion is inappropriate? Again, no. Here I think that Phil
doesn’t intend to make an assertion about the group’s information (again, he is asked about
what he knows). His assertion is completely warranted and appropriate. If he really intended
to make an assertion about the group’s information, we should expect the following
conversations to be equally appropriate, but they are not:
(Cookies Warrant) Susie and Lil: Phil, we’re interviewing people to get information
about who stole the cookies. Do you know who stole them?
Phil: No, but Angelica might be the thief.
Lil: # What makes you say that? You aren’t in a position to say
what is consistent with Susie and my information.
(Cookies Warrant*) Susie and Lil: Phil, we’re interviewing people to get information
about who stole the cookies. Do you know who stole them?
Phil: No, but it is consistent with the group’s information that
Angelica is the thief.
Lil: What makes you say that? You aren’t in a position to say
33
what is consistent with Susie and my information.
Lil’s retort is appropriate in the latter, but not the former conversation. This suggests that
Phil’s assertion is unwarranted only in the latter conversation. And this suggests that he
asserts different things – i.e. that he does not assert a proposition about the group’s
information by uttering the ‘might’-sentence.
34
We’ve surveyed Dowell’s three options and seen that none adequately describe the
case. But, one of these options would have to do so if the combination of Contextualism and
The Direct Approach were right. The reader may feel some pressure at this point to go back
and re-evaluate each of the options, thinking one of the three options just has to be right.
This leads one to start doing semantic calculations in one’s head, resulting in theory-laden
judgments. Instead, what we need to remember is that The Direct Approach is a very strong
thesis. It says that, given an utterance of a ‘might’-sentence, there is only one speech act that
is at-issue. But in actual cases of communication, pragmatic factors abound. So, it is no
wonder that accepting The Direct Approach leaves us unable to explain all conversations
involving uses of ‘might’-sentences. Before looking at some more liberal theories of at-issue
speech acts, I want to give a different argument against The Direct Approach – one that brings
out its restrictiveness.
This problem for the combination of Contextualism and The Direct Approach comes
from conversations where multiple responses are appropriate. We saw an example of such a
conversation in Chapter One. Suppose that Phil and Lil are playing battleship and that Lil, in
fact, has a PT boat, not a battleship, on C4 and knows she does:
(Battleship') Phil: There might be a battleship on C4.
Lil: Yeah. There might be. / No. There can’t be.
Given a particular utterance of the first sentence by Phil, Lil can respond in either of the two
ways listed. But The Direct Approach says that there is only one at-issue speech act for any
given utterance – the direct speech act. So, Contextualism and The Direct Approach can’t
explain the availability of multiple appropriate responses – one positive and one negative.
(Battleship') requires that there are at least two at-issue speech acts. So, The Direct Approach
can’t be correct. Dowell’s observations about manifest intentions and wavering judgments are
of no help here. We simply need more at-issue speech acts.
The attempt to salvage the combination of Contextualism and The Direct Approach
has failed. We saw two related but distinct problems. First, conversations like (Cookies) show
that hearers can respond to something other than the proposition that a speaker asserts with a
34
See Lennertz MS c for further discussion of this argument.
34
speaker-centric use of ‘might’. Second, sometimes a hearer can appropriately respond either
positively or negatively to a particular utterance of a ‘might’-sentence. The first problem
suggests that there is an at-issue speech act that is not the proposition that the speaker asserts.
The second problem suggests that, in some cases, there are at least two at-issue speech acts.
In the following sections I will canvass some replacement theories of at-issue speech acts that
heed these lessons.
2. The Implicature Approach
In this section I will introduce a more liberal theory of at-issue speech acts where a
hearer can respond not just to what a speaker asserts but also to what she implicates. Though
this theory is incorrect (as I will show), it will give the reader an idea of what alternatives to
The Direct Approach look like.
Before presenting the theory of at-issue speech acts, I will briefly explain conversational
implicatures. Often, when a speaker utters a sentence, she does not just make an assertion.
She also suggests or implies other things. A large swath of such suggestions fit under Grice’s
(1989b) model of conversational implicature. We can characterize conversational implicatures
as follows. A conversational implicature is a speech act which has a content that is not the
semantic content of the sentence that the speaker utters. For a speech act to be a
conversational implicature, a hearer must be able figure out what a speaker implicated even
though the sentence he uttered meant something different. She must be able to do this by
relying on what the speaker did say together principles of rational cooperation.
35
Let’s look at
a classic example from Grice to illustrate the phenomenon. Lil is standing by an obviously
immobilized car and is approached by Phil; the following exchange takes place:
(Petrol) Lil: I am out of petrol.
Phil: There is a garage round the corner.
As Grice notes, Phil suggests that the garage is open and carries petrol. The hearer, Lil, can
figure this out by relying on what Phil said together with the assumption that Phil is being
cooperative.
The full details of how conversational implicatures work aren’t important here. What
is important is that cases where there are implicatures are cases where the speaker performs
multiple speech acts – an assertion and an implicature. We might think, then, that both can
be at-issue speech acts:
36
35
Grice (1989b) doesn’t require that hearers actually perform this sort of derivation – just that such a derivation
is possible in every case.
36
Schaffer (2011, 219) likewise suggests that the appearance of disagreement in cases involving epistemic modals
“might only concern implicatures (or other matters downstream from the propositions at-issue).”
35
The Implicature Approach: An assertive utterance of a ‘might’-sentence gives rise to
multiple at-issue speech acts – the direct speech act that the speaker performed and any
implicatures that the speaker performed.
Using this theory of at-issue speech acts, we may be able to give explanations of the
problematic conversations – eg. relying on an at-issue speech act other than the asserted
content in (Cookies). And in many conversations there will be multiple at-issue speech acts,
which is what we need to explain conversations like (Battleship').
To make good on these explanations, we would have to say what is implicated in each
conversation and show how this would allow us to explain the appropriate response or
responses. I’m not going to undertake this task here because there is a more fundamental
problem with The Implicature Approach. Implicatures are not, in general, at-issue speech
acts. So, The Implicature Approach is not well-motivated as a general theory.
Let’s consider a variation on our example from above to see that what is implicated
isn’t generally an at-issue speech act. Lil is standing by an obviously immobilized car and is
approached by Phil. But in this variation Phil’s friend Dil is walking with him. Dil knows
that there is a garage around the corner. But he also knows that it is out of petrol. Consider
the following conversation:
(Petrol') Lil: I am out of petrol.
Phil: There is a garage round the corner.
Dil: # No (/You’re wrong). The garage is out of petrol.
Dil’s response is inappropriate. Phil implicates that the garage is open and sells petrol. But it
appears that Dil cannot respond to Phil’s implicature – otherwise his response would be
appropriate.
37
This is evidence that the implicature is not an at-issue speech act. The reader
can construct similar examples with her favorite cases of conversational implicature to see that
implicatures are not, in general available for response.
38
So, I propose to move on from The Implicature Approach. I should note, however,
that there may be some more-specific types of implicatures that are at-issue speech acts. I
have not ruled out this possibility. And I don’t want to speculate on what this means for
some amended version of The Implicature Approach. I just want to note that insofar as The
Implicature Approach is based on the idea that implicatures, in general, are at-issue speech
37
To respond to an implicature, one must choose some more indirect route. Dil could say, “Yeah. There’s a
garage there. But it’s out of petrol. There’s a garage with petrol three blocks down on the left.”
38
The same phenomenon obtains with conventional implicatures (if there are such things). A hearer cannot
appropriately respond to an utterance of “He’s poor but brave” by saying “No. There is no anti-correlation
between poverty and bravery.”
36
acts, the theory is not well-motivated.
39
3. The Hearer’s Context Approach
There is another theory that I want to reject for similar reasons – the principles behind
it make general predictions about when speech acts are at-issue that are not borne out. The
theory is motivated by Gunnar Björnsson and Stephen Finlay’s (2010) account of the deontic
modal, ‘ought’. Björnsson and Finlay think deontic ‘ought’-sentences are context-sensitive
with respect to information in the same way that Contextualism says that epistemic ‘might’-
sentences are. So we can easily adapt their idea for the case of ‘might’.
40
So-adapted, the idea
is that a hearer can respond to a proposition that is not the object of any speech act that the
speaker performed in uttering a ‘might’-sentence. This proposition is the one that the speaker
would have asserted if he uttered the sentence in the hearer’s context.
41
More precisely:
The Hearer’s Context Approach: An assertive utterance of a ‘might’-sentence gives rise
to the following at-issue speech act: the assertion that would be performed if the hearer,
in her context, assertively uttered the sentence that the speaker uttered.
The hope is that this theory will be both plausible and able to explain the responses in all of
the conversations. I will leave it to the reader to show how Contextualism and The Hearer’s
Context Approach can, indeed, explain the appropriateness (Cookies).
42
I want to focus,
instead, on the general plausibility of The Hearer’s Context Approach as it features in an
account of communication.
It is important to realize that The Hearer’s Context Approach does not lead to or even
suggest Relativism. It is not a thesis about what is conveyed by an assertive use of a ‘might’-
sentence. It is merely a thesis about what a hearer can respond to. And so, The Hearer’s
Context Approach is compatible with Contextualism (indeed Björnsson and Finlay take this
sort of approach to be central to their defense of a thesis like Contextualism). Nonetheless,
The Hearer’s Context Approach is a radical thesis in its own right. It takes a radical view of
39
Thanks to my audience at the Meaning, Context, and Implicit Content Conference in Cerisy, France, 2011 –
in particular, Tim Sundell, Thomas Hodgson, and Manuel Garcia-Carpintero – for discussion of this issue.
40
Björnsson and Finlay (2010, 36) say, “we also suspect that many of our strategies generalize effectively to
other normative and modal terms.”
41
Björnsson and Finlay (2010, 21) say the following about ‘ought’:
What now becomes conversationally relevant . . . are other ought propositions [ones that were not
asserted], which are related to the agents’ utterances (or judgments) in the following way: they are the
propositions that the agents would have asserted by their utterances if they had rather been made in
the new, improved context, relativized to the new, superior information. In contexts of advice, we
evaluate previous ought claims as if they had been made in our present context, evaluating relevant
propositions rather than the original propositions.
(The text in the square brackets is my clarification, while the text in the parentheses is in the original.)
42
The Hearer’s Context Approach won’t, by itself, explain the multiple responses in (Battleship'). But if we
added the idea that the direct speech act was at-issue as well, we could explain the multiple responses.
37
communication. It is intuitive to think that in communication, speakers communicate things
and hearers respond to the things speakers communicate. The Hearer’s Context Approach
does not respect this idea. It says that hearers can appropriately reply without responding
directly to anything that the speaker communicated. Rather, it prompts the hearer to take the
sentence that the speaker uttered and think about what it would be used to assert in the
hearer’s context. I think that we should be suspicious of this strategy. We can’t, in general,
respond to a speaker by responding to what would have been asserted in our context by
utterance of the sentence the speaker used:
(Hungry) Phil: I’m hungry.
Lil: # No (/You’re wrong). I’m not hungry.
Björnsson and Finlay (2010) attempt to justify The Hearer’s Context Approach by
saying that sometimes (though not in cases like (Hungry)) what best serves our conversational
purposes is not the proposition that the speaker asserted. Rather, it is the proposition that the
speaker would have asserted if he had the hearer’s information. So what a hearer may
appropriately respond to is partially dictated by the purposes of the conversation. It is not
completely determined by the speech acts performed by the speaker.
43
Though they make this
point in the context of using the deontic ‘ought’ to give advice, it carries over quite easily to
uses of the epistemic ‘might’ – which function to advise agents about their epistemic states.
For this advice to be useful, they think that the at-issue speech acts in a conversation should
track not the speech acts that speakers actually performed but, rather, the ones they would
have performed given updated information. And, so, these speech acts are at-issue, given The
Hearer’s Context Approach. More generally, Björnsson and Finlay see cooperative
communication as part of a larger project of achieving a shared goal. Embedded in this larger
framework, it may, indeed, make sense that a hearer responds to a proposition that the
speaker did not communicate because it helps achieve the shared goal.
43
Björnsson and Finlay (2010, 21-22) say:
Since ought propositions relative to the best available information always provide agents with the
best available bases for deciding what to do, ought claims with these contents have the pragmatic role
of recommendations. Recommendations have to be endorsed or rejected as new information
becomes available. . . Our primary conversational interest in these speech acts is with this pragmatic
role, and therefore, when endorsing or rejecting them as recommendations or decisions, the salient
propositions for assessment are the ought propositions relative to the new, improved information.
There are two different threads here. First, there is the idea that at least one thing that speakers are doing
when they utter ‘ought’-sentences – and, by extension, ‘might’-sentences – is recommending or advising. I
want to put this point to the side for the following reason. Recommending and advising are speech acts.
Giving an account on which these speech acts are at-issue is not at all radical and doesn’t require adopting
The Hearer’s Context Approach. In this section, I am focusing on the truly radical claim – that because of
the purposes of the conversation, a hearer can respond to a speaker’s utterance by targeting a speech act that
was not performed by the speaker. This involves evaluating a proposition that the sentence would be used to
assert if uttered in the hearer’s context.
38
The problem is that there are cases where conversational participants are engaged in a
project where responding to a proposition that is not the object of any speech act the speaker
performed would help achieve the goal of the project. But, in such cases, the hearer may not
appropriately give such a response. In particular, some are cases where what is relevant is
what would be asserted by a use of the sentence uttered in the hearer’s context but the hearer
still cannot appropriately respond to such a proposition. She can only respond to a speech act
that the speaker performed.
Consider the following case: Suppose that Chuckie is terrified of monsters and Tommy
wants to help him avoid them. They are talking over the phone. Chuckie is trying to decide
where to spend the night in order to avoid monsters. He can stay at his house or walk across
the street and have a sleepover at Phil and Lil’s. Chuckie’s dad does not want to drive him
over to Tommy’s house, and Chuckie knows this. They have the following conversation:
(Monsters) Chuckie: I’m trying to figure out where to spend the night to avoid
monsters. Where are there monsters?
Tommy: Well, there are no monsters here
Chuckie (seeing a monster emerge from under his bed): # Ahhh!.
You’re wrong. There is one under the bed.
This is a case where Tommy and Chuckie have a shared goal. What Tommy says is not
relevant to helping to achieve that goal (since spending the night at Tommy’s house is not an
option). What is relevant is the proposition that would be asserted by an utterance of
Tommy’s sentence, “There are no monsters here”, in Chuckie’s context. So, Chuckie should
be able to respond to this proposition. This case is exactly the sort of case in which, given
Björnsson and Finlay’s story, we should expect Chuckie’s response to be appropriate.
However, his response is clearly not appropriate.
44
This makes it look like general concerns of cooperation aren’t enough to allow hearers
to respond to the proposition that would be asserted by a use of the sentence in their context.
That is, it makes it look like the reason for accepting The Hearer’s Context Approach is not a
good one. Holding onto the theory absent of this motivation would be an ad hoc maneuver.
So, we should reject the theory (or at least put it to the side until given a reason to think
communication can work in the way it requires).
44
There are replies that Björnsson and Finlay might give. They might fall back on there being some other
feature of the conversation that makes the proposition that would be asserted by the sentence in Chuckie’s
context unavailable to be responded to. But it’s hard to see what such a feature would be (obviously, it can’t
be that it doesn’t further the goal of the conversation). Or they could restrict their story by saying that it only
applies to certain sorts of sentences. But, the account is supposed to be based on general pragmatic
principles. Restricting it to certain sorts of sentences would be inconsistent with what was supposed to be
plausible about the account in the first place.
39
4. The Prejacent Approach
In this section I will present a theory of at-issue speech acts that is more plausible as
part of an account of communication. This theory, together with Contextualism, succeeds in
explaining problematic cases like (Cookies) and (Battleship'). However, it cannot offer similar
explanations of cases where uncertain hearers give positive responses to utterances of ‘might’-
sentences. Furthermore, this replacement does not work in explaining conversations which
involve terms that are similar to ‘might’.
Contextualism says that when a speaker utters a ‘might’-sentence he asserts a
proposition about what is consistent with the relevant body of information in the context.
However, it is plausible that this isn’t all that the speaker does when he utters a ‘might’-
sentence. He doesn’t just perform a direct speech act of assertion. It is natural to think that
the prejacent is the object of some speech act he performs – a noncommittal one that serves
merely to raise a possibility. I’ll call such a speech act raising:
PREJACENT RAISING: A speaker who utters a sentence of the form
┌
It might be
that S
┐
, where p is the content of S in the context, typically raises p.
And PREJACENT RAISING is plausible. After all, in cases of ‘might’-communication,
conversational participants are really concerned with the prejacent. The speaker asserts an
MP because he isn’t in a position to assert the prejacent flat out. But he is in a position to
raise the prejacent. In the case of (Cookies), it is plausible that, in addition to asserting that it
is consistent with his information that Angelica is the thief, Phil raises the prejacent – the
proposition that Angelica is the thief. Of course, he doesn’t commit to her being the thief.
But he does put that proposition out there.
PREJACENT RAISING gives us additional resources for explaining conversations if
we liberalize our theory of at-issue speech acts. We need not restrict at-issue speech acts to
comprise just the direct speech act. We could claim that raising the prejacent of ‘might’ is an
at-issue speech act as well. Here is the theory:
The Prejacent Approach: An assertive utterance of a ‘might’-sentence gives rise to two
at-issue speech acts – the direct speech act that the speaker performed and the raising
of the prejacent.
45
It makes sense that a hearer would be able to focus on the prejacent in responding to
utterances of ‘might’-sentences. This is because our main concern is, almost always, the truth
45
PREJACENT RAISING and The Prejacent Approach are not novel suggestions. There are some
Contextualist accounts – in particular those of von Fintel and Gillies (2007) and Dietz (2008) where the
prejacent plays a significant role in explaining conversations.
40
of the prejacent.
The phenomenon of a hearer responding to the content of a that-clause is rather
robust; it does not just occur in the domain of ‘might’-sentences. Consider again our example
from Chapter One:
(Philosopher) Betty: Phil thinks that Angelica is a philosopher.
Lil: No. She’s not.
Lil responds as she does because she and Betty are concerned with whether Angelica is a
philosopher, not with what Phil thinks. In such a case, Betty might raise the complement
clause of the thought report. So, responding to the raising of the prejacent of ‘might’ may just
be an instance of a more general phenomenon.
It’s important to note that The Prejacent Approach does not say that other factors play
no role in determining which of the two at-issue speech acts is more appropriate to respond to
on a given occasion. It may be that general principles of cooperation or the goal of the
conversation make it preferable to respond to one of the at-issue speech acts and not the
other. Indeed, I think that this is often the case. So, a conversation where it is degraded to
respond to one or the other of the at-issue speech acts doesn’t doom The Prejacent Approach.
What would be a strong objection to The Prejacent Approach is a case where neither of the
at-issue speech acts present in a conversation explains why a response is appropriate. After
displaying the relative virtues of The Prejacent Approach, as compared with The Direct
Approach, we will see cases just like this.
46
So, let’s see how The Prejacent Approach is supposed to help with our problem cases.
Here is how it appears to explain (Cookies). In line with PREJACENT RAISING, in
making his utterance Phil raises the proposition that Angelica is the thief. According to The
Prejacent Approach, there are two at-issue speech acts – Phil’s assertion and his raising of this
proposition. We saw that we can’t make sense of Lil’s response as a response to his assertion.
But we can make sense of her response as a response to his raising this proposition. She
denies that Angelica is the thief. This is appropriate because The Prejacent Approach says
that raising the prejacent is an at-issue speech act – and because Lil knows Angelica is not the
thief. In general, problem cases like (Cookies) are explained by saying that the hearer
responds to the speaker’s raising of the prejacent.
The Prejacent Approach also gives us the resources to make sense of conversations
46
Why couldn’t we similarly hold that there is some other explanation for the infelicity in the examples I used
to argue against The Implicature Approach and The Hearer’s Context Approach, (Petrol') and (Monsters)
respectively? We couldn’t because there are no countervailing features in these cases. For instance, in both
cases the responses that the hearers offer should, if the theories were right, be relevant and serve to promote
the goal of the conversation. However, these responses are still inappropriate.
41
where multiple responses are appropriate, like (Battleship'):
(Battleship') Phil: There might be a battleship on C4.
Lil: Yeah. There might be. / No. There can’t be.
Phil performs multiple speech acts in making his utterance. He asserts that it is consistent
with his information that there is a battleship on C4 (by Contextualism) and raises that there is
a battleship on C4 (by PREJACENT RAISING). And The Prejacent Approach says that
both of these are at-issue speech acts. So, it is well-positioned to explain the two sorts of
appropriate responses. Lil’s first available response is to Phil’s assertion. It is consistent with
his information that there is a battleship on C4. Lil’s second available response is to his
raising the prejacent. It isn’t the case that there is a battleship on C4. Because The Prejacent
Approach allows that there are two at-issue speech acts, it is able explain these cases where
multiple responses are appropriate.
Unfortunately, The Prejacent Approach doesn’t explain all conversations in which
‘might’-sentences are used and these uses are responded to. In particular, it doesn’t explain
cases where a hearer responds positively to an utterance of a ‘might’-sentence, while having
different information from the speaker but not knowing whether the prejacent is true. I will
also argue that there is pressure on The Prejacent Approach because of its inability to explain
conversations involving uses of and responses to uses of terms very much like ‘might’. My
conclusion will not be that raising the prejacent is not an at-issue speech act. Perhaps it is
part of an explanation of all of the problematic conversations, but The Prejacent Approach
can’t be the whole story. Finally, I will show why the motivation I gave above for the
plausibility of The Prejacent Approach does not support it as strongly as advertised.
We want the account of why we can respond positively to uses of ‘might’-sentences to
be structurally the same as the account of why we can respond negatively to them. But The
Prejacent Approach does not adequately explain positive responses.
47
Suppose again that
Susie and Lil are investigating the cookie theft. But this time, suppose that Phil knows
nothing about who stole the cookies. Lil, though she does not know who stole them, knows
that they were stolen when only Angelica, Tommy, and Chuckie were in the house:
(Agree Cookies) Susie and Lil: Phil, we’re interviewing people to get information
about who stole the cookies. Do you know who stole them?
Phil: No, but Angelica might be the thief.
Lil: Yeah. She might be.
Again, we can say, from Contextualism and PREJACENT RAISING, that Phil asserts an MP
47
This point is originally due to von Fintel and Gillies (2011).
42
and raises the prejacent. Lil’s response is appropriate. According to The Prejacent Approach,
she must either be responding to Phil’s assertion or to his raising of the prejacent. But she is
not doing either of these. She is not agreeing with what Phil asserted – the proposition that it
is consistent with Phil’s information that Angelica is the thief. Her information is in play as
well; if she had known that Angelica wasn’t the thief, she would have responded negatively
(even if she knew that it was consistent with Phil’s information that Angelica was the thief).
And she is not agreeing with the prejacent – the proposition that Angelica is the thief. Lil
doesn’t know (and doesn’t take herself to know) whether Angelica is the thief or not. So, in
cases of agreement where the participants have different information, The Prejacent Approach
fails.
Other terms that signify epistemic uncertainty, like ‘likely’, should get treatments
similar to the one given for ‘might’ in Contextualism:
Likely Contextualism: The direct speech act performed by a speaker who
assertively utters
┌
It is likely that S
┐
, where p is the content of S in the context, is an
assertion of the proposition that the value that the credence function of the context
assigns to p conditional on the information of the context is greater than the value it
assigns to not-p conditional on the information of the context.
I’ll leave an explanation of the details of Likely Contextualism to the side for now. What is
important is that this makes ‘likely’ context-sensitive just like ‘might’ – both are sensitive to
information. And we encounter the same sort of puzzle in the case of ‘likely’ as we did in the
case of ‘might’. Suppose that, given Phil’s information, there is a high credence that Angelica
is the thief. But, given Lil’s richer state of information, there is a low – but nonzero –
credence that Angelica is the thief. Consider:
(Likely Cookies) Susie and Lil: Phil, we’re interviewing people to get information
about who stole the cookies. Do you know who stole them?
Phil: No, but it’s likely that Angelica is the thief.
Lil: No. It’s not likely that she is. I was with her almost all
afternoon.
In this case Lil responds appropriately. But she is not responding either to Phil’s assertion or
to his raising of the prejacent (assuming that there is some analogue of PREJACENT
RAISING for uses of ‘likely’-sentences). She is not denying what Phil asserted – that the
credence that Angelica is the thief given his information is higher than the credence that she is
not the thief given his information. Lil may know that, given Phil’s information, this relation
holds. But, Lil is also not evaluating the prejacent. Lil does not know (and does not take
43
herself to know) whether Angelica is the thief. So we have a similar result to what we found in
the positive response case. None of the at-issue speech acts explain Lil’s response.
48
The
pressure in this case to abandon The Prejacent Approach, which, as formulated, applies only
to ‘might’, is not direct. It relies on an extremely reasonable view – that other terms signifying
epistemic uncertainty should receive a semantic and pragmatic treatment similar to that given
for ‘might’ – that is, we should adopt Likely Contextualism along with a suitably generalized
theory of at-issue speech acts.
49
Above I noted that it is natural for hearers to respond to the speaker’s raising of the
prejacent. This is because speakers and hearers are usually concerned with the truth of the
prejacent and not with what is consistent with the speaker’s information. While this is right, it
is worth noting that there is a disanalogy between cases like (Cookies) and those that were
used to support The Prejacent Approach. Note that “You’re wrong” is an appropriate
response to utterances of ‘might’-sentences. Consider a slight variation on (Cookies):
(Wrong Cookies) Susie and Lil: Phil, we’re interviewing people to get information
about who stole the cookies. Do you know who stole them?
Phil: No, but Angelica might be the thief.
Lil: You’re wrong. She can’t be. I was with her all afternoon.
50
But Phil does not commit himself to the proposition that Angelica is the thief. So why could
Lil say that he is wrong about this? Notice the disanalogy with an amended version of the
case considered above:
(Wrong Philosopher) Betty: Phil thinks Angelica is a philosopher.
Lil: # You’re wrong. She’s not.
What Lil says is inappropriate. Betty is not committed to the proposition that Angelica is a
philosopher. So it is infelicitous to say that she is wrong about it.
Thus, it seems that something more is going on in (Wrong Cookies) than is going on
in (Wrong Philosopher). The obvious thing to say is that in (Cookies) and (Wrong Cookies)
there is some kind of disagreement between Phil and Lil. On the other hand, in
(Philosopher) and (Wrong Philosopher), there is no disagreement – at least not between Betty
and Lil. In Chapters Three and Four, I’ll present a view about the source of the disagreement
in (Cookies) and (Wrong Cookies). But, for now, we should note that this disagreement
48
A related, but distinct, argument is given by Dowell (2011).
49
For more extensive remarks on the desired parallel, see Swanson 2011. Epistemic terms like ‘likely’, are the
topic of Chapter Six. One could also doubt The Prejacent Approach using an analogy between ‘might’ and
predicates of personal taste. See Stephenson 2007, Egan 2010, and Shaffer 2011 for more.
50
Saying that Phil is wrong does not amount to saying that he is blameworthy. In (Wrong Cookies) Phil is
wrong, but he is not blameworthy. Through no fault of his, he lacked the information that Angelica was not
the thief. Thanks to Kai von Fintel and Anthony Gillies for discussion. See also von Fintel and Gillies MS.
44
between conversational participants suggests that, often, what is being responded to is not
simply the prejacent.
51
5. The Multiple MP Approach
In this section, I entertain and ultimately reject a more expansive theory about which
speech acts can, given an utterance of a ‘might’-sentence, be at-issue. This theory is based on
some suggestions by Kai von Fintel and Anthony Gillies (2008; 2011), though it is not precisely
their view.
52
We can start by asking what further sorts of at-issue speech acts will help us in the case
of (Agree Cookies). One way to see (Agree Cookies) is as follows: Phil utters a ‘might’-
sentence based on his information and Lil responds positively based on the information of the
group. This looks very much like the diagnosis of the case that a Relativist would give. But,
given the right theory of at-issue speech acts, we don’t need to be Relativists to explain this.
All we need is that Phil’s utterance gives rise to an at-issue speech act whose content is the
proposition that it is consistent with the group’s information that Angelica is the thief. We
need not say that Phil asserts this proposition; we only need to claim that it is the content of
some at-issue speech act. In general, we need speech acts whose contents are MPs which are
relative to at least the speaker’s information and the group’s information.
It seems unmotivated to simply say that any MP about the consistency of the prejacent
with some body of information or other is the object of an at-issue speech act. Von Fintel and
Gillies (2011) attempt to give an account of which MPs are available for response. Though
they don’t subscribe to the theory of direct speech acts, Contextualism, they do subscribe to a
contextualist semantic theory and their view has much in common with Contextualism. So, I
will address their view here. According to their contextualist semantics, the semantic content
of a ‘might’-sentence is an MP – the proposition that the prejacent is consistent with the
information of the context. But they claim that it is often not determinate in a conversation
what the relevant information is. Because of this, there is no single context of utterance, but
rather a cloud of admissible contexts – possible contexts which are candidates for being actual,
51
MacFarlane (2011a) suggests two further ways of seeing that the prejacent isn’t what the hearer responds to.
One involves the hearer responding with “You spoke falsely”. This is supposed to target what the speaker
asserted. As Dowell (2011) points out, “You spoke falsely” is a strange thing to say, and judgments about its
felicity aren’t particularly consistent. MacFarlane also suggests just asking the hearer what she disagrees with,
the proposition asserted or the prejacent. I’m skeptical about this approach for two reasons. First, people’s
communicative competence may not always be something they can properly reflect on. Second, asking the
hearer whether she is responding to the proposition asserted or the prejacent assumes that those are the only
things available to be responded to. In what follows, I’ll entertain a view which rejects this assumption.
52
Their view is more flexible than the one to be discussed here – combining the to-be-discussed Multiple MP
Approach with some of the insights of The Prejacent Approach. But this combination is of no help in
explaining positive response versions of the sort of conversation I consider here.
45
though none is.
53
In such a situation, when a speaker utters a ‘might’-sentence, he does not
make an assertion, as Contextualism would suggest (because there is no context to determine
which proposition he asserts). Rather he puts in play a proposition relative to the information
of each admissible context. That is, they suppose that with a single utterance of a ‘might’-
sentence a speaker can perform multiple speech acts. They posit a speech act called putting in
play. Putting a proposition in play involves proffering it for evaluation – acts of putting in
play are at-issue speech acts. However, a speaker can put a proposition in play without
committing to its truth.
54
For an utterance of a ‘might’-sentence to be appropriate, a speaker
only needs to be in a position to assert one of the propositions he puts in play. Here, then is
their account of the propositions put in play by an utterance of a ‘might’-sentence:
PUTTING IN PLAY: A speaker who utters a sentence of the form
┌
It might be that
S
┐
, where p is the content of S in the context, typically, for the body of information of
each admissible context, x, puts in play a proposition of the form p is consistent with x.
If PUTTING IN PLAY is on the right track, then we can see why there may be multiple MPs
that can serve as at-issue speech acts.
We are in a position to state the theory of at-issue speech acts that corresponds to the
view just outlined:
The Multiple MP Approach: An utterance of a ‘might’-sentence may give rise to many
at-issue speech acts – those of putting an MP in play.
55
In this section I will outline how PUTTING IN PLAY and The Multiple MP Approach
seem to explain conversations like (Agree Cookies). But I will argue that a natural extension
of this combination of views cannot explain conversations involving some embedded uses of
‘might’.
I will show how The Multiple MP Approach appears to give an explanation of (Agree
Cookies). Given this, it is straightforward to see how it applies to the other problem cases for
The Direct Approach – (Cookies) and (Battleship') – and I leave these details to the reader. In
(Agree Cookies), Phil puts in play the proposition that it is consistent with the group’s
53
We might make sense of their view of contexts in the following way: A context is an ordered n-tuple of
values needed to resolve possible context-sensitivity. It is determined by the physical speech situation. A
determinate context is one that has a single object in each position of the n-tuple. A context is not
determinate if there are parameters which don’t have a single value – if there are a range of admissible values.
In the case at hand, the information parameter is not determinate. So, an expression relying on that
parameter will not have a determinate content relative to that indeterminate context.
54
Putting in play is like the speech act raising from above. But putting a proposition in play happens only
when there is some contextual parameter that could be indeterminate. Raising can be done even if there is
no contextual parameter – like in the prejacent cases from above. I’m not claiming any deep distinction
between the two. But it is convenient to be able to separate putting in play from raising in the discussion.
55
Since assertion is a species of putting in play, The Multiple MP Approach rightly allows hearers to respond
to any MP that is asserted (according to von Fintel and Gillies, this will happen in the cases where there is no
indeterminacy about what information is relevant in the context).
46
information that Angelica is the thief (in addition to putting in play the proposition that it is
consistent with his information that Angelica is the thief).
56
Given The Multiple MP
Approach, his putting this proposition in play is an at-issue speech act. So Lil can respond to
it. And the proposition put in play is true; so Lil can respond positively. Furthermore, since
she is responding based on this proposition, there is vindication that she is responding based
on her greater state of information – not agreeing with Phil’s assessment of his own
information.
57
Despite its advantages, The Multiple MP Approach cannot explain all of the
conversations we want to explain. In particular, it does not adequately explain some of the
features of embedded ‘might’-sentences. Though von Fintel and Gillies (2011) are agnostic
about how their account extends to uses of embedded ‘might’-sentences, the natural way to
extend the account does not explain all of the relevant conversations.
Here is the case I wish to focus on. Suppose Lil, Susie, and Betty all know, as
background, that Betty’s friend Phil has never heard of Lil. Lil and Susie are searching for the
cookie thief. They call in Betty to interview her:
(Belief Cookies) Susie and Lil: Betty, we’re interviewing people to get information
about who stole the cookies. Do you know who stole them?
Betty: No, but Phil believes Angelica might be the thief.
Lil: He’s wrong in believing Angelica might be the thief. She can’t
be. I was with her all afternoon.
The exchange in (Belief Cookies) is appropriate.
It is important to note that, given the theories we are considering, PUTTING IN
PLAY and The Multiple MP Approach, we aren’t going to get any predictions at all about
(Belief Cookies). Throughout this dissertation I will refer to sentences where ‘might’ takes
widest logical scope as simple ‘might’-sentences and sentences where ‘might’ is embedded
under an operator complex ‘might’-sentences. We don’t yet get any predictions about (Belief
Cookies) because PUTTING IN PLAY only tells us only about the speech acts performed by
speakers who utter simple ‘might’-sentences. Von Fintel and Gillies recognize this, but the
problem is not simply that their theory is incomplete. It is that the natural way of completing
it fails to explain (Belief Cookies) in much the way that combination of Contextualism and
The Direct Approach failed to explain (Cookies).
Let’s investigate the natural way of extending PUTTING IN PLAY. We can start by
56
There may be more propositions put in play. This is how von Fintel and Gillies (2011) propose to make
sense of the eavesdropper cases of Egan, Hawthorne, and Weatherson (2005) and Egan (2007).
57
An of extension of The Multiple MP Approach to the case of epistemic terms like ‘likely’ would dissolve that
problem for the analogue of The Prejacent Approach. I’ll leave it to the reader to fill in the details.
47
thinking about what the contextualist account says about the semantics of ‘might’-sentences
when embedded under ‘believes’. The obvious answer is that the content of such a sentence is
the proposition that the subject of the belief report believes that it is consistent with the
information of the context that the prejacent is true. Given that, on von Fintel and Gillies
account, it can be indeterminate what the information of the context is, we are led to the
following extension of PUTTING IN PLAY:
PUTTING IN PLAYBelief: A speaker who utters sentence of the form
┌
A believes that
it might be that S
┐
, where p is the content of S in the context and o is the person
denoted by A, typically, for the body of information of each admissible context, x, puts
in play a proposition of the form o believes that p is consistent with x.
Putting together PUTTING IN PLAYBelief and The Multiple MP Approach will give us
predictions about conversations like (Belief Cookies), but, unfortunately the predictions are
incorrect.
In (Belief Cookies) Lil appropriately says that there is some proposition that Phil
believes and is wrong about. Given von Fintel and Gillies’s way of explaining warrant, a
number of acts of putting propositions in play are appropriate to perform only if the speaker
is warranted in asserting at least one of those propositions. The problem in (Belief Cookies),
however, is that Lil is not warranted in asserting any of the propositions that we might think
she puts in play. She is not warranted in asserting that Phil is mistaken in believing that it is
consistent with his information that Angelica is the thief, since she knows that such a
proposition is true (so, he is not mistaken).
58
And Lil is not warranted in asserting that Phil is
mistaken in believing that it is consistent with the information of some group that includes her
that Angelica is the thief, since she knows that Phil does not believe this proposition. Lil
knows that Phil does not have beliefs about a group that includes her since she knows that
Phil has never even heard of her. So, there is no proposition that Lil could put in play that
she is warranted in asserting. But The Multiple MP Approach requires there to be such a
proposition for Lil’s utterance to be appropriate. Thus, given The Multiple MP Approach,
her utterance should not be appropriate. But it is; so, The Multiple MP Approach must be
incorrect.
Here is one way to look at this problem. The reason that The Multiple MP Approach
helped in cases like (Cookies) and (Agree Cookies) was that it allowed us to explain the
58
We could extend this point for other suggested bodies of information by supposing that Lil knows that she is
the only one who is aware of Angelica’s innocence. In such a scenario, Lil knows, for any group that she is
not a part of, that it is consistent with that group’s information that Angelica is the thief. So, she would not
be warranted in saying that Phil is mistaken in having a belief about it being consistent with the information
of any such group that Angelica is the thief.
48
production of an utterance of a ‘might’-sentence in terms of one proposition put in play (the
one relative to the speaker’s information) and the evaluation of that utterance in terms of
another proposition put in play (the one relative to the group’s information). But in the
embedded case, we have both the production and the evaluation in a single utterance. So,
there needs to be a single proposition put in play that explains both. However, this
proposition is exactly what we could not find at the outset of our search; it is precisely what
led us to abandon The Direct Approach in the first place.
One could take this result to show that the combination of PUTTING IN PLAYBelief,
and The Multiple MP Approach is false, while holding that this does not impugn PUTTING
IN PLAY and The Multiple MP Approach. But the former combination is the natural way to
extend the latter combination to the belief case. They form part of a unified theory for
explaining conversations involving both simple and complex ‘might’-sentences. Chapter Five
is concerned with uses of and responses to uses of complex ‘might’-sentences quite generally.
There I will introduce a number of theories about the speech acts performed by assertive uses
of ‘might’-sentences embedded in various constructions (under connectives, attitude operators,
quantifiers, etc.), as part of a unified picture of communication.
6. Lessons Learned
In this chapter, I’ve combined Contextualism with a number of theories at-issue speech
acts. In each case, I found the combination to be unsatisfactory. There are a number of
lessons that we’ve learned along the way. First, a theory of at-issue speech acts should be
plausible in light of a general account of communication (The Implicature Approach and The
Hearer’s Context Approach ran afoul of this desideratum). Second, we need an account of
‘might’-communication that allows that there are sometimes multiple at-issue speech acts
performed by a speaker who utters a ‘might’-sentence. This is to account for cases – like
(Battleship') – where multiple responses are appropriate. Finally, we need an account of
‘might’-communication that allows there to be one thing that is both committed to and
negatively evaluated in cases like (Belief Cookies). Even the very liberal Multiple MP
Approach stumbled on this count.
In the rest of this dissertation, I will present a view which heeds these lessons. It heeds
the first by following the model of standardized indirect speech acts – like the request
performed by a speaker who utters, “Can you close the door?”. It heeds the second by pairing
Contextualism with an account of a further speech act that is typically performed – and is
typically at-issue – by a speaker who utters a ‘might’-sentence. And it heeds the third by
49
realizing that this speech act is of a different sort than the ones that we’ve seen so far. Instead
of thinking of a speech act with some propositional content, I will proceed by focusing on a
speech act of expressing an attitude of partial commitment A speaker like Lil in (Belief
Cookies) will be able to appropriately ascribe this attitude and negatively evaluate it.
Chapter Four will be devoted to motivating and making sense of such a view of simple
‘might’-communication. In Chapter Five, I will extend the view to uses of embedded ‘might’-
sentences, allowing for an explanation of cases like (Belief Cookies). Before getting to this
account of communication, we need to discuss the sorts of attitudes that are the objects of
these speech acts of expression. In the next chapter I will motivate their existence, discuss
their rational role in inquiry, and show how they feature in disagreement between agents.
50
Chapter Three: Reasoning with Uncertainty: Floats and Sinks
In Chapter One, I presented the challenge of explaining ‘might’-communication,
springing from the different ways ‘might’-sentences are used and responded to. In Chapter
Two, I showed that this challenge is quite persistent – it resisted explanatory attempts by some
more liberal theories of at-issue speech acts. In Chapters Four and Five I am going to give a
view of ‘might’-communication which meets this challenge, and in Chapter Six I will extend
the view to deal with other epistemic terms, like ‘must’ and ‘likely’. In order to prepare for
this, we need to go beyond issues in language and communication to think seriously about
why we use epistemic language.
Gunnar Björnsson and Alexander Almér (2010, 30) say, “the central conversational
purpose of claims to [the effect that some proposition, P, might be the case], is to guide
strategies of uncertainty relative to P: to direct the way we search for more evidence for or
against P, hedge our bets and direct our contingency planning.” More generally, epistemic
language is used to coordinate reasoning in situations of uncertainty. The view of
communication that I will present in the next chapter says that speakers typically use
epistemic sentences to express the attitudes that they use in reasoning with uncertainty. So,
we need to understand this reasoning and the attitudes agents use in it before we can give a
satisfactory theory of communication involving epistemic language. That is the task of this
chapter.
When I talk of reasoning with uncertainty I am talking about explicit reasoning with
attitudes that are, in some sense, accessible. This may only be a part of the total reasoning
that goes on in agents’ minds. I am not, for instance, concerned solely with attitudes like
those used in the decision theoretic conception of credences, where they are theoretical posits
to explain behavior. Rather, I am interested in psychologically real attitudes that agents can, if
reflecting properly, consult in their process of thinking about what is the case.
In §1 I present a picture of explicit reasoning with uncertainty according to which
agents take some subset of the total possibilities as serious options in reasoning. Agents have
an attitude toward these privileged possibilities – they take them as options in reasoning. In
filling out this picture of reasoning with uncertainty, I argue, in §2, that agents also have a
complementary attitude – one of ruling out a proposition as an option in reasoning. In §3, I
discuss the norms of rationality that apply to these attitudes. Next, I extend the account to
51
deal with conditional and quantificational versions of these attitudes – in §4 and §5. In §6, I
complete the account by accommodating the fact that agents can take propositions as options
in reasoning to different degrees. Finally, in §7, I discuss a hypothesis about how agents
evaluate each other’s float- and sink-like attitudes, and in §8, I show how this conception of
evaluation underpins how agents with these sorts of attitudes can disagree with each other.
1. Reasoning with Uncertainty and Floats
In this section I argue for a view of reasoning with uncertainty which says that agents
focus their reasoning on some privileged set of propositions. Agents have a special attitude
toward the propositions in this set. They take them as options in reasoning.
I’m going to investigate a particular example of reasoning. But the results apply
generally for reasoning agents with ordinary human abilities. Consider Phil. He wants to
figure out who stole the cookies. He has evidence about the crime, but it doesn’t show who
the thief is. Using some of the evidence, he decisively rules out that Lil is the thief and he
decisively rules out that Susie is the thief. He last checked the cookie jar eight hours ago. It
was then full of cookies, so he knows that the cookies must have been stolen in the last eight
hours. He rules out as the thief anyone who couldn’t get to the cookies in Los Angeles in
eight hours. Anyone else, other than Lil and Susie, could, in principle, be the thief. Though
many of these people are very unlikely to be the thief, it is consistent with his information that
one of them is.
It is natural to think that Phil proceeds in reasoning as follows: He focuses on a
privileged set of propositions – one for each suspect. These propositions are salient to Phil,
perhaps based on evidence he has, testimony from others, or habit. He treats these
propositions in a special way in reasoning. They are the ones he aims to gather evidence
about and the ones he spends time reflecting on.
59
They are the ones he takes as options in
reasoning about the question of who is the thief.
60
Let’s suppose that in our example case,
Phil takes it as an option that Tommy is the thief, takes it as an option that Chuckie is the
thief, and takes it as an option that Angelica is the thief. Usually, he will discover that one of
these propositions is true and which one that is. Sometimes, however, he will discover that
59
Saying that Phil focuses on these propositions in explicit reasoning is compatible with thinking that other
propositions play some role in reasoning, in an implicit way.
60
Compare with Toulmin 1958 (18): “The first stage after the stating of the problem will be concerned,
therefore, with setting out the possible solutions, the suggestions demanding our attention, or at any rate the
serious possibilities, which demand our attention most urgently.”
52
none of them is true. In this case, he will restructure his reasoning, taking some different
propositions as options.
We’ve said that agents focus their attention in reasoning on some privileged
propositions, which they take as options in reasoning. To take a proposition as an option in
reasoning is to have an attitude toward that proposition. We can call an attitude of taking a
proposition as an option in reasoning a float. And we can say one floats a proposition when
one takes it as an option in reasoning. In the example at hand, Phil floats that Tommy is the
thief, he floats that Chuckie is the thief, and he floats that Angelica is the thief.
It is important to note that floating a proposition involves more than simply having
that proposition be consistent with one’s information. For example, it is consistent with Phil’s
information that Brad Pitt is the thief. But Phil does not float that Pitt is the thief. If he
considers this proposition at all, he judges it too unlikely to be important. Nonetheless, this
doesn’t, by itself, show that we must posit a new, basic attitude to account for reasoning with
uncertainty. Floats may be reducible to some other sort of attitudes. For instance, an initially
plausible hypothesis is that floats can be reduced to credences of at least moderate strength. I
will explore the possibility of this sort of reduction in Chapter Seven. For now, I just want to
recognize that floats play a role in agents’ reasoning – whether or not they are, in the end,
reducible to a more basic sort of attitude.
2. Sinks
In this section, I will argue that agents distinguish the propositions that they rule out in
reasoning from those that they ignore. So, agents must have a complementary sort of attitude
to a float – a state of ruling out a proposition in reasoning. Together these attitudes allow an
agent to structure propositions into those she takes as options, those she rules out as options,
and those that are neither taken nor ruled out as options.
Let’s reflect again on the scenario from above, where Phil is reasoning about who stole
the cookies. Phil floats some propositions about who is the thief. So, there is a distinction
between the propositions that Phil floats and the ones that he doesn’t. But this distinction
doesn’t fully describe the situation that Phil is in when he reasons. It’s not the case that he
floats some propositions and all other ones he treats as equally less relevant. Rather, there are
some propositions that he rules out as options in reasoning. Above we saw that Phil has
evidence that allows him to prove that Lil is not the thief and that Susie is not the thief. So, in
53
our scenario, Phil rules it out as an option that Lil is the thief, and he rules it out as an option
that Susie is the thief. Since he has evidence that shows that each of Lil and Susie is not the
thief, Phil deliberately structures his reasoning so as not to take either of them as an option for
being the thief. He takes it for granted that Lil isn’t the thief and that Susie isn’t the thief.
This is different than, say, the proposition that Brad Pitt is the thief, which Phil may not even
consider or which, even if he does entertain it, is not considered likely enough to take as an
option.
Thus, agents have, in addition to floats, another sort of attitude that they use in
reasoning with uncertainty. This is the attitude an agent has toward a proposition that he
rules out in reasoning. I call such an attitude a sink.
61
Just as in the case of floats, I don’t
want to commit to sinks being a basic sort of attitude. They may be reducible to some other
attitude, as I will discuss in Chapter Seven.
So, in a scenario where an agent is reasoning under uncertainty, the space of
propositions can be divided into three groups: the propositions that the agent floats, the
propositions that she sinks, and the propositions that she neither floats nor sinks. We can
describe where an agent is in the process of deliberation about a question by describing which
propositions she floats and sinks. Let’s call the state an agent is in when deliberating about a
question her deliberative state with regard to that question.
62
There is one more important feature of floats and sinks that we must note before
moving on. Agents take part in different inquiries, and they may take different propositions as
options in different inquiries. Thus, they may float a proposition in one inquiry and not float
it in another inquiry. For instance, when trying to figure out who stole the cookies, Phil may
not float that Brad Pitt is the thief. But he may float that proposition in the following
scenario: a bookie offers Phil a dollar bet on whether Brad Pitt is the thief, offering to pay out
a billion dollars if Pitt is the thief. There is no inconsistency in Phil not floating this
proposition in the former scenario and floating this proposition in the latter one. He simply
floats that Brad Pitt is the thief when reasoning about the bet but not when doing his detective
61
This is similar to Huemer’s (2007) attitude, dismissal. He says, “One who dismisses a proposition is
disposed to refuse to take it seriously. Consequently, he does not take it as relevant in practical deliberation
nor does he bother to collect evidence about whether it is the case.” The reader can also compare a sink also
to what Toulmin (1958) calls ruling-out.
62
The initial structuring of an agent’s deliberative state need not be determined as the result of some previous
reasoning process. Otherwise we would face a sort of regress. Rather, an agent may float or sink
propositions as a result of the weight of evidence she takes to be in favor of or against them, their salience,
the habit of considering them in similar situations, and, perhaps, other factors. Thanks to Cecilia Stepp for
discussion.
54
work. That is, he has the attitude when participating in one inquiry but not when
participating in the other. It is important to keep this issue straight when investigating the
norms governing floats and sinks.
3. Norms on Reasoning
Now that we have accepted floats and sinks, we can ask the normative question of how
agents should reason with them. Which ways of reasoning are rational and which are not?
First, I’ll lay down a general norm on deliberative states. After that, I will discuss a norm that
holds only in particular cases. It is important that the characterization of how reasoning
should progress in one case is fundamentally diachronic – that is, what mental state an agent
should rationally be in depends not just on the properties of that state, but on its relation to
previous state the agent was in.
Let’s start with a general and highly intuitive norm on deliberative states:
Singlemindedness: For any time and inquiry, an agent should not both float and sink a
proposition, at that time, in that inquiry.
63
In a single inquiry, an agent should not both take a proposition as an option in reasoning and
rule out that proposition as an option in reasoning.
64
I take this norm to be very plausible on
its face, and I won’t argue for it further.
Now, I will build toward a norm that applies in almost all scenarios of reasoning with
floats and sinks. The exceptions are scenarios where an agent loses some of what she
previously took to be her evidence. In such cases, she may have to revise her state in a way
that would be irrational in most other cases. Let’s call these retraction scenarios.
65
I’m going
to put them off to the side for now.
The paradigmatic move in reasoning with floats and sinks is moving from floating to
63
So that we can see the rational relations between floats and sinks more clearly, I’m going to make a
simplifying assumption. On some views, propositions are coarse-grained to the degree that an agent can
entertain the same proposition in different ways. Such an agent may not realize that he is entertaining the
same proposition rather than two different ones. This might affect whether or not it is rational to take
various attitudes toward a proposition – for instance believing it under one mode of presentation and
disbelieving it under another. Whether this is rational is not central to my investigation. So, I will put aside
this issue in what follows. When I talk about having different attitudes toward the same proposition, I am
assuming the agent thinks of the proposition under the same mode of presentation.
64
It might seem not just irrational, but impossible for an agent to both float and sink a proposition. I want to
stay neutral about this. Even if it is impossible, it seems that this is because an agent like us couldn’t have
both attitudes simultaneously, since doing so is obviously irrational. Thanks to Mark Schroeder for
discussion.
65
These scenarios are closely related to difficult issue of characterizing the proper method for retraction in
formal systems of belief revision.
55
sinking a proposition. Suppose an agent floats a proposition, p, and later acquires evidence
that p is false. On acquiring such evidence, she will typically progress from floating to sinking
p – she will go from taking p as an option in her reasoning to ruling it out as an option. Such
a move is permissible. An agent may gain information that shows that a proposition she
previously floated is false. She can then permissibly come to sink that proposition.
This is not the only rational change an agent can make. An agent may gain evidence
that shows that some proposition she floats is unlikely – without showing that that proposition
is false. In this case the agent can rationally stop floating the proposition without going on to
sink it – she can go from taking a proposition as an option in reasoning to not taking it as an
option (without thereby ruling it out as an option).
Sometimes an agent will start to take seriously in reasoning a proposition that she
neither took nor ruled out as an option in reasoning. That is, sometimes an agent will begin
to float a proposition she previously had no opinion about. This may happen because the
agent happened upon some evidence about the proposition or because the proposition
became particularly salient (or, perhaps, for some other reason). This is particularly likely to
happen if an agent finds out that none of the propositions she previously floated are true.
Another permissible move in reasoning is similar. An agent might happen upon some
evidence that shows that a proposition that she neither floats nor sinks is false. She can then
rationally come to sink that proposition.
We’ve just seen a number of permissible changes in an agent’s deliberative state. Not
all changes in a deliberative state are permissible (in non-retraction scenarios). In particular,
an agent should not give up a sink in a proposition. That is, she should continue to sink – or
rule out as an option in reasoning – each proposition she previously sank. Here is an
argument for this conclusion. If she thinks that her evidence proves that a proposition is false,
she should not, unless she takes herself to have lost evidence or been wrong about what her
evidence supports, begin to take it as an option in reasoning. Doing so, without changing her
mind about some of her evidence, is problematic. For instance if Stu sinks that Angelica is the
thief and learns something new – say, that Tommy likes chocolate chip cookies – then Stu
should continue to sink that Angelica is the thief. He should not come to float that Angelica
is the thief on learning this new information. Of course if he learns that the information that
he previously took to rule out that Angelica is the thief is false or that it doesn’t really rule out
Angelica’s being the thief, then he can permissibly come to no longer sink that she is the thief.
56
This, however, is a retraction scenario. So, I propose the following general norm on reasoning
with sinks:
Steadfastness: For any inquiry, if an agent sinks a proposition in that inquiry, she
should, except in a retraction scenario, continue to sink that proposition in that
inquiry.
It is important to notice that Steadfastness is a diachronic norm – it is a norm
governing how an agent’s deliberative state should not evolve across time. Any proposition an
agent sinks must continue, at a later time, to be sunk (except in a retraction scenario). Note,
however that the reverse is not true. It is rational for an agent to move from not sinking to
sinking a proposition. Thus there is an asymmetry between what deliberative states agents can
rationally reach in reasoning. Agents can permissibly gain but not lose sinks in non-retraction
scenarios.
Retraction scenarios call for an agent to backtrack – to retrace her epistemic steps. In
such a case, an agent can permissibly come to again take as options or neither take nor rule
out as options propositions that she already ruled out as options. Thus, Steadfastness doesn’t
issue a prohibition in retraction scenarios. If one learns, for example that the testimony of a
key witness who said Stu is not the thief is unreliable, the agent may permissibly come to float
or neither float nor sink proposition that Stu is the thief (as long as there isn’t some other
evidence that proves that Stu is not the thief). Again, however, our focus will mostly be on the
ordinary cases of reasoning, not retraction scenarios. And in such cases, Steadfastness puts a
constraint on how to proceed.
In this section we saw how agents may and may not proceed in reasoning with floats
and sinks. We laid down two norms of rationality, Singlemindedness and Steadfastness. We
also saw that Steadfastness is a fundamentally diachronic norm.
4. Conditional Floats and Sinks
We’ve seen that when agents reason in situations of uncertainty, they take and rule out
propositions as options in that reasoning. Agents also engage in contingency reasoning. They
have thoughts about what to take and rule out as options under various suppositions. In
order to explain this sort of reasoning, we need to accept that agents have attitudes concerning
whether to take a proposition as an option in reasoning, on some condition, and whether to
rule out a proposition as an option in reasoning, on some condition.
57
Suppose that Phil is still reasoning about the cookie theft. The situation is as before,
except that Phil has some further evidence that lets him know that if Stu has crumbs in his
pocket, it is somewhat likely that he is the thief. How should Phil respond? In response to
this sort of evidence, he should have an attitude that I call a conditional float. This is an
attitude that will help Phil structure his reasoning if he finds out that Stu has crumbs in his
pocket. It tells him to float that Stu is the thief if he learns that Stu has crumbs in his pocket.
So, by having this conditional float, Phil can reason effectively in a way that conforms to his
evidence.
Having a conditional float is useful for Phil in two ways. First, it streamlines Phil’s
reasoning if he comes to learn that Stu has crumbs in his pocket. He can form the conditional
float when he has time to deliberate fully about his evidence, so that he need not re-deliberate
on learning that Stu has crumbs in his pocket. Second, having a conditional float can guide
Phil in figuring out which other propositions to investigate. If he conditionally floats that Stu
is the thief given that he has crumbs in his pocket, Phil may take steps to figure out whether or
not Stu has crumbs in his pocket. Thus, conditional floats help agents decide where to
allocate resources since they allow agents to see which new facts would help the inquiry
progress and which would not.
We can understand conditional floats by analogy to conditional credences. Credences
are attitudes which encode an agent’s degree of confidence in propositions. It is commonly
assumed that agents have conditional credences as well. These are not attitudes that encode
an agent’s confidence in a proposition. Rather, they determine what credence an agent should
have in a proposition on learning some other proposition. I aim to understand conditional
floats by analogy. Conditional floats are not attitudes of taking some proposition as an option
in reasoning (like floats are). Rather, they are attitudes that determine that an agent should
take a proposition as an option in reasoning on learning some other information. I won’t aim
to give a complete account of either conditional credences or conditional floats here.
66
I will
just understand conditional floats by analogy to the commonly accepted sort of attitude –
conditional credences.
Just as it makes sense that agents have conditional floats, it makes sense that they have
what I call conditional sinks. Just as conditional floats are not floats, but determine what an
agent should float on learning some information, conditional sinks are not sinks, but they
66
For more on my view of conditional credences, see Lennertz forthcoming a.
58
determine what an agent should sink on learning some information. For example, suppose
that Phil knows that Stu can’t tell a lie. So, Phil has evidence that rules out that Stu is the
thief if he says he isn’t the thief. Then the appropriate way for Phil to respond to this
evidence is to conditionally sink that Stu is the thief, given that Stu says he isn’t the thief.
Such a conditional sink determines that Phil should sink that Stu is the thief if he learns that
Stu says he isn’t the thief.
We’re now in a position to enrich our picture of reasoning. Agents can, in addition to
floating and sinking propositions simpliciter, also conditionally float and conditionally sink
them. It is possible for an agent to conditionally float or sink propositions that she neither
floats nor sinks. For instance, Phil may conditionally float that Brad Pitt is the thief if he was
in the kitchen in the last eight hours, while neither floating nor sinking (simpliciter) that Brad
Pitt is the thief. It is also possible for an agent to conditionally float or sink propositions that
she already floats or sinks. For instance, an agent can conditionally sink p given q while
(unconditionally) floating p. Such an agent takes p as an option in reasoning but thinks that
learning q should make her rule out p as an option in reasoning. For example, Phil may float
that Tommy is the thief, but also conditionally sink that Tommy is the thief if he was not at
the scene of the crime when the cookies were stolen.
As in the unconditional case, there are norms on reasoning with conditional floats and
sinks. The first is just the conditional analogue of Singlemindedness:
Conditional Singlemindedness: For any time and inquiry, an agent should not both
conditionally float and conditionally sink a proposition on the same condition, at that
time, in that inquiry.
She should not both take a proposition as an option in reasoning on some condition and rule
out that proposition as an option in reasoning on that condition. For instance, if, in
reasoning about the thief, Phil conditionally floats that Stu is the thief, given that he says he
isn’t and conditionally sinks that Stu is the thief, given that he says he isn’t, then Phil is
irrational.
In characterizing conditional floats and sinks, I’ve claimed that they are attitudes that
determine norms about how to treat a proposition on learning some other proposition. For
example, having a conditional float that Stu is the thief if he has crumbs in his pocket makes it
the case that Phil should float that Stu is the thief if Phil accepts that he has crumbs in his
pocket. And having a conditional sink that Stu is the thief if he says he isn’t makes it the case
59
that Phil should sink that Stu is the thief if Phil accepts that he said he isn’t the thief. We can
formalize this in the following norm on conditional floats and sinks:
Conditional Detachment: For any inquiry, if an agent, in that inquiry, conditionally
floats (sinks) p if q and accepts q, then she should, in that inquiry, float (sink) p.
Conditional Detachment has the flavor of modus ponens. It says that putting together a
conditional state with acceptance of the triggering condition makes it the case that you should
have the corresponding unconditional state. We might wonder whether conditional floats and
sinks also give rise to a norm with the flavor of modus tollens. The answer is no. It is not the
case that, for any inquiry, if an agent, in that inquiry, conditionally floats (sinks) p if q and
sinks (floats) p, then she should, in that inquiry, accept q’s negation. The reasons why this
sort of principle fails for the case of conditional floats are different from the reasons why this
sort of principle fails for the case of conditional sinks.
Let’s take the case of conditional sinks first. It is perfectly rational to conditionally sink
p if q, float p, but not accept q’s negation. This just means that you take p as a serious option
and are committed to ruling out p if you learn q. But this is consistent with leaving it open
whether q. If you come to accept q, then you need to stop floating p and come to sink it. For
instance, Phil can take it as an option that Angelica is the thief and conditionally rule it out as
an option that she is the thief, if she was not at the scene of the crime, while being agnostic
about whether she was at the scene of the crime. There is no rational problem with this
combination of attitudes. Indeed it seems like an eminently reasonable state to be in.
Let’s now discuss the case of conditional floats. The issue here is different. Remember
we are trying to see whether an agent who conditionally floats p if q and sinks p should accept
the negation of q. However, it is already irrational to conditionally float p if q and sink p. If
you sink p, you have ruled it out under any condition – so you should rule it out on the
condition that q. This is similar to how, in the realm of credences, if you have a credence of 0
in p, you should, for any q, have a conditional credence of 0 in p given q. So, the suggested
norm simply doesn’t apply since there are no rational agents who meet the antecedent
conditions. Thus, when considering both conditional sinks and conditional floats, there are
no rational norms with the flavor of modus tollens.
67
Now let’s move on to a diachronic norm that holds in non-retraction scenarios. We
saw above that there is an asymmetry between how floats and sinks should feature in such
67
I said above that Conditional Detachment has the flavor of modus ponens. But, given what we’ve just seen, a
more accurate analogy is to a conditionalization norm on conditional credences.
60
reasoning – characterized by Steadfastness. There is also an asymmetry between how
conditional floats and conditional sinks should feature in reasoning. It is permissible, on
gaining strictly more information, to move from a conditional float in p if q to a conditional
sink in p if q. But the reverse is not permissible. If I already conditionally sink p if q, I take
myself to have information that shows that learning q would show that p is false. It only
makes sense to give up this conditional sink, if I come to think I was wrong in forming it – i.e.
if I come to reject some of what I previously took to be evidence. And this, as we’ve seen, is a
retraction scenario. More generally, if I take q to show that p is false, I either must continue
to conditionally sink p if q, come to sink p (if I learn q), or come to think my evidence was
wrong and backtrack. So, we get the following diachronic condition:
Conditional Steadfastness: For any inquiry, if an agent has a conditional sink in p if q
in that inquiry, she should, except in a retraction scenario, retain that conditional sink
in that inquiry or come to have a sink in p in that inquiry.
In this section we have seen that agents have conditional floats and sinks for the
purpose of engaging in contingency reasoning. I presented some rational norms that apply to
combinations of conditional and unconditional floats and sinks.
5. Quantificational Floats and Sinks
In this section I will present another sort of float-like attitude – what I’ll call a
quantificational float. Suppose Phil does not know who stole the cookies, but he knows that
there was only one thief. And he does not know who was in the kitchen in the eight hours
since he last saw the cookies, but he knows that there were at least two people who went into
the room. Furthermore, he has evidence that suggests that, for anybody who was in the
kitchen during that period, it is somewhat likely that that person is the thief. In order to
properly respond to this evidence, Phil should take it as an option, for each person who was in
the kitchen in the last eight hours, that that person is the thief. Let’s call such a state a
quantificational float, for each person who was in the kitchen in the last eight hours, that that
person is the thief.
We can ask what it is for someone to have a quantificational float. This is analogous to
a question about the conditional case – the question of what it is for someone to have a
conditional float. Above I suggested that we understand conditional floats as distinct attitudes
from ordinary floats. This was a relatively conservative move since I introduced conditional
61
floats by analogy to a more familiar conditional attitude – conditional credences.
Unfortunately, there is nothing in the orthodoxy about quantificational credences.
68

Nonetheless, I will argue in what follows that we need to accept that quantificational floats are
independent float-like states.
In order to support this claim, I will show that two natural attempts to reduce
quantificational floats to ordinary floats fail. I’m going to use the example of Phil’s particular
quantificational float for ease of discussion, but the results are general. Consider what it is for
Phil to have a quantificational float, for each person who was in the kitchen in the last eight
hours, that that person is the thief. There are two natural ways to try to reduce this
quantificational float to ordinary floats. These correspond to two possibilities for where the
quantification is in relation to the term denoting the attitude in a description of the state –
either in the scope of the term denoting the float (in the term denoting the content floated) or
outside the term denoting the float. The first hypothesized reduction says that for Phil to have
his quantificational float is for him to have a single, ordinary float in the following
quantificational proposition: that each person who was in the kitchen in the last eight hours is
the thief. According to this hypothesis, having a quantificational float amounts to having a
single, ordinary float with a quantificational content. The second hypothesized reduction says
that for Phil to have his quantificational float is for him to be such that, for each person who
was in the kitchen in the last eight hours, he has a float that that person is the thief.
According to this hypothesis, having a quantificational float amounts to having a combination
of floats about some quantity of objects that have the property at-issue. I will evaluate the two
hypothesized reductions in turn.
The first reduction said that for Phil to have his quantificational float is for him to have
an ordinary float in the proposition that each person who was in the kitchen in the last eight
hours is the thief. However, this reduction can’t be right, since Phil can have the
quantificational float without having the corresponding float in the quantificational
proposition. Indeed, this is so in the very case that I described to start this section.
Remember that Phil thinks that there is only one thief but that there were multiple people in
the kitchen in the last eight hours. But Phil does not float that each of these people is the
unique thief. It is incoherent to float that multiple people are the unique thief. Thus, Phil
doesn’t have this float, though he does have the quantificational float. So, we have a
68
See Lennertz forthcoming a for an argument that there are quantificational credences.
62
counterexample to the first suggested reduction of quantificational floats. A quantificational
float cannot just be a float in a quantificational proposition.
Let’s see if our second hypothesized reduction fares any better. It said that Phil’s
having his quantificational float is just his being such that, for each person who was in the
kitchen in the last eight hours, he floats that that person is the thief. According to it, Phil has
a combination of floats – each about some particular person being the thief. But, as we said,
Phil does not know which people were in the kitchen in the last eight hours. So, there may be
people who were in the kitchen in the last eight hours that he does not have the required floats
about. Furthermore, Phil may not be acquainted with some of these people. So, it may not
even be possible for Phil to have a float about them. Nonetheless, Phil does have the
quantificational float. So, we have a counterexample to the second suggested reduction of
quantificational floats. Having a quantificational float does not consist in having a
combination of ordinary floats.
69
We seem to be in a bind. It seems that in a description of a quantificational credence,
the quantification should either be inside of the scope of the term denoting the attitude (first
hypothesis) or outside of it (second hypothesis). But we’ve seen problems for both of these
options. However, there is another option. In the case of conditional states, we accepted an
independent sort of float-like attitude – a conditional float. Having a conditional float in p
given q is neither having a float in a conditional if q, then p, nor being such that, if q, you have
a float in p. Rather, it is an independent attitude (though one which is related to floats
through interesting normative principles like Conditional Detachment). We should accept
that quantificational floats are independent attitudes in the same way. In the rest of this
dissertation I will take for granted that there are quantificational floats that are not reducible to
ordinary floats and that they can be understood as analogous to conditional floats.
70
Agents can have quantificational sinks, as well. For instance, Phil may
69
There is a further argument against this hypothesis, on the assumption that a similar view is taken about
quantificational sinks. Assume that Phil thinks that no one who is currently out of the state was in the
kitchen in the last eight hours and, so, could be the thief. So, he sinks, for each person who is currently out
of the state, that she is the thief. It happens that, unbeknownst to Phil, Didi has traveled to Arizona but was
in the kitchen in the last eight hours. According to the suggested hypothesis, then, Phil has both a float that
Didi is the thief (since she was in the kitchen in the last eight hours) and a sink that she is the thief (since she
is currently out of the state). So he is, according to our Singlemindedness norm, irrational. But it seems
perfectly reasonable for Phil to both float, for each person who was in the kitchen in the last eight hours, that
she is the thief and sink, for each person who is currently out of the state, that she is the thief. If he has
made any mistakes, they are purely factual – about the present whereabouts of people who were in the
kitchen in the last eight hours. He is not irrational.
70
For further discussion and development of these ideas see Lennertz forthcoming a. Though the discussion
there focuses on quantificational credences, the lessons apply to quantificational floats.
63
quantificationally sink, for each person who was out of town when the cookies were stolen,
that that person is the thief. The examples of both quantificational floats and quantificational
sinks have involved only universal quantification. Later, I will discuss what I will call less-
than-universal quantificational floats and sinks. But first, I will discuss how universal
quantificational floats and sinks fit into a picture of rationality.
In order to put forward general norms, we need a way of talking more abstractly about
quantificational credences. We’ve focused on the example of Phil’s quantificational float, for
each person who was in the kitchen in the last eight hours, that that person is the thief. But
we can talk more generally about quantificational floats as follows: for a quantifier, Q, and
properties, F and G, we can speak of a quantificational float, for Q of the Gs, that it is F. In
talking about universal quantificational credences, we can focus only on universal quantifiers.
With this notation in hand, we are in a position to talk about norms.
The norms on universal quantificational floats and sinks are analogous to those on
conditional floats and sinks. Just as there was Conditional Singlemindedness, there is:
Universal Quantificational Singlemindedness: For any time and inquiry, an agent
should not, at that time, in that inquiry, have both a quantificational float, for each G,
that it is F and a quantificational sink for each G, that it is F
For example, if Phil has a quantificational float, for each person who was in the kitchen in the
last eight hours, that she is the thief and a quantificational sink, for each person who was in
the kitchen in the last eight hours, that she is the thief, then he is irrational. And just as there
was Conditional Detachment, there is:
Universal Quantificational Detachment: For any inquiry, if an agent, in that inquiry,
has a quantificational float (sink), for each G, that it is F and, for some object, o,
accepts that o is G, then she should, in that inquiry, float (sink) that o is F.
For example, suppose that Phil has a quantificational float, for each person who was in the
kitchen in the last eight hours, that she is the thief. And suppose that he thinks Reptar was in
the kitchen in the last eight hours. Then, he should float that Reptar is the thief. In this way,
Universal Quantificational Detachment makes sense given the purposes for which we have
universal quantificational floats and sinks.
There is also a quantificational version of the steadfastness norm for non-retraction
scenarios:
Universal Quantificational Steadfastness: For any inquiry, if an agent has, in that
64
inquiry, a quantificational sink, for each G, that it is F, then she should, except in a
retraction scenario, retain that quantificational sink in that inquiry or come to have, for
each object she thinks is G, a sink that it is F in that inquiry.
In other words, an agent should not release herself from the pressure of a quantificational sink
unless she loses some piece of her evidence. The reason for this norm is the same as the
reason for the simple and conditional versions of Steadfastness. Once an agent has a universal
quantificational sink, she has decided that some objects are ruled out from having the target
property. Simply getting more information shouldn’t change this. Only in cases where some
evidence is lost can an agent permissibly give up a universal quantificational sink.
In our discussion so far, we have been thinking of the quantificational aspect of
quantificational floats and sinks as universal – where we look for each object having some
property. But there are also less-than-universal quantificational floats and sinks. Suppose ten
people are in the next room and Phil has evidence that multiple of them (he doesn’t know
which ones) were in the kitchen during the time the cookies were stolen. Phil should have a
quantificational float, for multiple people in the next room, that they are the thief. The idea is
similar for other quantifiers. For instance, Phil could have a quantificational float, for a few
(most, etc.) people next door, that they are the thief.
One reason why it makes sense for an agent to have a universal quantificational float is
that such attitudes are governed by Universal Quantificational Detachment. These attitudes
tell her what to float when she learns that some object has some property. This is not the case
for less-than-universal quantificational floats and sinks. For an agent with a less-than-
universal quantificational float or sink, knowing that an object has some property does not
give her enough information to know if she should have a float or sink about that object. She
still doesn’t know which object with that property is (are) the one(s) to have a float or sink
about. For instance, suppose that Phil has a quantificational float, for multiple people in the
next room, that they are the thief. And suppose he learns that Dil is in the next room. He
need not float that Dil is the thief. This is because Dil could be one of the other people (those
that he need not have a float about). So, no norm analogous to Universal Quantificational
Detachment applies to these attitudes.
To understand the usefulness of less-than-universal quantificational floats and sinks,
we need to recognize the wider rational role that quantificational floats and sinks play. They
commit agents to having or not having particular attitudes on learning information –
65
sometimes a single piece, sometimes a large body. Unlike the universal case, having a less-
than-universal quantificational float or sink, for Q of the Gs, that it is F, commits the agent to
having floats or sinks about some objects being F only once they learn which objects are all
and only those that are G. Again, Phil isn’t required, based on his less-than-universal
quantificational float, for multiple people in the next room, that she is the thief, to have a float
that Dil is the thief upon learning that Dil is in the next room. But if Phil learns who all and
only the people in the next room are, he is committed to having floats about multiple of them
– and so is under rational pressure to do so.
71
Because of this, there are constraints on which quantificational floats and sinks can be
rationally held at the same time. Less-than-universal quantificational floats and sinks lead an
agent to regulate how her mental state coheres. Let’s return to our example where Phil has a
quantificational float, for multiple people next door, that they are the thief. This rationally
forbids him from also having a quantificational sink, for each person next door, that that
person is the thief. Satisfying the commitments that flow from these two quantificational
floats and sinks require him to be inconsistent, according to Singlemindedness.
72
In general
the following norm applies:
Commitment Satisfaction: An agent should not have combinations of quantificational
floats and sinks that engender commitments that are impossible to simultaneously
fulfill without violating Singlemindedness.
73
Universal Quantificational Singlemindedness is a sub-norm of this – one dealing only with the
universal case.
I should also note that a steadfastness principle holds for quantificational sinks of any
71
The commitment engendered by a less-than-universal quantificational float or sink is pro tanto, not all-
things-considered. Phil may not be all-things-considered irrational if he doesn’t have these multiple floats in
the scenario just mentioned – even if he knows which people are all and only those in the next room. This is
because he may have no grounds for differentiating them and, so, requiring him to have the suggested floats
or sinks would lead to violating some norm about properly apportioning his attitudes to the evidence or
treating symmetrical cases symmetrically. See Lennertz forthcoming a (§4.4) for more on this issue. Thanks
to Mark Schroeder, Shyam Nair, and Kenny Easwaran for discussion.
72
Something similar is true in a case where Phil has the original less-than-universal quantificational float and
the following combination of sinks: for each person next door (or all but one), Phil has a sink that that
person is the thief. In such a state, it is impossible to fulfill the commitment bred by his quantificational
float without violating Singlemindedness.
73
Commitment Satisfaction can be generalized in two ways. First, it can be generalized to predict that a
quantificational float should not be had along with ordinary sinks that preclude satisfying the commitment it
engenders. See note 72. Second, it can be generalized to predict certain sorts of inconsistencies between
quantificational and conditional floats and sinks. For instance, it is inconsistent to have a quantificational
float, for each person in the next room, that that person is the thief and a conditional sink that Stu is the thief
given that he is in the next room. See Lennertz forthcoming a (§4) for the details of these generalizations in
the analogous case of credences.
66
strength of quantification. The support for such a principle is the same as for all other sorts of
sinks. If an agent has decided that a quantity of objects with a certain property can’t have
some other property, then she shouldn’t change her mind about this unless she is in a
retraction scenario – where she realizes that she was wrong in what she took as evidence or
what she took her evidence to support.
This means that there are general versions of steadfastness and singlemindedness
principles that apply to less-than-universal and universal quantificational floats and sinks
(Commitment Satisfaction is the official statement of the latter). But there is no general
version of detachment. This applies only to universal quantificational floats and sinks.
In this section we’ve seen that agents have novel sorts of attitudes – quantificational
floats and sinks. We have seen that these states engender commitments to have particular
floats and sinks about some quantity of objects based on learning that these objects have some
property. They also put constraints on what sorts of float-like attitudes can rationally be held
together. Quantificational floats and sinks allow agents to effectively regulate some of their
more complex reasoning with uncertainty.
6. Floats of Different Strengths
The account of reasoning with uncertainty and the corresponding picture of the mind
that I’ve developed is quite rich. But it is lacking in one important respect. The current state
of the account implies that any proposition that is taken as an option is treated in the same
way. But, in reality, not all options are equal when reasoning with uncertainty. An agent can
take one proposition as a more likely option than another proposition, while taking both as
options.
For instance, Phil may take Angelica’s being the thief to be more likely than he takes
Tommy’s being the thief, while taking both of those propositions as options in reasoning.
This may result in Phil allocating more resources to searching for evidence about Angelica
than he allocates to searching for evidence about Tommy. We can say that he floats that
Angelica is the thief to a greater degree than he floats that Tommy is the thief. It seems that
an agent can take a proposition as somewhat likely or very likely – i.e. she can have a
somewhat likely float or a very likely float. It even seems that an agent can float propositions
to particular, real-numbered degrees. For instance, Phil may take it as 50% likely in his
reasoning that Angelica is the thief, 25% likely in his reasoning that Tommy is the thief, and
67
25% likely in his reasoning that Chuckie is the thief. In such a case, it would make sense for
Phil to send two of his detectives to investigate Angelica, and one each to investigate Tommy
and Chuckie.
74
In this case, we can say that Phil floats that Angelica is the thief to degree .5,
floats that Tommy is the thief to degree .25, and floats that Chuckie is the thief to degree .25.
75

In general, floats can come in qualitative and quantitative degrees.
This need not collapse into an account where every float an agent has is a float to some
degree. It is possible that in some situations, an agent merely floats some propositions, while
not floating them to any particular degree. I call such floats bare floats, while I call floats that
aren’t bare degreed floats. Imagine you are a detective just assigned to a case. You are shown
the three suspects. The natural thing to do is, for each suspect, float that he or she committed
the crime. But these floats need not have particular strengths.
76
There are simply three
people who are serious options for having committed the crime. As you go forward you may
come to float some propositions to a greater degree and some to a lesser degree. But it is
natural to have a bare float in each at the outset.
As I mentioned, the precision of the strengths of degreed floats can vary. Some are
qualitative – like somewhat likely floats or very likely floats – and some are quantitative – like
three-quarter likely floats or 60% likely floats. The variation in strengths of floats should not
be surprising. Sometimes we take a proposition as three-quarters likely in our reasoning. But
sometimes we aren’t this precise, so that we merely take it as very likely. The limiting case is
the bare float – where we take a proposition as an option without taking it as likely to any
particular quantitative or qualitative degree.
Just like the states that we have been discussing, degreed floats enter into rational
relations. We can generalize the Singlemindedness condition:
Generalized Singlemindedness: An agent should not both float a proposition to any
74
This is an idealized example, where I am assuming that each detective offers about the same amount of help
to finding the answer and each detective costs the same amount. I don’t take Phil’s actions here to support
that he has precisely these degrees of float. Rather, it is plausible that, if he has floats to these degrees, he will
act in approximately this way in such a scenario.
75
The idea that floats can come in degrees makes floats seem very much like degrees of confidence or
credences. We might wonder whether floats just are credences. I will return to this issue in Chapter Seven .
76
This allows us to avoid the following issue that arises for standard views of credences: How should an agent
assign initial credences to an exclusive, exhaustive set of possibilities about which she has no evidence? One
answer is to follow the principle of indifference, which says that she should distribute her credence equally
among such propositions. Others argue that any initial distribution of credences is allowed (provided it is
representable by a probability function). The above picture of floats avoids this issue altogether. One can
take a proposition as an option without having a specific degree of confidence that it is true. As one gains
more information, one can become more decided and, so, have more precise degreed floats. See White 2010
for a defense of the principle of indifference for credences.
68
degree and sink that proposition, nor should she float a proposition to two separate
degrees (at the same time).
So, if an agent has a float – bare or degreed – in a proposition and a sink in that proposition,
then that agent is irrational. Furthermore, if an agent floats a proposition to two different
degrees, she is irrational. For example, suppose that Phil somewhat likely floats that Angelica
is the thief and also very likely floats that Angelica is the thief. He takes it as a (merely)
somewhat likely option that Angelica is the thief and also takes it as a very likely option that
Angelica is the thief. This is irrational.
77
We’ve seen that agents have degreed, as well as bare, floats. Above we saw that we
needed conditional versions of bare floats. The same is true of degreed floats. We need
conditional degreed floats. The reason is the same as above. Agents engage in contingency
reasoning with uncertainty. For example, Phil may have a conditional very likely float that
Angelica is the thief, if the fingerprints on the cookie jar are hers. The same sort of motivation
applies to accepting quantificational degreed floats. And the rational relations involving
conditional and quantificational degreed floats will follow as expected from the principles
about conditional, quantificational, and degreed floats as well as Generalized
Singlemindedness.
In this section, I’ve further developed our picture of reasoning with uncertainty. To
the picture of the mind that we started with, we added the insight that floats – as well as
conditional and quantificational floats – can come in degrees. This allows us to account for
the difference between options we take as more or less likely.
7. Epistemic Evaluations
In this chapter, I have posited a number of attitudes that agents use in reasoning with
uncertainty and I have discussed the rational norms that apply to having and reasoning with
these attitudes. In the rest of the chapter, I will discuss how agents can be in disagreement in
virtue of having combinations of float-like attitudes. My account says, roughly, that
disagreement between agents with float-like attitudes arises because of agents’ dispositions to
give negative evaluations of others’ attitudes. In order to set the stage for such an account, I
will, in this section, investigate how agents perform epistemic evaluations of others attitudes –
77
Note that there won’t be any generalized Steadfastness condition. This is because Steadfastness concerns just
sinks and nothing about floats being graded changes what we have said about sinks.
69
with an eye toward typical ways of evaluating floats and sinks.
78
It is natural to think that epistemic evaluations are not based on a single factor but,
rather, take place along at least two dimensions – an accuracy (or truth-related) dimension and
an evidential dimension.
79
We care about whether epistemic attitudes accurately represent the
world. And we care about whether they are well-supported by evidence.

In order to
understand the overall evaluation, we need to be familiar with the component evaluations. I
don’t aim to give a complete account of how these component evaluations work. A rough
picture suffices in order to understand the total evaluation that results from their combination.
Let’s first discuss the notion of accuracy. An attitude is accurate if it involves
acceptance of the truth (falsity) of a proposition and the proposition is true (false). And it is
inaccurate if it involves acceptance of the truth (falsity) of a proposition and the proposition is
false (true). If an attitude involves only partial acceptance of the truth (falsity) of a
proposition, then it will be only partially accurate.
80
A sink is an attitude that involves
acceptance of the falsity of a proposition. So, it can be completely accurate. A float, however,
does not involve acceptance of the truth or falsity of a proposition. So it cannot be completely
accurate.
How do agents evaluate the accuracy of others’ attitudes? They do so based on their
own thoughts about the contents of those attitudes. For example, in evaluating the accuracy of
a sink in a proposition, p, an agent will see if she accepts p’s negation. If she does, she will
positively evaluate the sink. If she accepts p, she will negatively evaluate the sink. And if she
accepts neither p nor its negation, she will refrain from making judgments about the accuracy
of the sink. Likewise, in evaluating the accuracy of a float in p, an agent will consult her own
attitudes. She will negatively evaluate a float in p if she accepts p’s negation, since she will
think a float in p is not fully accurate and there is some determinate way in which it could be
made more accurate.
81
Again, she will refrain from making accuracy judgments if she
78
The discussion in Chapter Seven will make clear that floats are not purely epistemic attitudes. So, I will later
slightly amend the picture of evaluation that I now give. But we aren’t in a position to undertake that
discussion yet.
79
Perhaps, truth is the sole fundamental epistemic value, and evidential value is derivative of it. Nonetheless,
respecting the evidence would still be epistemically valuable.
80
This can be generalized to finer-grained attitudes to coincide with Joyce’s (1998) notion of gradational
accuracy. Note, however, that in Joyce’s argument for probabilism, he assumes that we evaluate the expected
accuracy of an attitude. I am concerned with judgments of accuracy itself.
81
This last portion is important because I don’t negatively evaluate a float in p if I have no attitude about p,
just because I know such a float is not completely accurate. Without the latter clause, all states of partial
acceptance would be evaluated negatively by reasonable agents (even ones that are uncertain about that very
matter). But this is not, in fact, how evaluation works.
70
suspends judgment about p. In general an agent’s evaluations of an attitude’s accuracy
depends on the agent’s own take on the truth of the content of the attitude.
What about the evaluations of an attitude along the evidential dimension? Such
evaluations are less straightforward than evaluations of accuracy. The general idea is that an
agent will positively evaluate an attitude along this dimension just in case she thinks that the
attitude has evidential virtues. Luckily, it won’t be necessary for us to come up with a precise
list of these virtues (different agents likely hold different features of an epistemic strategy to be
evidentially virtuous or weight different features differently). Here are two candidate epistemic
virtues that go into an evidential evaluation, which will be useful for our discussion. An agent
might judge that an attitude is well-proportioned to the evidence. If the agent believes the
evidence strongly supports a proposition, then she’ll positively evaluate a belief that it is true
and negatively evaluate a belief that it is false. If she thinks the evidence gives some degree of
support then she’ll positively evaluate a float of similar degree. The other factor that will
primarily concern us is that of false or misleading evidence. If an agent thinks that an attitude
is based on false or misleading evidence, then she’ll negatively evaluate that attitude on the
evidential dimension.
One interesting issue regards what the evidence is for these evaluations. That is, when
agent A makes an epistemic evaluation, does she evaluate agent B’s attitude based on whether
it respects B’s evidence, whether it respects A’s evidence, or whether it respects some other set
of evidence? While A could evaluate B’s attitude with respect to any set of evidence, it is
natural for her to evaluate it relative to the evidence of whoever she takes to be the most
informed agent in the group investigating the question at hand. So, in a joint deliberation with
B, if she thinks that she has at least as much information as B, she will evaluate it relative to
her own evidence. If she thinks that B has more information than she has, she will refrain
from making this sort of evaluation. This makes sense of why agents tend to correct others
when they think they are more informed but defer to others when they think they are less
informed.
Now that we have a feel for evaluations of accuracy and evidential evaluations, let’s see
how they fit together to yield an overall epistemic evaluation. Here is my hypothesis:
Evaluations of accuracy always take precedence; evidential evaluations only come into play
when agents do not make an evaluation of accuracy. What this means is that, if an agent
evaluates the accuracy of an attitude positively, then her overall epistemic evaluation will be
71
positive. And if she evaluates the accuracy of an attitude negatively, then her overall epistemic
evaluation will be negative. If she does not evaluate the accuracy either way, then her overall
epistemic evaluation will depend on her evidential evaluation of the attitude – though this too
can be indeterminate, so that an agent can have an overall epistemic evaluation of an attitude
that is neither positive or negative. This structure of evaluation priority is natural when we
think about the epistemic enterprise. Though we care about respecting the evidence, we
generally care, first and foremost, about whether an agent is correct in having an attitude.
I take it that an overall evaluation of an agent’s attitude can be positive or negative.
Imagine that you are engaging in an evaluation of another agent’s attitude toward the
proposition, p. The evaluation will be positive if you evaluate the attitude to be just as
accurate as your attitude toward p. It will be negative if you evaluate it to be less accurate than
your attitude toward p. It will also be negative if you aren’t in a position to evaluate the
respective accuracies of both attitudes toward p, but you evaluate the other agent’s attitude
negatively on some evidential dimension. Similarly for a positive evaluation based on
evidential concerns. Otherwise, you will not evaluate the attitude either positively or
negatively.
82
In the case of floats and sinks, this plays out as follows: A sink in p involves acceptance
of p’s negation. An agent will evaluate the sink positively if she also accepts p’s negation. She
will evaluate it negatively if she accepts p (since she will think the sink is less accurate than her
attitude of accepting p). And if she suspends judgment as to whether p is true (perhaps by
floating p) and evaluates its evidential dimension negatively, then she will evaluate the sink
negatively. A float in p does not involve acceptance of any proposition. It will be positively
evaluated if the agent also floats p. But if the agent accepts either p or p’s negation, it will be
evaluated negatively.
The story about degreed floats is the same, but there is one important issue to note.
An agent with a degreed float will not make an evaluation of accuracy of another agent’s float-
like attitude, but she may evaluate it on the basis of whether it is well-proportioned to the
evidence. That is, she will offer a positive evaluation if she thinks that other agent’s attitude is
82
It seems that an agent can also evaluate her own attitude toward p. In cases where she is not conflicted in
her opinion about p, she will trivially evaluate her attitude toward p positively (since she will evaluate it
positively on either an accuracy or evidential dimension on the basis of that very attitude constituting her
opinion about p). But there may be cases where she is conflicted, of two minds, or irrational with respect to
p. She may, for instance, both think that p is true and take p’s negation as an option. If she evaluates one of
these attitudes on the basis of the other, she will give a negative evaluation of that attitude. Thanks to Scott
Soames for raising this issue.
72
well-proportioned to the evidence, and she will offer a negative evaluation if she thinks that it
is not.
Conditional floats and sinks are a bit more complicated, but the procedure for
evaluation is much the same as above. Let’s start with conditional sinks. An agent will judge a
conditional sink in p if q to be accurate (inaccurate) just in case, supposing q, she accepts that
p is false (true). For example, suppose that Phil has a conditional sink that Stu is the thief if
he says he isn’t. Lil will judge his conditional sink to be accurate if, supposing that Stu says
that he isn’t the thief, she accepts that it is false that he is the thief. And she will judge his
conditional sink to be inaccurate if, supposing that Stu says he isn’t the thief, she accepts that
it is true that Stu is the thief. If she accepts neither that it is false nor true that Stu is the thief
on the supposition that he says he isn’t, then she will refrain from evaluating the accuracy of
Phil’s conditional sink. But, as in the unconditional case, she may still make an epistemic
evaluation of Phil’s conditional sink on evidential grounds. This would involve evaluating the
evidential responsiveness of a sink that Stu is the thief, on the supposition that he said he isn’t.
An agent will judge a conditional float in p if q to be inaccurate if, supposing q, she
accepts that p is false. For example, suppose that Phil conditionally floats that Angelica is the
thief if she has crumbs in her pocket. Lil will judge this conditional float to be inaccurate if,
on the supposition that Angelica has crumbs in her pocket, she accepts that it is false that
Angelica is the thief. As in the unconditional case, an agent will not evaluate a conditional
float as accurate, since a conditional float does not involve taking a position about how the
world is – even on a some supposition. Nonetheless, conditional floats are available for
evidential evaluations in the same way that conditional sinks are.
The case of quantificational sinks can be more of a mouthful. An agent will judge a
quantificational sink, for Q of the Gs, that it is F, to be accurate just in case, supposing X is
the set of all and only the Gs, for each of Q of the members of X, she accepts that it is false
that that object is F. For instance, consider a case where the cookie theft was performed by a
group of people. Our agent has evidence that three of the people next door can’t be part of
the theft. She will have a quantificational sink, for three of the people next door, that they are
part of the theft. Another agent will evaluate this attitude positively on an accuracy dimension
if she has the same quantificational sink or one that is stronger than it (i.e. if she has a
quantificational sink, say, for four of the people next door, that they are part of the theft). In
such a case she will, supposing that X is the set of all and only the people next door, for each
73
of three of the members of X, accept that it is false that that member is part of the theft. As
usual, quantificational floats will never be positively evaluated for accuracy, since, on the
relevant suppositions, the floats can’t be completely accurate. As we’ve said before, they aren’t
the right sorts of things for complete accuracy to apply to.
I will state how negative evaluations of accuracy go for quantificational floats and sinks
in a more general way – drawing on principles from our earlier discussion. An agent will judge
a quantificational float or sink to be inaccurate when rationally updating her deliberative state
with that attitude yields a state that is in violation of the principle from §5, Commitment
Satisfaction. For instance, suppose Phil has a quantificational float, for at least three people
next door, that they are the thief. And suppose that Lil has a quantificational sink, for all but
two people next door, that they are the thief. Then Lil will evaluate Phil’s quantificational
float as inaccurate, since rationally adding it to her attitudes will result in a violation of
commitment satisfaction. She can’t rationally both retain her quantificational sink and take on
his quantificational float.
83
As expected, there can also be evidential evaluations of quantificational floats and
sinks. Consider a quantificational float, for Q of the Gs, that it is F. An agent will evaluate its
evidential responsiveness by evaluating, on the supposition that X is the set of all and only the
Gs, the evidential responsiveness of, for each of Q of the members of X, a float that that object
is F. Evidential evaluations work in the same manner for quantificational sinks.
In this section, I’ve given a picture of epistemic evaluations as comprising two
component evaluations – one of accuracy and one of evidential virtues. My hypothesis is that
evaluations of accuracy always take precedence, and it is only when no evaluation of accuracy
is made that an overall epistemic evaluation depends on the evidential evaluation of an
attitude. I closed the section by mentioning how the account applies to other sorts of float-
like attitudes.
8. Disagreement
Consider the following case: Phil is reasoning about who stole the cookies. He floats
that Angelica is the thief. Lil is also reasoning about who stole the cookies. However, she has
83
Note that it doesn’t follow from this that Phil will evaluate Lil’s quantificational sink as inaccurate. This is
because, when Phil updates his attitude with Lil’s, it is rational for him to relinquish his quantificational float
in favor of her quantificational sink. The reason it is irrational for Lil to do something similar is another
principle that we saw in §5 – Quantificational Steadfastness – which says that it is never rational (outside of
retraction scenarios) to give up quantificational sinks.
74
evidence that shows that Angelica is not the thief. So, Lil sinks that Angelica is the thief. That
is, Lil rules out that Angelica is the thief while Phil takes that proposition as an option. It
seems that in such a case Lil is committed to thinking that Phil is wrong. It seems that Lil
disagrees with Phil. What is the nature of this disagreement?
The first thing to notice is that this disagreement cannot be explained by a flat-footed
account. Suppose that Tommy believes that Angelica is a philosopher and Chuckie believes
that she isn’t a philosopher. Tommy and Chuckie disagree, and the flat-footed account says
that this is because one of them thinks the proposition that Angelica is a philosopher is true
and the other thinks it is false. This is an instance of the following, very popular theory of
disagreement:
Inconsistentism: Suppose that agent, A, has a belief in a proposition, p. Another
agent, B, disagrees with A if p is inconsistent with the contents of B’s beliefs.
Inconsistentism cannot explain the disagreement that is present in the case of floats and sinks.
In the case we have been discussing, Phil floats that Angelica is the thief and Lil sinks that
Angelica is the thief. Floating a proposition is not just a matter of believing some proposition.
So, in such a case, Inconsistentism simply won’t apply.
Though Inconsistentism cannot explain how floats can feature in disagreement, it is
important to note that it does not preclude floats from so-featuring. This is because
Inconsistentism is a sufficient condition for disagreement and is silent on what conditions are
necessary. In what follows, I will explore a way of enriching this account of disagreement,
which will retain Inconsistentism as a sufficient subcondition for disagreement, while allowing
for cases of disagreement featuring floats.
My account explains the sort of disagreement that arises between Lil and Phil in terms
of the evaluations that agents make of others’ attitudes – using tools from the last section. In
particular, I will claim that giving a negative evaluation of an agent’s attitude is a way of
disagreeing with that agent. And I will suggest that one is in disagreement with an agent if one
is disposed to give a negative epistemic evaluation of that agent’s attitude. Likewise for
positive evaluations and agreement. This notion of disagreement will not only explain the
case just discussed. It will also, as we will see, allow a surprising fact that Inconsistentism did
not – the fact that disagreement can be asymmetric.
The act of evaluating an attitude negatively is, I submit, an act of disagreeing. When I
evaluate your attitude negatively I think that you are wrong. That is, I disagree with you.
75
This is the notion of disagreeing as an action; what I am doing is performing an act of
disagreeing. But, it is possible, at some time, for an agent to be in a state of disagreement with
another even if there is no act of disagreement taking place at that time.
84
How is this possible
if I want to base disagreement on negative evaluations – which are acts? The answer is that
the state of disagreement obtains in virtue of an agent being disposed to negatively evaluate an
attitude in circumstances where an evaluation based on the best available evidence is
appropriate. Here, more precisely, is the characterization:
Evaluationism: Suppose that agent, A, has an epistemic attitude, x. Another agent, B,
disagrees with A if B is disposed to give a negative epistemic evaluation of x.
85
Evaluationism allows an agent to think another is wrong in having an attitude – even if that
attitude is not a belief whose content is inconsistent with her beliefs (as Inconsistentism
requires).
We can also give a characterization of agreement in terms of dispositions toward
positive evaluations:
Evaluationism Agree: Suppose that agent, A, has an epistemic attitude, x. Another
agent, B, agrees with A if B is disposed to give a positive epistemic evaluation of x.
I’ll be focusing primarily on disagreement in what follows.
We are now in a position to use Evaluationism to explain the disagreement in our
target case involving Phil and Lil. Remember that an agent, A, will be disposed to negatively
evaluate (on accuracy dimensions) a float in p if A accepts that p is false. If another agent, B,
has a float in p, then A disagrees with B by Evaluationism. This is the standard sort of
disagreement involving floats and sinks. Our example involving Phil and Lil is just this sort of
case. Lil thinks it is false that Angelica is the thief, while Phil floats that she is. So, Lil is
disposed to negatively evaluate Phil’s attitude (she is disposed to think it is not fully accurate).
So, according to Evaluationism, Lil disagrees with Phil. Thus, Evaluationism allows us to
explain our central example.
However, when we think about the ways that agents negatively evaluate attitudes, we
can see that there other ways in which one agent can disagree with another according to
Evaluationism. In the example above, what underpinned disagreement was a negative
evaluation of accuracy, but an agent can also disagree based on a negative evidential evaluation
84
See Cappelen and Hawthorne 2009 (60-61) and MacFarlane MS (4) for more on this act/state distinction.
85
Perhaps, we should be more cautious in our statement of Evaluationism – in order to avoid bad predictions
stemming from nonstandard or irrational dispositions. I won’t concern myself with whether qualifications
are necessary and, if so, what exactly they should look like.
76
(when she is undecided about the accuracy of the attitude in question). For instance, an agent,
A, is disposed to negatively evaluate a sink in q if A is undecided about q but knows that the
evidence used in forming the sink in q is misleading. If another agent, B, has a sink that was
formed on the basis of misleading evidence, then A disagrees with B by Evaluationism. Here
is such an example. Suppose that Susie sinks that Tommy is the thief on the basis of Betty
telling her that Tommy isn’t the thief. Stu knows that Betty told Susie this, but he also knows
that Betty didn’t really know whether or not Tommy is the thief. That is, he knows that the
evidence that Susie takes to rule out the possibility that Tommy is the thief doesn’t really rule
out this possibility. So, Stu floats that Tommy is the thief. Evaluationism predicts that Stu
disagrees with Susie since he is disposed to negatively evaluate Susie’s float (since he thinks it
was formed on the basis of misleading evidence).
We’ve now seen an example where Evaluationism predicts that an agent with a sink in
a proposition disagrees with an agent with a float in it and an example where Evaluationism
predicts that an agent with a float in a proposition disagrees with an agent with a sink in it. It
is important to realize that Evaluationism makes these predictions for different reasons and to
realize that the former prediction is, in a sense, much more general than the latter.
Evaluationism predicts that an agent with a sink in a proposition will, in general, disagree with
an agent with a float in that proposition. This is because such an agent will, in general, be
disposed to negatively evaluate such a float as not fully accurate. On the other hand,
Evaluationism doesn’t make a general prediction that an agent with a float in some
proposition disagrees with an agent with a sink in it. Whether Evaluationism makes such a
prediction depends on whether, in the particular case, the agent is disposed to offer a negative
evidential evaluation of the other agent’s sink (since she won’t be disposed to offer a negative
evaluation of the sink’s accuracy). Such negative evaluations are comparatively rare, so
disagreement in that direction is predicted to be much more rare.
In particular, Evaluationism does not predict that agent A disagrees with agent B in the
following case: A floats r and is not disposed to give either a positive nor a negative evaluation
of B’s sink in r, since A does not know whether it is accurate, nor does she know who is better
informed about r. This structure is actually instantiated by our example involving Phil and
Lil. Though we’ve seen that Lil disagrees with Phil, Evaluationism does not predict that the
disagreement goes both ways. Phil floats that Angelica is the thief and is not disposed to
negatively evaluate Lil’s sink on either an accuracy or an evidential dimension.
77
This means that Evaluationism does not rule out the following, prima facie surprising,
claim – that disagreement can be asymmetric.
86
I can disagree with you without you
disagreeing with me. This is because I can be disposed to give a negative epistemic evaluation
of your attitude without you being disposed to give a negative epistemic evaluation of mine. It
is important to note, however, that Evaluationism gives only a sufficient and not a necessary
condition for disagreement. So, it could be that a full account does rule out this asymmetry.
But, though I won’t argue for it here, I suspect a full account won’t rule out asymmetric
disagreement. This is because, as surprising as it might sound at first, on reflection it seems
like a natural characterization of the disagreement in our example case. Though it seems that
Lil disagrees with Phil, it doesn’t seem that Phil disagrees with Lil. Phil is not committed to
thinking that people who sink that Angelica is the thief are wrong. He leaves it open that he
may, in the future, learn that Angelica isn’t the thief. He is just undecided as of now. So it
does not seem like Phil disagrees with Lil. That is, it seems that the disagreement between
Phil and Lil is asymmetric. In general an agent with a float in a proposition will be open to
coming to have a sink in that proposition if she acquires evidence of the right sort.
87
It is
natural to think that such an agent doesn’t, in general, disagree with people who sink that
proposition (the exception is the case of thinking that that person has false or misleading
evidence).
One may worry that, unlike the disagreement between agents with contradictory beliefs,
the disagreement between an agent who sinks a proposition and an agent who floats it is not
particularly deep or persistent. Lil need not think that Phil has made any particular mistake
in his reasoning. Nor need she think that he is worthy of blame for anything.
88
Phil and Lil
won’t encounter any deep impasse regarding how to resolve their differences. It is important
to remember that I am going to use Evaluationism to help explain why agents can respond in
particular ways to utterances of epistemic sentences. And it will turn out that the shallowness
and asymmetry of the disagreement we have just seen mirrors the shallowness and asymmetry
of the sort of disagreement evidenced in conversations like (Cookies).
It is also important to keep in mind that Evaluationism is a strictly more general
account of disagreement than Inconsistentism. Though it predicts more instances of
86
See Ross and Schroeder 2013 and Dietz 2008 for discussions of asymmetric disagreement involving modal
thoughts (though Dietz is primarily concerned with what he calls correct disagreement).
87
This hearkens back to a conclusion from §3 – that in non-retraction cases agents can rationally proceed from
floating to sinking a proposition but not vice-versa.
88
This may help to account for a judgment that many theorists have – that the sort of disagreement that takes
place in (Cookies) is faultless. See Kölbel 2004.
78
disagreement – as we have discussed – it also predicts that agents with inconsistent beliefs
disagree. This is because any agent who believes a proposition will be disposed to negatively
evaluate a belief in that proposition’s negation (since she will find it to be inaccurate). And
since inconsistency is a symmetric relation, this sort of disagreement – that involving beliefs
with inconsistent contents – will be symmetric as well. Thus, Evaluationism is conservative
with respect to Inconsistentism over the domain to which they both apply, and it adds
plausible predictions in cases where Inconsistentism does not apply. So, Evaluationism looks
like a plausible sufficient condition on disagreement.
89
Given our conclusions at the end of last section about epistemic evaluations of other
sorts of float-like attitudes – degreed floats, conditional floats, and quantificational floats – we
have an explanation of how they land their bearers in disagreement with others. For instance,
an agent with a sink can disagree with an agent with a degreed float in the same way that she
can disagree with an agent with a bare float. More interestingly, an agent with a float to some
degree can disagree with an agent with a float to a different degree. Such disagreement is
based on an evaluation of whether the other agent’s float is well-proportioned with respect to
the evidence (and not on whether it is judged to be accurate). For example, suppose that Phil
has a just-a-little-bit likely float that Tommy is the thief and Lil has a very likely float that
Tommy is the thief. They can be in disagreement if each is disposed to negatively evaluate the
other’s attitude – on the basis of thinking that it is not well-proportioned to the evidence.
90
There can also be disagreement between agents who have conditional floats and sinks.
For example, suppose that Phil conditionally floats that Stu is the thief if he has crumbs in his
pocket and Lil conditionally sinks that Stu is the thief if he has crumbs in his pocket. Then
Lil will be disposed to negatively evaluate Phil’s conditional float. Remember we saw in the
last section that such an evaluation proceeds by supposing the antecedent and then evaluating
a float in the consequent. When Lil does this, she will judge the resulting float to be
inaccurate and, so, will give a negative evaluation of Phil’s conditional float. So
89
The account of disagreement that I have just outlined shares important general features with the account in
Finlay 2014 (Ch. 8). Both allow that disagreement can arise from dispositions to engage in a certain type of
action (response or evaluation). Both allow that disagreement may be asymmetric. And, as will become clear
for my account in the next chapter, both allow that disagreement in attitude may be paired with a
contextualist account of the relevant portion of language. Thanks to Stephen Finlay for discussion and for
pressing me about some of the details of my picture.
90
One might hope that the sort of explanation offered here of disagreement between agents with different
degreed floats will extend to explain disagreement between agents with different credences. I think it does
extend to these attitudes in a straightforward way. For some discussion see Lennertz MS a (§5).
79
Evaluationism, correctly predicts that she disagrees with Phil in this case.
91
There are also
cases of disagreement stemming from evidential evaluations, but I will leave the construction
of such examples to the reader.
92
In this section, I used the notion of epistemic evaluations from the last section in
forming a condition on disagreement, Evaluationism, which explains how agents can disagree
in virtue of having floats and sinks. Evaluationism can explain interesting phenomena like
asymmetric disagreement. I also showed that Evaluationism is conservative with respect to
Inconsistentism over the domains to which they both apply.
9. The Plan Going Forward
In this chapter, I developed a picture of the sorts of attitudes that agents use in
reasoning with uncertainty. In the next three chapters I will explore the connection between
the picture of the mind that I’ve sketched here and epistemic language. My hypothesis is that
speakers use epistemic language to express different float-like attitudes. This makes sense
because epistemic language is used to coordinate reasoning in situations of uncertainty and
float-like attitudes are used in reasoning in situations of uncertainty. In the next chapter, I’ll
show how speakers typically express floats when they use simple ‘might’-sentences. In the
following chapter, I’ll explore complex ‘might’-sentences – ‘might’-sentences which are
embedded in larger sentences. In various cases, we’ll see that speakers express sinks,
conditional floats, and quantificational floats in using such sentences. In Chapter Six, I’ll
move beyond the word ‘might’ and the notion of possibility to language that is used to talk
about necessity and probability. We’ll see that epistemic ‘must’ sentences are used to express
sinks and sentences about probability are used to express degreed floats. In each of these three
chapters, the norms on rationality involving floats and sinks, as well as the conditions on
disagreement, will play a role in explaining some otherwise challenging linguistic data. In the
final chapter, I will return to discuss whether float-like attitudes are reducible to some other
sort of attitude that we already accept.
91
Indeed, she would also disagree with him if she had an ordinary sink that Stu is the thief, since her
evaluation would proceed in a similar manner (this time the supposition would have no effect on her
evaluation of his attitude).
92
An agent who has a quantificational sink can also disagree with an agent who has a quantificational float. For
example, Lil, who has a quantificational sink, for each person next door, that she is the thief, can disagree
with Phil, who has a quantificational float, for some person next door, that she is the thief.
80
Chapter Four: How Simple ‘Might’-Communication Works
There is connection between epistemic language and the mind. We use ‘might’-
sentences to communicate when we are reasoning collectively in situations of uncertainty.
Since floats allow agents to structure their reasoning under uncertainty, it is useful for
members of a group to coordinate what propositions they float in an inquiry. So, it is natural
to think that agents use ‘might’-sentences to coordinate which propositions to float. This is a
general description of the view to be presented in this chapter.
I want to begin by giving a brief sketch of my theory of communication involving
‘might’-sentences. In the later sections, I will present the theoretical basis for this theory. My
theory starts with Contextualism:
Contextualism: The direct speech act performed by a speaker who assertively utters
┌
It might be that S
┐
, where p is the content of S in the context, is an assertion of the
proposition that p is consistent with the information of the context.
We saw in Chapter One that Contextualism fit well with simple and unified compositional
semantic accounts of natural language modals.
However, there is more going on in ‘might’-communication than Contextualism
describes. A speaker of a ‘might’-sentence usually doesn’t just assert an MP about what is
consistent with his information. To figure out what else he conveys, we should think about
the purpose of typical conversations in which ‘might’-sentences are used.
93
Speakers in these
conversations want to aid in collective reasoning. As we saw in the last chapter, what is
important in reasoning with uncertainty is not what propositions are consistent with some
agent’s information. Rather, the propositions the agent floats are what matter. So, in order to
really aid in reasoning, a speaker will not want to convey merely what is consistent with his
information. He will also aim to convey that some possibility is an option in reasoning.
94

93
In focusing on the purposes of using particular sorts of language, I follow the strategy that Björnsson and
Finlay (2010) pursue for deontic modals and Björnsson and Almér (2010) pursue for many types of context-
sensitive language.
94
There is much in common between this approach and Stephen Toulmin’s (1958, 18) in The Uses of
Argument:
[I]n dealing with any sort of problem, there will be an initial stage at which we have to admit that a
number of different suggestions are entitled to be considered. They must all, at this first stage be
admitted as candidates for the title of ‘solution’, and to mark this we say of each of them, ‘It may (or
might) be the case that . . . .’ At this stage, the term ‘possibility’ is properly at home, along with it’s
associated verbs, adjective and adverb: to speak of a particular suggestion as a possibility is to
concede that it has the right to be considered.”
And Toulmin (1958, 36-37) says:
From the standpoint of mathematics, we may be justified in treating some notion as a possibility
simply in the absence of any demonstrable contradiction – this is the converse of contradictoriness,
the mathematical criterion of impossibility. In most cases, however, to call something a possibility is
to claim much more than this. . . . In order for a suggestion to be a ‘possibility’ in any context,
81
More precisely:
EXPRESSING FLOATS: A speaker who utters a sentence of the form
┌
It might be
that S
┐
, where p is the content of S in the context, typically indirectly expresses, as a
standardized speech act, a float in p.
The rest of this chapter will be devoted to making sense of EXPRESSING FLOATS – in
particular, what it means to indirectly express an attitude as a standardized speech act – and
applying it to different sorts of ‘might’-communication.
More precisely, the plan for the chapter is the following: In §1, I will characterize the
speech act, expression, and discuss why speakers express floats. In §2, I will explain the sense
in which this expression in the case of ‘might’-sentences is indirect. And in §3, I will argue
that this indirect expression is what we can call a standardized speech act and show how this
supports a new theory of at-issue speech acts. Together, §1-3 provide the theoretical
underpinning for my explanation of conversations involving uses of and responses to uses of
simple ‘might’-sentences. In §4, I’ll apply the framework in explaining the different types of
uses of ‘might’-sentences discussed in Chapter One and Chapter Two.
1. Expression
My explanation of many conversations involving uses of ‘might’-sentences relies on the
notion of expressing an attitude. So what is it to express an attitude? I am interested in the
ordinary sense of ‘expression’ – the one we talk about in everyday scenarios. We talk about
expressing one’s beliefs, one’s pain, one’s regrets, one’s gratitude, etc. This is in contrast to a
technical use of ‘express’ that is common in the philosophy of language – that a sentence in a
context expresses a proposition.
therefore, it must ‘have what it takes’ in order to be entitled to genuine consideration in that context.
To say, in any field, ‘Such-and-such is a possible answer to our question’, is to say that, bearing in
mind the nature of the problem concerned, such-and-such an answer deserves to be considered.
Toulmin (1958, 18) does not think that he is giving a semantic analysis of these words: “In connecting up the
words ‘possible’, ‘possibly’, ‘may’ and ‘might’ with this initial stage in the presentation of an argument, I do
not see myself as presenting a formal analysis of the term ‘possible’.” Thanks to Jonathan Dancy for drawing
my attention to Toulmin’s work.
This important way of thinking is gaining popularity and is a major part of some of the new
accounts of epistemic modals. Braun’s (2012) account incorporates the idea that uses of ‘might’-sentences are
importantly linked to agents’ taking propositions seriously. And Willer (2013, 50) says, “might-statements are
designed to change possibilities that are merely compatible with the agent’s evidence into ‘live possibilities’ –
possibilities that are compatible with the agent’s evidence and that the agent takes seriously in inquiry.” The
insight is also affirmed in Bach 2011 (57): “When you use a bare EP sentence [simple ‘might’-sentence]
assertively, ordinarily you do not assert a mere idle possibility. If it is worth mentioning, presumably you
take it to be a more serious possibility than that and intend it to be taken as such” and in von Fintel and
Gillies 2008 (83): “the speaker [of a ‘might’-sentence] is highlighting that possibility as one that should not be
ignored.” And Stephenson (2007, 516) suggests something similar: “pragmatic factors typically require there
to be some reason for bringing up a particular epistemic possibility, for example if there is reason to believe
that it’s fairly likely.” I believe this sort of idea was first developed in the recent literature in Swanson 2006
(§2.3.2), though it is mentioned in passing in Hawthorne 2004 (Ch.1 fn. 63).
82
Expression, in this sense, is a type of act, and often is a speech act. I can express my
belief that Hillary Clinton will win the 2016 presidential election by saying, “Hillary Clinton
will win the 2016 presidential election.” I can express my pain by saying “Ouch!”. I can
express my gratitude by saying “Thanks.” Expressing is something agents do, and something
they often do with words. One can express mental states without words. For instance, one
can express one’s pain with an intentional grimace. However, I will largely be concerned with
expression as a speech act. The other important feature of expression is that its objects are
mental attitudes. The things that we express are states of mind like beliefs, pains, regrets,
gratitude, etc.
It may be helpful to give an example of an analysis of expression. I don’t mean to
commit to the following account from Wayne Davis (2003, 59), but it is in the right ballpark:
For a subject, S, and an attitude, Ψ, “S expresses Ψ iff S performs an observable act as an
indication of occurrent Ψ . . .”
95
This account respects our two observations about expressing
– that expressing is an act (often a speech act) and that the things we express are attitudes.
Davis’s account is neutral with respect to the sort of attitude that can be expressed. For
example, to express a belief is to perform an observable act as an indication of an occurrent
belief, to express pain is to perform an observable act as an indication of occurrent pain, and,
we might think, to express a float is to perform an observable act as an indication of an
occurrent float. If this is right, then expressing a float is no more mysterious than expressing
any other sort of attitude.
Still, we might wonder why an agent would express a float. To answer this, it is
important to say a bit more about how expressing attitudes fits into a broader theory of
communication and rational action. Why do we express attitudes? In some cases it is merely
to inform our listeners about them. In others it is to get them to adopt these very attitudes.
Saying ‘Ouch!’ is a nice example of the first sort. A speaker does this merely to indicate her
pain. In many cases of saying “Hillary Clinton will win the 2016 presidential election”,
however, the reason the speaker expresses her belief is to inform her listeners for the purpose
of their sharing the belief expressed.
Most cases in which we express floats are attempts at coordination. Why, in general,
would we want to coordinate our attitudes with others? Well, there really isn’t a reason in the
case of pain, which is good because we don’t try coordinate in those cases. But in the case of
belief, it makes sense why we want to coordinate. We want our listeners to best achieve their
95
Davis goes on to say that this act must be performed “without thereby covertly simulating an unintentional
indication of Ψ.” But we can safely ignore this part of the account which is intended to deal with obscure
counterexamples. See Davis 2003 (§3.4) for discussion.
83
ends. And their ends will, by and large, be best achieved if they have the correct beliefs about
the facts.
96
Why would we express floats, in particular? We use language for many purposes.
One of them is to coordinate about what to believe. Another is to coordinate about what to
do. One important purpose of using language is to coordinate about how to best reason – and
how to do so in situations of uncertainty. In the case of reasoning under uncertainty, we may
try to influence not just a hearer’s beliefs, but what she floats. As we’ve seen, what she floats
plays a pivotal role in reasoning with uncertainty. And we can influence what she floats by
expressing a float. If she treats us as an authority and accepts our expression, then she will
come to have that float. And we will have affected her reasoning as desired.
Whether we express a float with a use of a ‘might’-sentence depends on what our
purposes are. In one kind of case, the speaker is engaged in or an advisor to some
deliberation about some question – say, the question of whether q. And he wants to help get
the right answer to q. Suppose the speaker floats some proposition, p, relevant to whether q.
It seems plausible that he would be interested in getting those deliberating to share his float in
p, since he thinks it would aid them in finding the correct answer to q. Because the speaker is
concerned with coordinating the group’s reasoning, it makes sense that he expresses his float
in p. Most speaker-centric and group-centric uses of ‘might’-sentences – examples of which we
saw in Chapter One – are of this sort.
The other sort of case is one where the speaker does not intend to help his audience
with some reasoning by getting them to share a float that he has. Since the speaker does not
intend to affect the course of reasoning by coordinating floats with his audience, he doesn’t
express a float by using a ‘might’-sentence in such cases. Exo-centric uses – which we also
surveyed in Chapter One – fall into this second category.
In this section, we saw that it sometimes makes sense for agents to express floats. It is
reasonable to hypothesize that they often do this by uttering ‘might’-sentences. Still, we need
to investigate the precise connection between ‘might’-sentences and the expression of floats.
This connection is rather systematic, though it’s not as simple as having a float be either the
semantic value of a ‘might’-sentence or the object of the direct speech act performed by a use
of a ‘might’-sentence. Rather, as I will argue in the next two sections, a speaker who uses a
‘might’-sentence typically expresses a float as a standardized indirect speech act.
96
This also explains why we sometimes express beliefs we don’t have. In those cases we know that our
interlocutor’s ends will be better achieved if she acts on a false belief (the case of lying to someone, as we say,
for her own good).
84
2. Expression of a Float as an Indirect Speech Act
EXPRESSING FLOATS says that when a speaker utters a simple ‘might’-sentence, he
typically indirectly expresses a float. I take an indirect speech act to be one that is not
determined by the meanings of the words, the mood of the sentence, the way in which the
words are combined, and contextual factors resolving demonstrative reference. In this section,
I will show how the expression of a float by a speaker who utters a ‘might’-sentence has three
general features of indirect speech acts.
Our model will be the standardized indirect request performed by a speaker who uses
the sentence, “Can you close the door?”.
97
Here are three characteristic properties of an
indirect speech act. First, it is often the speech act of primary importance – for instance, a
speaker who utters “Can you close the door?” primarily requests that the addressee close the
door, even though the direct speech act associated with this sentence is an act of asking about
the addressee’s abilities. Second, the indirect speech act is not always performed by a speaker
who forcefully uses that sentence – for instance, when a physical therapist uses “Can you close
the door?” to merely ask about her patient’s abilities. Third, that an indirect speech act is
performed can be figured out based on the meaning of the sentence and general pragmatic
features of the context. Searle (1975) gives a detailed explanation of how this is so for “Can you
close the door?”.
98
As we will see, Phil’s expression of a float in uttering, “Angelica might be the thief”,
appears to be an indirect speech act, since it shares all of these features.
99
The indirect speech
97
This is a classic sort of example in the literature. Many of the features to follow in this subsection and the
next are noted in Sadock 1974, Searle 1975, Davison 1975, Morgan 1978, Clark 1979, Bach and Harnish 1979,
and Asher and Lascarides 2001. Here are some other examples of sentences that are often used to perform
standardized indirect speech acts: “Do you have change for a dollar?”, “Do you want to do the dishes?”, and
“I want you to close the window.”
98
A hearer can calculate that the speaker has issued a request to close the door in the manner that has the
flavor of the examples in Grice 1989b. Suppose Lil is in class sitting next to the open door. There is a lot of
noise in the hallway, and she cannot hear her teacher very well. In a loud voice, he utters, “Can you close the
door?” to Lil. She could go through the following reasoning: he uttered “Can you close the door?”. So he’s
asked if I have the ability to close the door. But our exchange does not have the goal of figuring out my
abilities. So his contribution is not relevant unless he meant to perform some other speech act. Since it is
noisy outside, we desire quiet, and closing the door is a way to achieve that, it makes sense that he wants the
door closed. One way to achieve this is to request that I close the door. Having the ability to close the door
is necessary for fulfilling a request to close the door. And being asked if I have this ability makes such a
request salient. Such a request would be cooperative while a simple question about my abilities would not
be. He knows this and knows that I know that he knows this. And he has done nothing to stop me from
thinking he has made such a request. So he must have requested that I close the door. In this way Lil can
derive the request in a Gricean manner.
99
Something like this account has been put forward independently by Martin Montminy (2012). His view is
mistaken about the indirect speech act that is performed – he says it is a weak suggestive rather than an
expression of a float. In the next chapter, we will see how this mistake causes him to miss out on the
explanatory power of my account for embedded ‘might’-sentences. Nonetheless, our views share a kernel of
truth. We both realize that in the case of simple ‘might’-sentences, speakers typically perform standardized
indirect speech acts related to the attitudes used in uncertainty. So, I will take on many of his observations in
this subsection and the next.
85
act performed by a speaker who uses a simple ‘might’-sentence is often the speech act of
primary importance. For instance, in the case of (Cookies), Phil asserts that it is consistent
with his information that Angelica is the thief, but the speech act of primary importance –
since it is more useful in guiding the group’s reasoning – is his expression of a float that
Angelica is the thief.
But an expression of a float is not always performed by a speaker who utters a ‘might’-
sentence (forcefully – eg. not in a play). Though we typically use ‘might’-sentences primarily
for the purpose of expressing floats, we don’t always do so. Exo-centric uses yield examples of
such cases. The following case is very much like the battleship case from above. Suppose that
there is a contest to see who can figure out who stole the cookies. Lil is the judge of the
contest and, so, already knows who is the thief:
(Cookies Contest) Phil: Angelica might be the thief
Lil: Yeah. She might be.
Lil does not express a float in this case. She does not perform the indirect speech act. She
merely asserts that it is consistent with Phil’s information that Angelica is the thief.
The third property of indirect speech acts says that a hearer can use reasoning with a
Gricean flavor to figure out that a speaker who uses a ‘might’-sentence expresses a float,
though I don’t mean to claim that a complete Gricean derivation is always available. Here is
the reasoning that Lil could go through in (Cookies): Phil asserted that it is consistent with his
information that Angelica is the thief. But there are many, many propositions that are
consistent with his information (and, indeed, many, many propositions about who is the thief
that are consistent with his information). Given that our concern is whether Angelica is the
thief, Phil’s assertion would not be very relevant unless he meant to convey his float that
Angelica is the thief. Since he aims to help in reasoning in a case of uncertainty, it would be
natural and much more relevant for him to express his float that Angelica is the thief. Since I
am assuming that he is cooperative, I should conclude that he has expressed his float that
Angelica is the thief. Again, I don’t mean this to be a full Gricean derivation. Indeed, we will
see in the next subsection that properties of standardization, applied to both ‘might’ and our
model question/request, speak against a purely Gricean treatment. I have just tried to display
the Gricean flavor of the reasoning that a hearer can go through.
In this section, we’ve seen that the expression of a float by a speaker who utters
“Angelica might be the thief” has these three characteristic features of indirectness. So, it
seems to be an indirect speech act.
86
3. Expression of a Float as a Standardized Indirect Speech Act
In this section, I will argue that the indirect expression we’ve been discussing is what
Bach and Harnish (1979) call standardized – by showing that it has three common properties
of standardization. In this way, the indirect expression is different from one paradigm sort of
indirect speech act – a Gricean conversational implicature. Implicatures do not usually have
the three features that follow and, so, are not standardized.
100
Again, I will use the connection
between the sentence “Can you close the door?” and the standardized indirect request that the
addressee close the door as a model for understanding standardization.
We saw in the last section that hearers can calculate which indirect speech act is
performed. But, just because hearers can do this does not mean that they always do. Indeed,
uptake of indirect speech acts that are standardized typically does not involve such reasoning.
Bach and Harnish (1979, 174) describe this property of standardized speech acts by saying that
the possible inference is ‘short-circuited’. This choice of terms is in one way unfortunate,
since it suggests that something goes wrong and the communication fails. This is not so.
Rather, the term is supposed to suggest that communication succeeds directly by a shortcut
around any sort of inference or calculation. Our example of an indirect request performed by
a speaker who utters “Can you close the door?” seems to have just this sort of property. No
process of reasoning seems necessary to figure out that such a request was performed.
101
A second property of standardized speech acts is that they are detachable – that
utterances of sentences that are performances of the same direct speech act in the same
context aren’t always performances of the same indirect speech act. So, two speakers who
assert the same propositions in the same context using different sentences may perform
different standardized indirect speech acts. In our example, we can contrast “Can you close
the door?” with “Do you possess the ability to close the door?” Both sentences are used to
directly ask about the addressees abilities, but the first and not the second is typically used to
request that the addressee close the door.
102
It is worth realizing that this property seems
100
These features have led many theorists – eg. Sadock (1974), Searle (1975), Morgan (1978), and Asher and
Lascarides (2001) – to see standardized indirect speech acts as being conventional in some way. Sadock (1974,
158) draws an analogy between this case and that of metaphors and idioms:
[T]he original meaning gradually grows weaker, the new meaning correspondingly gains strength.
The number of formal properties that correlate with the apparent sense compared to the number
that depend on the idiomatic sense would be a measure of the relative strengths of the two
underlying semantic representations. Whimperatives [sentences like “Can you close the door?”], for
example, behave almost entirely like the requests for which they are used, but also display a few of
the formal properties of the questions that they appear to be. This might be taken as an indication
that the request sense of such sentences has grown very strong, but that a trace of the historically
prior question sense remains.
101
I should note that this may also be true of some instances of familiar ways of conveying conversational
implicatures, but is not true of such implicatures in general.
102
Perhaps this latter sentence can be used to issue a sarcastic indirect request in the right context. It is,
87
pretty peculiar to standardized indirect speech acts. As Grice (1989b) notes, ordinary
conversational implicatures, by contrast, are not detachable (unless they depend on the maxim
of manner).
The third property is that standardized speech acts can be at-issue. So a hearer can
respond to a standardized speech act in a straightforward way (again, this appears to contrast
with conversational implicatures – see (Petrol') from Chapter Two for an example). For
instance, a hearer can refuse to perform the action requested by a speaker who utters, “Can
you close the door?”, by saying, “No, I can’t” (even if she has the ability to close the door).
103

As Clark (1979), Bach and Harnish (1979), and Asher and Lascarides (2001) note, the response
can take the form that is typically appropriate as a response to the direct and not the indirect
speech act. The hearer’s response in the example just given – “No, I can’t” – seems to be a
refusal to satisfy the request dressed up as an answer to the question about her ability.
The three aforementioned properties of standardized speech acts hold of the
expressions of floats by speakers who utter ‘might’-sentences (as Montminy (2012) notes).
First, when a speaker utters a ‘might’-sentence, a hearer can interpret the utterance and grasp
the indirect speech act without calculation. Hearers seem to interpret utterances of ‘might’-
sentences as expressions of a float by default (though this default can be overridden – say, in
exo-centric uses).
Furthermore, the acts of expression that speakers perform in using these sentences are
detachable. Consider the sentence, “It is consistent with my information that Angelica is the
thief.” This sentence is used to perform the same direct speech act as a speaker-centric use of
“Angelica might be the thief.” But it is not used to express the speaker’s float. It is merely
used to assert that it is consistent with the speaker’s information that Angelica is the thief.
104
Finally, a hearer can straightforwardly respond to the indirect expression performed by
a speaker who utters a ‘might’-sentence. When Phil expresses his float in (Cookies), Lil can
however, less natural to interpret a use of this sentence as an indirect request (as compared to “Can you close
the door?”). Thanks to Stephen Finlay reminding me of this point.
103
If one thought, as Stephen Finlay (pc) does, that ‘can’ gets a bouletic, rather than an ability reading, one
could maintain that, in this case, the response is to the direct speech act – the question about what is
consistent with your desires. But thinking of the other examples of indirect speech acts from note 97 suggests
that this won’t succeed as a general strategy. For example, when I walk into a restaurant and say, “Do you
have change for a dollar?”, the woman can appropriately reply, “Nope. Sorry. We only give change to
paying customers.”
104
This feature appears to be strongly in tension with a purely Gricean account of the phenomena. But, again, I
do not aim to give a purely Gricean account. Rather, I aim to assimilate this distinctive data involving
‘might’ to that involving sentences like “Can you close the door?”. I take the distinctiveness of the data to be
evidence in favor of my treatment of ‘might’. Nonetheless, I don’t want to rule out a Gricean account of this
detachability – relying on Grice’s (1989b) Maxim of Manner – where the indirect speech act is performed by
uttering the shorter, more common expression and not by uttering the longer, more obscure one. Finlay
(2005) pursues this sort of strategy in defending the position that uses of sentences containing evaluative
terms conversationally implicate something about the speaker’s goals, purposes, or standards.
88
respond to his expression in a straightforward way. She can say “No.” She must be
responding to what he has indirectly expressed and not what he asserted (since, as we
belabored in Chapters One and Two, she has no reason to think that what he asserted is false).
Furthermore, she can respond to what he expressed in a way that is seemingly appropriate to
respond to what he asserted: “No. She can’t be.”
105
Thus, expressing floats works in a similar
way to other standardized indirect speech acts with regard to how a hearer can respond. And
we can see that expressing floats works like our model standardized indirect speech act with
regard to all three of these properties. This is strong evidence in favor of EXPRESSING
FLOATS.
The last feature of standardized speech acts is the theoretical underpinning for the
replacement for The Direct Approach – the problematic theory of at-issue speech acts from
above:
The Direct/Standardized Indirect Approach: An utterance of a ‘might’-sentence gives
rise to the following at-issue speech acts: any direct or standardized indirect speech
act(s) performed by the speaker in making the utterance.
106
In the case of simple ‘might’-sentences, this standardized indirect speech act is an expression
of a float in the prejacent of ‘might’, as EXPRESSING FLOATS says.
4. Applying the Theory to the Data
I’ll close this chapter by showing how the theory I’ve laid out applies to the examples
105
The ability to respond to the indirect speech act in a guise that is appropriate for responding to the direct
speech act may help to explain some controversial data, from the literature:
(False Cookies) Susie and Lil: Phil, we’re interviewing people to get information about who stole
the cookies. Do you know who stole them?
Phil: No, but Angelica might be the thief.
Lil: ? That’s false. She can’t be. I was with her all afternoon.
It is contentious whether Lil’s response here is appropriate. Judgments about this case vary more than in the
cases where Lil says, “No” or “You’re wrong.” (For experimental data on a case where truth and falsity are at-
issue, see Knobe and Yalcin MS.) If “That’s false” is an appropriate response, the above feature of indirect
speech acts can help to explain why. “That’s false” has the form of a response that would be appropriate to
the direct speech act. But, as mentioned above, we can use this form in responding negatively to the indirect
speech act (just as we can say, “No. I can’t.” to respond to the indirect request to close the door). Thanks to
Stephen Finlay for discussion and disagreement here.
106
Other pragmatic factors may make it highly preferable to respond to one speech act rather than another.
Compare (Cookies) with:
(Weird Cookies) Susie and Lil: Phil, we’re interviewing people to get information about who stole
the cookies. Do you know who stole them?
Phil: No, but Angelica might be the thief.
Lil: ?? Right. It is consistent with your information that she is the thief. But I
know that she isn’t.
Lil’s response in (Weird Cookies) is more degraded than her response in (Cookies) due to the fact that the
speech act of primary importance is the indirect expression of a float. So it makes more sense for her to
respond to the expression of the float – as in (Cookies) – than to the assertion – as in (Weird Cookies). This
is compatible with The Direct/Standardized Indirect Approach.
89
of different sorts of uses of ‘might’-sentences. First, consider, yet again, the scenario where Lil
and Susie are investigating the cookie heist. They know Angelica isn’t the thief. And they
think that Phil might have some more information about the stolen cookies. Phil walks in:
(Cookies) Susie and Lil: Phil, we’re interviewing people to get information about
who stole the cookies. Do you know who stole them?
Phil: No, but Angelica might be the thief.
Lil: No. She can’t be. I was with her all afternoon.
As we saw in Chapter One, we cannot explain this conversation with only the resources of
The Direct Approach and Contextualism. But we can explain it with Contextualism,
EXPRESSING FLOATS, and The Direct/Standardized Indirect Approach. Here is how.
Phil is an advisor to Lil and Susie in their inquiry into who is the thief. He wants to help
them succeed by getting them to share his float that Angelica is the thief. To this end he
expresses his float that Angelica is the thief as a standardized indirect speech act (from
EXPRESSING FLOATS). This means, among other things, that Lil can respond to it in a
straightforward manner – she can respond with ‘Yes’ or ‘No’ depending on whether she agrees
or disagrees with Phil (from The Direct/Standardized Indirect Approach). In fact, Lil
disagrees with Phil according to Evaluationism. She is disposed to negatively evaluate his float
as inaccurate. Because of this, it is appropriate for Lil to respond negatively to what Phil
indirectly expressed.
A similar explanation is available for group-centric uses of ‘might’-sentences, like in the
conversation from Chapter One – (Murder). Cases of positive responses, like the conversation
from Chapter Two – (Agree Cookies) – also receive similar explanations, with the interlocutors
agreeing rather than disagreeing. The details of these cases are straightforward, and I will leave
it to the reader to fill them in.
More interesting is that the current approach also has the resources to explain multiple
response cases. These cases were a problem for Contextualism and The Direct Approach
since, according to that combination, there simply aren’t enough at-issue speech acts to make
sense of the multiple appropriate responses. Suppose, again, that Phil and Lil are playing
battleship and that Lil, in fact, has a PT boat, not a battleship, on C4 and knows she does:
(Battleship') Phil: There might be a battleship on C4.
Lil: Yeah. There might be. / No. There can’t be.
Phil both asserts that it is consistent with his information that there is a battleship on C4
(from Contextualism) and expresses his float that there is a battleship on C4 (from
EXPRESSING FLOATS). The first is the direct speech act, and the second is a standardized
90
indirect speech act. By The Direct/Standardized Indirect Approach, both his direct speech act
– his assertion – and his indirect speech act – his expression – are at-issue. So the proposition
that it is consistent with his information that there is a battleship on C4 and his float that there
is a battleship on C4 can be evaluated. Since Lil agrees with Phil about the proposition he
asserts, she can respond positively, and since she disagrees with Phil in virtue of the float, she
can respond negatively. Thus the combination of Contextualism, EXPRESSING FLOATS,
and The Direct/Standardized Indirect Approach can explain multiple response cases.
We might think that much-discussed eavesdropper cases, like those found in Egan,
Hawthorne, and Weatherson 2005 and Egan 2007, require independent treatment. In these
cases the speaker is not concerned with cooperating with the eavesdropper in reasoning. And
so, we might think, the speaker will not indirectly express his float in some proposition. This
would be bad, since we need some explanation of why eavesdroppers can respond in the way
that Lil responds in (Cookies).
Here is a classic eavesdropper case from Egan (2007). Suppose that Bond, when in the
headquarters of his enemies, Blofeld and Number 2, planted a bug and evidence that he is in
Zürich. He then escaped to London to listen to the bug. Blofeld and Number 2 are now
trying to figure out Bond’s whereabouts:
(Eavesdropper) Number 2: Bond might be in Zürich.
Blofeld: Yeah.
Bond: (Eavesdropping from London) No. Number 2 is wrong.
I’m not in Zürich.
We might think that Number 2 cannot be expressing a float because he is not cooperating
with Bond or trying to help him in reasoning.
While it is correct that Number 2 is not cooperating with Bond, it does not follow that
he does not express a float that Bond is in Zürich. Here is the way eavesdropper cases are set
up: The speaker aims to coordinate with his audience (not the eavesdroppers) in deliberation.
He utters a sentence. The eavesdropper overhears and responds negatively to this utterance.
But, given this set-up, it is easy to see that that the speaker is expressing his float. He does
this in order to aid his audience in their reasoning. The eavesdropper can respond to the
expression even if it is not motivated by a desire to aid him in reasoning.
107
In (Eavesdropper) Number 2 expresses his float, as a standardized indirect speech act,
because of his shared goal with Blofeld. Given The Direct/Standardized Indirect Approach,
Number 2’s expression of a float that Bond is in Zürich is an at-issue speech act. Since Bond
107
This is similar to how an eavesdropper can respond negatively to an assertion even if the assertion is not
intended to get the eavesdropper to accept the proposition asserted.
91
disagrees with Number 2 in virtue of this float (according to Evaluationism), he can respond
negatively to Number 2’s utterance.
In general the story about eavesdropper cases goes as follows. There is nothing special
about them. If a speaker indirectly expresses a float and an eavesdropper overhears him, the
eavesdropper can appropriately respond based on whether she agrees or disagrees with the
speaker in virtue of the float that was expressed. If the speaker does not express a float, then
the eavesdropper cannot appropriately respond to any floats. Eavesdropper cases are just
ordinary cases on my account.
In this section I’ve shown how my account is unified – offering straightforward,
plausible explanations of a variety of different ways in which speakers can use and hearers can
respond to uses of simple ‘might’-sentences. In the chapter as a whole, I motivated the idea
that, in addition to asserting an MP, a speaker who utters a ‘might’-sentence often indirectly
expresses a float as a standardized speech act – EXPRESSING FLOATS. I explained why a
speaker would want to do this. And I explained why a hearer could be expected to realize a
speaker has expressed a float. I motivated a view that allows expressions of floats to be at-issue
speech acts – The Direct/Standardized Indirect Approach – based on theoretical
considerations about standardized, indirect speech acts. My view is not only quite natural; it
also helps to explain the sorts of conversations that have been used in objecting to the
conjunction of Contextualism and The Direct Approach. Contextualism together with
EXPRESSING FLOATS and The Direct/Standardized Indirect Approach is, therefore, quite
plausible.
There is still a large and salient obstacle. In Chapter Two, I objected to one view – the
combination of Contextualism, PUTTING IN PLAY, and The Multiple MP Approach –
because its natural extension can’t explain uses of and responses to uses of complex ‘might’-
sentences. But, in this chapter I’ve only dealt with uses of and responses to uses of simple
‘might’-sentences. For my account to be complete – and for the advantage of my view over
that competitor to be apparent – I need to explain communication involving complex ‘might’-
sentences. This task is the topic of the next chapter.
92
Chapter Five: Complex ‘Might’-Sentences
In the last chapter, I presented a theory of communication involving uses of simple
‘might’-sentences – sentences like “Angelica might be the thief.” But I have yet to address
complex ‘might’-sentences – larger sentences in which ‘might’-sentences are embedded. How
are sentences like, “If she has crumbs in her pocket, then Angelica might be the thief”, “Phil
believes Angelica might be the thief”, and “Angelica might have been the thief” used in
communication? In this chapter I give an account of these and other complex ‘might’-
sentences.
My account of uses of and responses to uses of complex ‘might’-sentences is similar to
my account of uses of and responses to uses of simple ‘might’-sentences. A speaker who uses
a complex ‘might’-sentence asserts a proposition. Typically, she also performs a standardized
indirect speech act. For utterances of many complex ‘might’-sentences, a speaker will
indirectly express a float-like mental state. This makes sense because complex ‘might’-
sentences – just like simple ones – are used to coordinate reasoning with uncertainty. And
float-like states are what agents use in reasoning with uncertainty.
The position about the indirect speech acts that speakers typically perform in uttering
complex ‘might’-sentences will allow us to explain some otherwise problematic conversations.
We can use our theory of at-issue speech acts from the last chapter to explain responses to
uses of complex ‘might’-sentences as responses to either the direct or a standardized indirect
speech act performed. Though we only applied it to uses of simple ‘might’-sentences in the
last chapter, the theory of applies equally well to uses of complex ‘might’-sentences. Here it is
again:
The Direct/Standardized Indirect Approach: An utterance of a ‘might’-sentence gives
rise to the following at-issue speech acts: any direct or standardized indirect speech
act(s) performed by the speaker in making the utterance.
In this chapter, I won’t only be concerned with positive and negative responses to uses of
‘might’-sentences. I will also be concerned with explaining how uses of such sentences
function in pieces of reasonable inference. And I will address a problem from Yalcin 2007;
2011 about the infelicity of uses of certain sorts of complex ‘might’-sentences.
1. Methodology and Compositionality
93
Before I address particular types of complex ‘might’-sentences, I am going to give an
overview of the account of embedding. I’ve said that in uttering a complex ‘might’-sentence, a
speaker asserts a proposition and typically performs an indirect speech act. For these acts to
be communicatively successful – so that they are the type of acts that it would make sense for
speakers to perform with these sentences – it must be possible for a hearer to figure out that
the speaker is performing them.
The direct speech act performed by a speaker who uses a ‘might’-sentence (simple or
complex) has a close relationship with the semantic content of the sentence in the context. In
Chapter One, I showed that Contextualism fits nicely with two semantic theories of ‘might’,
and I stayed neutral between the two theories. The proposition asserted by a speaker who
utters a ‘might’-sentence of any complexity can be derived largely by compositional means
with obvious pragmatic features to resolve demonstrative reference or fill in the gaps
(depending on which view is taken). A hearer can figure out the direct speech act performed
by such an utterance from her grasp of the meanings of the words, the rules for composition,
and the objects demonstrated in the context (and, perhaps, the pragmatic principle of
completion used by Bach (1994)).
But, ‘might’-sentences (simple and complex) are also typically used to perform
standardized, indirect speech acts. Since they are standardized, competent speakers will be
able to figure out what speech act was performed. But, we still need an explanation of how
that speech act could have become standardized in the first place. That is, we need an
explanation of how hearers could come to understand that the indirect speech act was being
performed in virtue of knowing what direct speech act was performed and what the context is
like. A hearer could not come to understand the indirect speech act solely from the semantic
values of the terms and compositional principles. Rather, as we saw in the last chapter,
indirect speech acts allow hearers to reason to what is communicated in a way that has a
Gricean flavor – based on the role that uses of ‘might’-sentences play in conversations.
For example, consider the sentence, “If Angelica has crumbs in her pocket, then she
might be the thief.” Such a sentence is useful for a speaker to utter in order to help his
interlocutors in contingency reasoning about the question of who the thief is. And, we saw in
Chapter Three that the sort of attitude that is useful in such contingency reasoning is a
conditional float. So, it makes sense that a speaker who uses this sentence indirectly expresses
a conditional float that Angelica is the thief given that she has crumbs in her pocket. In
94
general, since ‘might’-sentences are used to coordinate reasoning with uncertainty, it makes
sense that speakers use them to express float-like attitudes or convey information about float-
like attitudes.
This point about how hearers could figure out what indirect speech act was performed
extends to multiply complex sentences – like “If Angelica wasn’t at the scene of the crime,
then it’s not the case that she might be the thief” (where a ‘might’-sentence is embedded under
negation and that sentence is embedded in conditional). A speaker who uses this sentence will
typically try to coordinate group reasoning by expressing a conditional sink that Angelica is
the thief given that she wasn’t at the scene of the crime. I won’t further investigate multiply
embedded ‘might’-sentences. But the reader should be able to work out what is indirectly
expressed by following the aforementioned strategy.
In this section, I’ve laid out the strategy for extending the account of ‘might’-
communication that I gave in the last chapter to complex ‘might’-sentences. And I’ve noted
that the indirect speech acts performed by speakers who use these sentences – which are often
the most important speech acts performed – aren’t generated in a compositional manner. But
wouldn’t it be simpler, we might think, to have an account where the semantic values of all of
these sentences correlate with the most important speech acts that speakers perform in using
them? The idea would be to give an expressivist semantic theory – i.e. to assign mental states
to simple ‘might’-sentences as their semantic values and then derive the mental states
expressed by complex ‘might’-sentences (as their semantic values) from the simpler parts
together with compositional principles. There is, on first thought, something attractively
simple about this, but I think that, ultimately, we must give the sort of account that I do in
this chapter for a number of reasons.
First, with respect to at least one theoretical virtue, the expressivist semantic account
comes out behind the traditional contextualist view. This was discussed in Chapter One, so I
will just remind the reader of it here. The dominant accounts of modals are contextualist
ones. Allowing for a contextualist treatment of epistemic uses of modals allows for a unified
treatment. If we, on the other hand, went for an expressivist picture, we would be introducing
a sort of ambiguity across the class of words, including ‘might’, ‘must’, and ‘should’. I take
such an ambiguity to be theoretically undesirable, and we should avoid it if we can.
There is also reason to think that the expressivist semantic account has less of an
advantage over the pragmatic account than we might originally think in the theoretical virtue
95
of systematicity. This is because the pragmatic account does, in fact, have a degree of
systematicity. It is true that each type of embedding must be treated separately. But there are
some general things that we can say about embedding. For instance, as was noted a few
paragraphs back in the discussion of multiply embedded ‘might’-sentences, a speaker who uses
a conditional with a negated ‘might’-sentence in the consequent typically expresses a
conditional sink. Once we see the accounts of conditional and negated ‘might’-sentences (in
§2 and §5, respectively), we will see that combining these two accounts leads us to expect that
a conditional sink is expressed. I do admit that the account is, on the whole, not as systematic
as a purely compositional one. However, this affords it the flexibility to account for the variety
of linguistic data without large complications, as I will now review.
In Chapter One, we saw some data that were quite hard to make sense of on the theory
of direct speech acts that is closely related to an expressivist semantic picture – Expressivism.
These data involved conversations where external bodies of information seemed relevant to
the interpretation of ‘might’ and conversations where multiple sorts of speech acts appeared to
be performed. Such data, while not inconsistent with Expressivism, require a number of
complications, hampering one of the main virtues of the expressivist semantic picture –
simplicity.
Finally, and perhaps most importantly, we saw in Chapter One that giving a
compositional expressivist semantic picture is a difficult, if not impossible, task. In that
chapter, we saw that two prominent attempts – that of Blackburn (1988) and that of Yalcin
(2007; 2011) (in the spirit of Gibbard 1990; 2003) – don’t succeed in giving the right semantic
values to all complex sentences. If, as it seems, there can be no successful compositional
expressivist account, then a non-compositional, pragmatic expressivist account which correctly
explains what is expressed by speakers who use complex ‘might’-sentences is the only game in
town. So, I take the recent discussion to have reminded the reader both that it is doubtful that
we can form a satisfactory compositional expressivist semantics and that, even in the presence
of such a semantics, we might still prefer the combination of contextualism and a non-
compositional expressivist pragmatics for its theoretical conservatism and flexibility in
explaining the conversational data.
Here is a quick overview of what is to come in the rest of this chapter. Since our theses
about communication from the last chapter – Contextualism and EXPRESSING FLOATS –
concern only simple ‘might’-sentences, we don’t yet have an account of the direct or the
96
indirect speech acts performed by speakers who use complex ‘might’-sentences. In this
chapter, I will give these accounts for a number of types of complex ‘might’-sentences. I will
address different sorts of complex ‘might’-sentences, though the cases covered do not exhaust
the ways in which ‘might’ can embed.
108
But I will address the most common cases. The
spirit of my account – that we need to think about how speakers are trying to coordinate
reasoning in particular ways – can be applied to the other ways of embedding ‘might’-
sentences. Furthermore, in this chapter, I will show how the account of the indirect speech
act performed helps to explain uses of and responses to uses of complex ‘might’-sentences in
conversation, as well as (in some cases) how such sentences function in reasonable inferences.
2. Conditional ‘Might’-Sentences
In this section I give an account of what is communicated by utterances of conditional
‘might’-sentences – sentences in which a ‘might’-sentence is embedded in the consequent of a
conditional. Here is a theory – based on Contextualism – of the direct speech act performed
by a speaker who uses a conditional ‘might’-sentence:
ContextualismConditional: The direct speech act performed by a speaker who
assertively utters
┌
If S, then it might be that T
┐
, where p is the content of S and q is
the content of T in the context, is a conditional assertion of the proposition that q is
consistent with the information of the context, conditional on the supposition of p.
This account of conditionals is in line with suppositional accounts like those of Edgington
(1995), DeRose and Grandy (1999), Bennett (2003), and Barnett (2006). It makes clear how a
conditional is used in hypothetical reasoning to affect reasoning on some supposition.
109
In uttering a conditional ‘might’-sentence, a speaker does more than just perform the
direct speech act predicted by ContextualismConditional. This is because conditional assertions
of propositions about what possibilities are consistent with an agent’s information are not
particularly useful in reasoning with uncertainty. There are many, many propositions that are
108
I don’t address the case where a ‘might’-sentence is embedded under conjunction. In uttering a conjunction,
a speaker asserts the content of each conjunct. So this works the same as the unembedded case. A speaker
who utters a conjunctive ‘might’-sentence typically expresses a float in the prejacent of ‘might’.
109
One might reject ContextualismConditional in favor of an account based on the material conditional (Grice
1989a; Jackson 1979) or some other truth-conditional analysis (for instance, those in Stalnaker 1975 or Kratzer
1986). See Kratzer 1986 and Edgington 1995 for arguments against the material account. Other truth-
conditional accounts are more plausible, though they encounter problems dealing with partial beliefs in
conditionals – as shown by Lewis (1976) and discussed by Edgington (1995). Nonetheless, there may be a
version of these latter accounts which is immune to these criticisms and can serve the role I want in my
overall theory of ‘might’-communication. However, I will avoid ultimately settling this vexed issue here and
proceed with the account in the text.
97
consistent with an agent’s information under a supposition. So it makes sense that a speaker
would do something more relevant and useful for contingency reasoning under uncertainty –
express a conditional float:
EXPRESSING CONDITIONAL FLOATS: A speaker who utters a sentence of the
form
┌
If S, then it might be that T
┐
, where p is the content of S and q is the content
of T in the context, typically indirectly expresses, as a standardized speech act, a
conditional float in q given p.
Given that speakers use simple ‘might’-sentences to indirectly express ordinary floats, it is
extremely natural to think that they use conditional ‘might’-sentences to indirectly express
conditional floats.
Consider the familiar scenario of searching for the cookie thief:
(Cond Cookies) Susie and Lil: Phil, we’re interviewing people to get information
about who stole the cookies. Do you know who stole them?
Phil: No, but if Angelica has crumbs in her pocket, then she
might be the thief.
Lil: No. If she has crumbs in her pocket, she can’t be the thief.
The thief does not have crumbs in her pocket.
Phil conditionally asserts that it is consistent with his information that Angelica is the thief, on
the supposition that Angelica has crumbs in her pocket. He also indirectly expresses a
conditional float that Angelica is the thief given that she has crumbs in her pocket.
We can understand why Phil expresses a conditional float in a way similar to the
unembedded case. Phil wants to help coordinate contingency reasoning and does so by
expressing a conditional float. Phil’s expression has all of the features of a standardized,
indirect speech act. Let’s first address the three features of indirectness: First, in (Cond
Cookies), Phil’s expression is the speech act of primary importance. Second, one could use
that sentence without performing the indirect speech act – for instance, by using it exo-
centrically.
110
Finally, a hearer can figure out what indirect speech act Phil performed based
the direct speech act and general features of the context – in a way that has a Gricean flavor.
She could think Phil’s conditional assertion that it is consistent with his information that
110
Here is an example using the familiar battleship background:
(Battleship') Phil: There might be a battleship on C4.
Lil (as an aside to someone else): If he hasn’t figured out yet that it’s a PT boat,
then, yeah, there might be a battleship on C4.
Lil does not express a conditional float here. She merely makes a comment on Phil’s epistemic position.
98
Angelica is the thief on the supposition that she has crumbs in her pocket is not very relevant
to the inquiry. First, it would be consistent with Phil’s information that Angelica is the thief
whether or not she has crumbs in her pocket.
111
Second, knowing that a proposition is
consistent with his information on the supposition that Angelica has crumbs in her pocket is
not very useful, since there are many, many propositions that are consistent with his
information under this supposition (and, even, many, many such propositions about who is
the thief). What is relevant for the inquiry and, so, what is relevant for Phil to express is a
conditional float that Angelica is the thief if she has crumbs in her pocket. Since Phil is
aiming to help the group in reasoning, a hearer can conclude that he expresses this conditional
float.
Phil’s expression also has our three characteristic properties of standardization. First,
the indirect speech act is recoverable without the sort of reasoning I just outlined. Second, the
indirect speech act is detachable from the direct speech act. For example, in circumstances
that are similar to those surrounding (Cond Cookies) but involve Phil instead uttering, “If
Angelica has crumbs in her pocket, then it is consistent with my information that she is the
thief”, Phil does not express a conditional float. He merely conditionally asserts that it is
consistent with his information that Angelica is the thief, on the supposition that she has
crumbs in her pocket. Finally, the indirect speech act is at-issue. We can see this by
recognizing that it allows straightforward replies. For example, as I will show in the next
paragraph, Lil replies in a straightforward manner to Phil’s expression (she can’t be
responding to what he conditionally asserted). So, Phil’s expression has all of the properties of
a standardized indirect speech act.
Just as our account of simple ‘might’-sentences allowed us to explain conversations like
(Cookies), this account of conditional ‘might’-sentences allows us to explain (Cond Cookies).
In (Cond Cookies), Lil can’t be responding to what Phil conditionally asserted – that it is
consistent with his information that Angelica is the thief, given that she has crumbs in her
pocket. For all Lil knows, this is true. Rather, she is responding to what Phil indirectly
expressed – a conditional float in Angelica’s being the thief if she has crumbs in her pocket.
Lil has a conditional sink that Angelica is the thief if she has crumbs in her pocket. We saw
in Chapter Three that Lil disagrees with Phil in this situation. She fulfills the condition in
Evaluationism – being disposed to negatively evaluate his conditional float in virtue of having
111
For a detailed discussion of this problem about the interaction of ‘might’ and conditionals, see Lennertz
forthcoming b.
99
her conditional sink. Since Phil indirectly expressed his conditional float as a standardized
indirect speech act (and given The Direct/Standardized Indirect Approach), Lil can respond
to it by saying “No”.
This account of conditional ‘might’-sentences also allows us to explain why some
inferences are compelling:
(4) If Angelica has crumbs in her pocket, then she might be the thief.
(5) Angelica has crumbs in her pocket.
(6) So, she might be the thief.
Suppose Phil and Lil have the following conversation:
(Inf Cookies) Phil: If Angelica has crumbs in her pocket, then she might be the thief.
Lil: Well, Angelica has crumbs in her pocket. So, she might be the thief.
This is a reasonable conclusion for Lil to draw. And it seems that she is doing more than
saying that it is consistent with Phil’s or their shared information that Angelica is the thief
when she says, “she might be the thief.” Indeed, as is typical with utterances of simple
‘might’-sentences, she is expressing a float that Angelica is the thief. But, we need some
explanation of how she rationally arrives at that float from what she gets from Phil’s utterance
together with her knowledge that Angelica has crumbs in her pocket. We can’t get this
explanation solely based on the logical properties of the contents of the direct speech acts that
the speakers perform, since these do not involve floats.
Rather, we can explain Lil’s inference in (Inf Cookies) in terms of the standardized,
indirect expressions of float-like attitudes. In making his utterance, Phil indirectly expresses,
in a standardized way, a conditional float that Angelica is the thief given that she has crumbs
in her pocket. Lil comes to agree with Phil – i.e. comes to have that conditional float.
Additionally, Lil knows that Angelica has crumbs in her pocket. As we saw in Chapter Three,
being in these two states rationally requires Lil to float that Angelica is the thief by the
principle that I called Conditional Detachment (which says that, for any inquiry, if an agent,
in that inquiry, conditionally floats p if q and accepts q, then she should, in that inquiry, float
p). The attitude of floating that Lil is the thief is standardly expressed by a speaker who uses
“Angelica might be the thief.” So, the inference that Lil makes to have this float is compelling
in virtue of the rational commitments she incurs by agreeing with what is communicated
through standardized, indirect speech acts performed in uttering the premises.
In this section I extended the story about simple ‘might’-sentences to conditional
100
‘might’-sentences. We saw that this account can help explain conversations involving uses of
and responses to uses of conditional ‘might’-sentences, as well as uses of conditional ‘might’-
sentences in inferences.
3. Disjunctive ‘Might’-Sentences
A disjunctive ‘might’-sentence – a disjunction which contains a ‘might’-sentence as a
disjunct – is typically used to assert a proposition and to express a float-like attitude. In this
section, I will develop a theory according to which a different float-like attitude is expressed
depending on the sort of reading the disjunction gets. In particular, if the disjunctive ‘might’-
sentence gets an inclusive reading, then the attitude typically expressed is an ordinary float in
the prejacent of ‘might’. And if the disjunctive ‘might’-sentence gets an exclusive reading, then
the attitude typically expressed is a conditional float. Like in the case of conditionals, this
allows us to explain uses of and responses to uses of ‘might’-sentences, as well as how uses of
such sentences function in inferences. The difference in what attitude is expressed should not
be surprising, since differences in the propositions conveyed result in differences in what a
reasonable hearer can infer about what else the speaker communicated.
Let’s start by giving an account of the direct speech act performed by a speaker who
uses a disjunctive ‘might’-sentence:
ContextualismDisjunction: The direct speech act performed by a speaker who
assertively utters
┌
S or it might be that T
┐
, where p is the content of S and q is the
content of T in the context, is an assertion of the proposition that p is the case or q is
consistent with the information of the context.
It is plausible that speakers also indirectly express float-like states when uttering disjunctive
‘might’-sentences.
As I said, which indirect speech act is performed by a speaker who uses a disjunctive
‘might’-sentence depends on whether the disjunction gets an inclusive or exclusive reading.
On an inclusive reading of a disjunction, it is allowed that both of the disjunctions are true.
On an exclusive reading, it is assumed that one or the other of the disjuncts is true, but not
both. I must be clear that I am not assuming that there is a sort of semantic ambiguity in ‘or’
that determines these two readings. Rather, as Grice (1989b) taught us, it is a matter of
pragmatics. Nonetheless, the sort of reading that a ‘might’-disjunction gets (even as a
pragmatic output) affects what indirect speech act it is reasonable to conclude that a speaker
101
performs in uttering it.
I’m going to begin with the inclusive disjunction case (and return to the exclusive case).
A common way to get an inclusive reading is to use a sentence in which both disjuncts are
‘might’-sentences. In uttering “Angelica might be the thief or Chuckie might be the thief”, a
speaker appears to express two floats – one that Angelica is the thief and one that Chuckie is.
In such cases, it looks like ‘or’ is functioning like conjunction. As a number of authors have
noted, an utterance of “Angelica might be the thief or Chuckie might be the thief” puts one in
a position to utter “Angelica might be the thief” (and “Chuckie might be the thief”).
112
There is a good reason why this is so. When one utters a disjunction one conveys that
it is compatible with what one knows that the first disjunct is true and it’s compatible with
what one knows that the second disjunct is true. So when one utters the above disjunction,
one conveys that it’s compatible with what one knows that Angelica might be the thief and it’s
compatible with what one knows that Chuckie might be the thief. If these double epistemics
collapse, then we get the reading that Angelica might be the thief and Chuckie might be the
thief. It’s plausible that they do collapse because each looks at what is consistent with the
speaker’s information (remember, this is a speaker-centric use). If the speaker is in a position
to convey that it is consistent with his information that it is consistent with his information
that Angelica is the thief, then it seems that he is in a position to convey that it is consistent
with his information that Angelica is the thief.
113
So it makes sense that both floats are
expressed. More generally:
EXPRESSING FLOATSDISJUNCTION: A speaker who utters a sentence of the form
┌
S
or it might be that T
┐
, where p is the content of S and q is the content of T in the
context, and where the disjunction receives an inclusive reading, typically indirectly
expresses, as a standardized speech act, a float in q.
114
The expression of a float by a speaker who uses a disjunctive ‘might’-sentence with an
inclusive reading is a standardized indirect speech act. I gave the sort of reasoning a hearer
could go through to figure out that such a speech act was performed before stating
EXPRESSING FLOATSDISJUNCTION. The expression also has the other two features of
112
See Zimmermann 2000, Geurts 2005, and von Fintel and Gillies 2008; 2011.
113
We might doubt that a collapse occurs in cases of mixed information. For example, consider DeRose’s
(1991) case where a patient takes a test that rules out having a disease if it comes up negative but issues false
positives. Waiting for the results, which the doctors have already seen, the patient says, “I don’t know the
results of the test. It’s possible that I might have the disease. Only the doctors know for sure whether I
might have it.”
114
Thanks to Mark Schroeder for discussion.
102
indirectness – often being the speech act of primary importance and not being performed by
every use of that sentence. Additionally this expression also has the features of
standardization. A hearer can grasp that the indirect expression was performed without
calculation. The act of expression is detachable – for instance, it is not performed by an
utterance of “It is consistent with my information that Angelica is the thief or it is consistent
with my information that Chuckie is the thief.” Finally, the expression is at-issue. A hearer
could respond to an utterance of “Angelica might be the thief or Chuckie might be the thief”
by saying “No” if she knew, for instance, that Angelica isn’t the thief.
We’ve seen that ‘might’-disjunctions, when they get an inclusive reading, are typically
used to express ordinary floats. But there are many cases where a disjunctive ‘might’-sentence
gets an exclusive reading. In such a case, a speaker typically expresses a conditional float in
the prejacent of ‘might’ given the negation of the other disjunct. For example, assume the
standard background from (Cookies):
(Dis Cookies) Susie and Lil: Phil, we’re interviewing people to get information about
who stole the cookies. Do you know who stole them?
Phil: No, but either Angelica has a good alibi or she might be the thief.
Lil: No. Even if she doesn’t have a good alibi, she can’t be the thief.
Phil asserts that Angelica has a good alibi or it is consistent with his information that Angelica
is the thief. We might think that we can get the same sort of reasoning as the inclusive case to
kick in here – leading to the conclusion that Phil expresses an ordinary float that Angelica is
the thief. But this isn’t so. The above reasoning is canceled since, in making his utterance,
Phil also conveys that its not the case that both of the disjuncts are true.
115
It is reasonable that
this fact blocks the inference to the conclusion that he floats that Angelica is the thief.
But, we can still infer that Phil expressed some float-like attitude. This is because it
follows from what he said that he thinks that on the supposition that Angelica does not have a
good alibi, it is consistent with his information that Angelica is the thief. But this puts us in
exactly the same position that we were in when considering conditionals. So it makes sense
that Phil indirectly expresses a conditional float that Angelica is the thief if she doesn’t have a
good alibi. In general:
EXPRESSING CONDITIONAL FLOATSDISJUNCTION: A speaker who utters a
sentence of the form
┌
S or it might be that T
┐
, where p is the content of S and q is the
115
’Either’ is a strong marker of a disjunction getting an exclusive reading.
103
content of T in the context, and where the disjunction receives an exclusive reading,
typically indirectly expresses, as a standardized speech act, a conditional float in q given
p.
This expression has all the features of a standardized, indirect speech act. Since this can be
shown in a way that mirrors the conditional case, I will leave it to the reader to carry out the
exact details.
Given this account, we can explain (Dis Cookies). Phil indirectly expresses, as a
standardized speech act, a conditional float that Angelica is the thief, if she doesn’t have a
good alibi. Lil, on the other hand, has a conditional sink that Angelica is the thief given that
she doesn’t have a good alibi. Lil is disposed to negatively evaluate Phil’s conditional float, so,
by Evaluationism, she disagrees with him. And, since he expressed his conditional float as a
standardized indirect speech act, she can, according to The Direct/Standardized Indirect
Approach, reply negatively.
We can also use the account just given to explain why the following inference is
compelling:
(7) Either Angelica has a good alibi or she might be the thief.
(8) Angelica does not have a good alibi.
(9) So, she might be the thief.
A speaker who uses “Either Angelica has a good alibi or she might be the thief” standardly
indirectly expresses a conditional float that Angelica is the thief given that she does not have a
good alibi. This allows us to explain the inference to the float communicated by a speaker
who utters (9) in the same way we did in the last section. Conditionally floating that Angelica
is the thief given that she does not have a good alibi – the state expressed by a use of (7) – and
accepting that she does not have a good alibi – as evidenced by a use of (8) – rationally require
– by Conditional Detachment – one to float that Angelica is the thief – the state standardly
expressed by a use of (9).
In this section we’ve seen the two ways that disjunctive ‘might’-sentences are used to
coordinate reasoning with uncertainty. When a disjunctive ‘might’-sentence is used
inclusively, a speaker typically expresses a float in the prejacent of ‘might’. And when it gets
an exclusive reading, a disjunctive ‘might’-sentence is used to express a conditional float. This
explains the appropriateness of some conversations in which such sentences are used and the
compellingness of some inferences in which they function as premises. All of these
104
phenomena can be explained given the pragmatic facts about when it makes sense to utter
such disjunctions.
4. Quantificational ‘Might’-Sentences
In this section, I’ll give my account of sentences in which ‘might’ is embedded under a
quantifier.
116
Since ‘might’-sentences are generally used to help coordinate reasoning in
situations of uncertainty, it makes sense that quantificational ‘might’-sentences would be used
to express a float-like state – a quantificational float – in addition to being used to assert some
proposition.
Let’s start with an account of the direct speech act performed by an assertive use of a
quantificational ‘might’-sentence:
ContextualismQuantifier: The direct speech act performed by a speaker who
assertively utters a sentence of the form
┌
Q of the Gs might be Fs
┐
, where a is the
content of the quantifier expression Q, B is the content of F, and C is the content of G
in the context, is an assertion of the proposition that λy(it is consistent with the
information of the context that B applies to y) is a-ly instantiated among the Cs of the
domain of the context.
This deserves some clarification, as it looks more complex than the preceding accounts. The
idea behind this approach to quantifiers is that they predicate higher-order properties of
lower-order properties. These are properties of being instantiated in some way. For instance,
“Everything is extended” contains a quantifier expression, “Everything”. Its content is a
higher-order property that is predicated of the property being extended. And its content
predicates universal instantiation of the property of which it is predicated. So “Everything is
extended” is used to assert that being extended is universally instantiated. Likewise,
“Something is blue” is used to assert that being blue is existentially instantiated (i.e. is
instantiated by at least one thing). The other twist in ContextualismQuantifier is the use of
lambda abstraction. This is just a way of talking that allows us to refer to the property that
applies to an object just in case it is consistent with the information of the context that a
different property applies to that object. When we put these together, we see that in using a
116
Von Fintel and Iatridou (2003) argue that a quantifier can’t take scope over an epistemic modal – a thesis
they call the Epistemic Containment Principle, or ECP. But, as Swanson (2010) shows, the ECP is false. I
won’t go into the details of the final characterization of the ECP – which actually is a principle restricting
certain sorts of quantifier movement – nor will I speculate here on why von Fintel and Iatridou’s original
examples seem to support the ECP.
105
quantificational ‘might’-sentence, a speaker asserts a proposition which ascribes a higher-order
property, determined by the quantifier, of a lower-order one, which is represented by the
lambda abstract.
Let’s consider an example. Lil and Susie are investigating the cookie heist. They call
in Phil to get more information. They all know there is exactly one thief:
(Quant Cookies) Susie and Lil: Phil, we’re interviewing people to get information
about who stole the cookies. Do you know who stole them?
Phil: No. But each of the suspects reported in last night’s
newscast might be the thief.
Lil: No. One of them can’t be. The new evidence rules it out.
Consider the sentence that Phil utters: “Each of the suspects reported in last night’s newscast
might be the thief.” We need to first check the relative scope of the modal and the quantifier.
Phil would be saying something contradictory if we thought this were a case where the modal
scoped over the quantifier – that he asserts that it is consistent with his information that each
of the suspects reported in last night’s newscast is the (unique) thief. So, the quantifier must
take scope outside of the modal in this example.
117
According to ContextualismQuantifier, Phil
asserts that λx(it is consistent with his information that x is the thief) is universally instantiated
among the domain of suspects reported in last night’s newscast – or, more colloquially, for
each suspect reported in last night’s newscast, it is consistent with Phil’s information that that
person is the thief.
Agents also use quantificational ‘might’-sentences to express quantificational floats:
EXPRESSING QUANTIFICATIONAL FLOATS: A speaker who utters a sentence
of the form
┌
Q of the Gs might be Fs
┐
, where a is the content of the quantifier
expression Q, B is the content of F, and C is the content of G in the context, typically
indirectly expresses, as a standardized speech act, a quantificational float, for a of the
Cs, that it is B.
In (Quant Cookies) Phil indirectly expresses a quantificational float, for each suspect reported
in last night’s newscast, that she is the thief. Phil’s indirect expression in (Quant Cookies) is
motivated by a desire to aid in reasoning. He aims to get his audience to share his
quantificational float. Phil’s expression has the features of a standardized indirect speech act.
It is the speech act of primary importance, though the sentence could be used without
117
So this a case that witnesses the failure of Epistemic Containment Principle. See note 116.
106
performing the speech act (I’ll leave it to the reader to construct an exo-centric case that
witnesses this claim). And Lil can work out what Phil expressed in the standard Gricean
manner – in a way similar to the unembedded and conditional cases. Again, what Phil
asserted – that for each suspect reported in last night’s newscast, it is consistent with his
information that that person is the thief – is not very helpful in the reasoning. There are
many propositions that are consistent with his information – including many about people
other than those reported in last night’s newscast being the thief. But it would be very helpful
for him to express a quantificational float, for each suspect reported in last night’s newscast,
that she is the thief. So, a hearer can, on the assumption of cooperativeness, conclude that he
has expressed this quantificational float. However, as in the other cases, hearers can
understand that this speech act has been performed without going through this reasoning.
Furthermore, the speech act is detachable (as can be shown, just like in the other cases, by
comparing with the explicit consistency sentence). Finally, the expression is at-issue. We can
see this by noting that Lil straightforwardly responds to Phil’s expression in (Quant Cookies).
We can explain Lil’s response in (Quant Cookies) as a response to the quantificational
float that Phil indirectly expresses (as usual, we cannot make sense of Lil’s negative response
in terms of her denying what Phil asserted). According to The Direct/Standardized Indirect
Approach, Lil can respond to Phil’s standardized, indirect speech act. And she replies
negatively because she disagrees with Phil, as predicted by Evaluationism. She is disposed to
negatively evaluate his quantificational float.
The idea that quantificational ‘might’-sentences are used to standardly express
quantificational floats allows us to explain why the following inference is compelling:
(10) Each of the suspects reported in last night’s newscast might be the thief.
(11) Angelica is a suspect reported in last night’s newscast.
(12) So, Angelica might be the thief.
(10) is used to express a quantificational float, for each suspect reported in last night’s
newscast, that she is the thief. Being in this state and accepting that Angelica is a suspect
reported in last night’s newscast – as evidenced by (11) – rationally require – by Universal
Quantificational Detachment – one to float that Angelica is the thief – the state indirectly
expressed by a typical use of (12).
So, we have a satisfying account of what goes on in cases where a ‘might’-sentence is
embedded under a quantifier. In uttering such sentences, a speaker typically, in addition to
107
performing the direct speech act, indirectly expresses a quantificational float. I’ve dealt here
solely with the case of universal quantification. But the account works in the same way for
other strengths of quantification. The only difference is that we don’t get the same compelling
inferences. But this is exactly as it should be since, as we saw in Chapter Three, Universal
Quantificational Detachment only holds for universal quantificational floats and sinks.
5. Negated ‘Might’-Sentences
A negated ‘might’-sentence is typically used to perform multiple speech acts: asserting
that the prejacent is not consistent with the speaker’s information and expressing a sink in the
prejacent. The former is the direct speech act; the latter is indirect. It makes sense that the
latter, as well as the former, is performed since, when we think something can’t be the case, we
rule it out as an option in our reasoning.
118
Throughout this section, I use the colloquial way
of negating the epistemic ‘might’, ‘can’t’, rather than the obscure locution, ‘it’s not the case that
it might . . .” I still call sentences containing an epistemic use of ‘can’t’, negated ‘might’-
sentences.
119
Here’s the account of the direct speech act:
ContextualismNegation: The direct speech act performed by a speaker who assertively
utters
┌
It can’t be that S
┐
, where p is the content of S in the context, is an assertion of
the proposition that p is not consistent with the information of the context.
Additionally, a speaker who assertively utters a negated ‘might’-sentence typically performs an
indirect speech act as well:
EXPRESSING SINKS: A speaker who utters a sentence of the form
┌
It can’t be that
S
┐
, where p is the content of S in the context, typically indirectly expresses, as a
standardized speech act, a sink in p.
It makes sense that a speaker does this as ‘might’-sentences are used to coordinate group
reasoning with uncertainty and, so, are used to express float-like mental states. When a
speaker uses a negated ‘might’-sentence, she attempts to get the group to rule out the prejacent
of ‘might’ as an option in their reasoning.
118
Compare with Toulmin 1958 (29): “But the common implication of all of these statements, marked by the
use of the word ‘cannot’, should be clear by now. In each case, the proposition serves in part as an
injunction to rule out something-or-other – to dismiss from consideration any course of action involving this
something-or-other”.
119
The reader can easily verify that in uses of the epistemic ‘can’t’, the negation scopes outside the modal.
Unlike in ‘can’t’ and ‘cannot’, the negation in ‘might not’ scopes inside the modal, conveying the possibility
of the negation of what follows – not the negation of the possibility.
108
Let’s consider an example. Suppose that our characters are, again, trying to figure out
who stole the cookies. But this time Phil has incorrectly ruled out that Angelica is the thief on
the basis of misleading evidence. Lil knows that the evidence that Angelica is the thief is
misleading. They have the following conversation:
(Not Cookies)Susie and Lil: Phil, we’re interviewing people to get information about
who stole the cookies. Do you know who stole them?
Phil: No, but Angelica can’t be the thief.
Lil: No. She might be. The evidence that she isn’t is misleading.
Phil asserts, in line with ContextualismNegation, that it is not consistent with his information
that Angelica is the thief. He also, in line with EXPRESSING SINKS, indirectly expresses a
sink that Angelica is the thief.
Phil’s expression has the properties of a standardized, indirect speech act. The
expression of his sink is the speech act of primary importance. It can be used without
expressing a sink (as usual, think of an exo-centric case). And a hearer could figure out that
Phil expresses his sink based on knowledge of what he asserted and the context, as well as the
assumption that he is cooperative (more on this below). However, as usual, a hearer need not
actually go through the reasoning, since the expression is standardized. The expression of a
sink is detachable from the direct speech act; it isn’t performed by a speaker who utters, “It is
not consistent with my information that Angelica is the thief.” And Phil’s expression is an at-
issue speech act – Lil can give a straightforward response. So, since Lil negatively evaluates
Phil’s sink based on evidential considerations (and given The Direct/Standardized Indirect
Approach), she can respond negatively to Phil’s utterance.
Lil can figure out that Phil expresses his sink based on her understanding of the direct
speech act he performs together with her knowledge of the context in which Phil makes his
utterance. But this is by different means than she could derive that he expresses a float in
(Cookies). There are two kinds of reasoning going on in epistemic modal communication.
One was presented in Chapter Four. When Phil utters, “Angelica might be the thief”, the
proposition asserted isn’t very relevant because there are many propositions that are consistent
with Phil’s information. We need to know more than this if his contribution is to be
cooperative. There are similar relevance concerns at play in the conditional, disjunctive, and
quantificational cases. But this is not so in the negated case. Once we know that a
proposition is inconsistent with Phil’s information, we know – if we trust him and his
109
information – not to use that proposition in reasoning.
So a different way of reasoning is present here – one that is extended to necessity and
probability modals in the next chapter. Lil can reason in the following way. She knows that
it’s not always transparent to a subject what is or is not consistent with his information. So, he
may take something as an option even if it is not consistent with his information.
120
Absent
his utterance, it’s possible that it is not consistent with Phil’s information that Angelica is the
thief, but that he does not realize this and takes it is an option that she is (or just does not
consider the proposition in reasoning). However, his utterance shows that this is not the case
– it shows that he does realize that it is not consistent with his information that Angelica is the
thief. If he is rational, he will rule out Angelica’s being the thief as an option – i.e. he will sink
it. Furthermore, Lil knows that Phil wouldn’t utter, “Angelica can’t be the thief” unless he
thought that Angelica’s being the thief is taken as an option by someone else in the group.
She knows that such an utterance is used to dispel the thought that that proposition is an
option. So she can conclude that Phil aims to express his sink that Angelica is the thief.
There are two aspects to this sort of derivation. I explain them in terms of the
negation case. But, they will apply straightforwardly to the case of ‘must’ and, with some
changes, to talk about probability in the next chapter. The first aspect depends on how the
relations of consistency among the propositions that make up our information are not always
transparent to us. So, we could have some relation of consistency or inconsistency hold
without knowing it. However, by uttering a negated ‘might’-sentence, we show that we take
ourselves to know that the prejacent is not consistent with our information. Since we take
ourselves to know this and point it out, it is reasonable to infer that we rule out the prejacent
as an option in reasoning. The second aspect of this derivation is the following: We don’t
utter negated ‘might’-sentences unless we think that the prejacent is being taken as an option
by someone in the group. Such an utterance is used to dispel the thought that that
proposition is an option.
121
Above, I explained why some modus ponens-like inferences involving sentences that
are used to express conditional and ordinary floats and sinks were compelling. We can see
that the same is not true for the following modus tollens-like inferences. Consider:
120
This is a lesson from Hacking’s (1967) shipwreck case and von Fintel and Gillies’s (2011) Schmolmes case.
121
Thanks to Scott Soames for discussion of this second aspect. Huemer (2007, 131) says, “When S dismisses
[sinks] P, S will typically not consider or discuss P unless a third party brings it up, at which point S will
reject it”. Really, a third party need not bring P up, but S at least needs to think the third party is taking P as
an option.
110
(13) If Angelica has crumbs in her pocket, then she might be the thief.
(14) She can’t be the thief.
(15) So, Angelica doesn’t have crumbs in her pocket.
And
(16) If Angelica wasn’t at the scene of the crime, then she can’t be the thief.
(17) Angelica might be the thief.
(18) So, she was at the scene of the crime.
Neither inference is compelling – for reasons discussed in Chapter Three. (13) and (14) are
used to express attitudes that are already irrational to have jointly – a conditional float that
Angelica is the thief given that she has crumbs in her pocket and a sink that she is the thief.
Once one has the sink expressed by a speaker who uses (14) one should give up the
conditional float expressed by a speaker who uses (13). As for the second example, one can
clearly have the conditional sink expressed by a speaker who uses (16) and the float expressed
by a speaker who uses (17) without accepting (18). Similar points apply to disjunctive ‘might’-
sentences – as these sentences are used to express conditional floats and sinks – and
quantificational ‘might’-sentences.
So, in this section we’ve seen that, in uttering a negated ‘might’-sentence, a speaker
asserts that the prejacent is not consistent with his information and typically indirectly
expresses a sink in the prejacent. I showed that the expression performed in uttering such a
sentence is a standardized, indirect speech act – though the derivation is somewhat different
from the case of utterances of simple ‘might’-sentences.
6. Attitude Verbs
‘Might’-sentences can embed under attitude verbs of various sorts. I am going to start
this section by considering the case of embedding ‘might’ in belief reports. Then I will discuss
some other attitude verbs. As we’ve seen, ‘might’-sentences are typically used to coordinate
reasoning in situations of uncertainty. Sentences where ‘might’ is embedded under an attitude
verb are no different. In these cases speakers typically perform a standardized, indirect speech
act. However, unlike the cases we’ve seen so far, these sentences are not used to express a
float-like state. But they do still have a connection to float-like states.
We can start by stating the account of the direct speech act performed by a speaker
who uses a ‘might’-belief report:
111
ContextualismBelief: The direct speech act performed by a speaker who assertively
utters
┌
A believes that it might be that S
┐
, where p is the content of S in the context
and o is the person denoted by A, is an assertion of the proposition that o believes that
p is consistent with the information of the context.
Usually, the information of the context will be that of the subject of the belief report (this is in
contrast to most other ‘might’-sentences, in which it is typically the speaker’s information).
122

As with the other sorts of ‘might’-sentences we’ve seen, ‘might’-belief reports are also used to
perform an indirect speech act.
We can see what indirect speech act is performed by revisiting the following example:
(Belief Cookies) Susie and Lil: Betty, we’re interviewing people to get information
about who stole the cookies. Do you know who stole them?
Betty: No, but Phil believes Angelica might be the thief.
Lil: He’s wrong in believing Angelica might be the thief. She can’t
be. I was with her all afternoon.
According to ContextualismBelief, Betty asserts that Phil believes that it is consistent with his
information that Angelica is the thief. What else does Betty communicate with her utterance?
Well, Betty utters this sentence about Phil rather than just saying, “Angelica might be the
thief” because she doesn’t float that Angelica is the thief and doesn’t intend to express such a
float. She merely wants to communicate that Phil has this float and let Susie and Lil decide
for themselves whether or not to adopt it as they go forward in their reasoning. So, she
conveys that Phil has this float. And she does so as a standardized indirect speech act. In
general:
CONVEYING FLOATS: A speaker who utters a sentence of the form
┌
A believes
that it might be that S
┐
, where p is the content of S in the context and o is the person
denoted by A, typically indirectly conveys, as a standardized speech act, that o floats p.
In our example, Betty conveys that Phil has his float as a standardized indirect speech
act. It is the speech act of primary importance, though not every use of this sentence is a
performance of that speech act.
123
And a hearer can calculate that this indirect speech act is
122
Given the framework that I am using, I need to allow the context to change in the middle of the utterance of
a sentence. This is to account for uses of sentences like “Phil might believe that Angelica might be the thief.”
The first ‘might’ will typically be relative to the speaker’s information while the second will typically be
relative to Phil’s.
123
It is more difficult to get an exo-centric use under an attitude verb (Stephenson 2007). It is easier to see this
second point by thinking of a use of ‘might’ that is stressed and in a higher register: “Phil believes Angelica
might be the thief.” A speaker who uses this sentence in this way will generally convey only that Phil believes
that it is consistent with his information that Angelica is the thief (without conveying that Phil floats that she
112
performed in a way that has a Gricean flavor. Conveying that Phil floats that Angelica is the
thief is far more relevant than what Betty asserts – that Phil believes that it is consistent with
his information that Angelica is the thief. And Betty could expect Lil to infer that she intends
to convey this. Lil knows that a belief that a proposition is consistent with one’s information
is not one that people go out of their way to form without reason. Usually this belief comes
from wondering whether that proposition is true in the hope of taking it as an option in
reasoning. So, since it is reported that Phil found the proposition to be consistent with his
information and believes it, Lil can conclude he floats it.
124
Note that, for the reasons given in
the last paragraph, Betty conveys that Phil has this float and does not express her own float.
This indirect act of conveying is also standardized. Hearers need not actually go
through the above reasoning. The indirect speech act is detachable from the direct speech act
(think of “Phil believes that it is consistent with his information that Angelica is the thief”).
And, as shown in our example, Lil can straightforwardly respond to Betty’s indirect speech act
(saying Phil is wrong), showing that Betty’s expression is at-issue.
This account also allows us to explain the compellingness the following sort of
inference:
(19) Phil believes Angelica might be the thief.
(20) Angelica might be the thief.
(21) So, what Phil believes is true.
By uttering (19), a speaker typically indirectly conveys that Phil floats that Angelica is the thief.
By uttering (20), a speaker typically indirectly expresses a float that Angelica is the thief.
Someone who accepts that Phil floats that Angelica is the thief and also floats that Angelica is
the thief is rationally committed to thinking that Phil is right in having this float (by
Evaluationism Agree). That person can voice her agreement by saying that Phil’s belief is true
because of the special property of standardized indirect speech acts – that one can respond to
a standardized indirect speech act in a manner that seems appropriate for responding to the
direct speech act.
In general, when a speaker utters a ‘might’-belief report, he performs the direct speech
act described by ContextualismBelief and the indirect one described by CONVEYING
FLOATS. This applies not only to belief reports, but also to thought and indirect speech
reports. The reader is invited to check this by amending (Belief Cookies) accordingly. In
is).
124
Thanks to Scott Soames for discussion.
113
what follows I’m going to discuss how ‘might’ embeds under some other attitude verbs.
125
‘Might’-knowledge reports should receive a similar treatment to belief, with the
appropriate changes. In the case of the direct speech act, such sentences are used to assert a
proposition about what the subject knows (rather than just believes) is consistent with his
information:
ContextualismKnowledge: The direct speech act performed by a speaker who
assertively utters
┌
A knows that it might be that S
┐
, where p is the content of S in the
context and o is the person denoted by A, is an assertion of the proposition that o
knows that p is consistent with the information of the context.
We make the appropriate change from ContextualismBelief to ContextualismKnowledge by
changing the occurrence of ‘believes’ in ContextualismBelief to an occurrence of ‘knows’.
But things are not so simple for the speech acts involving floats. The word ‘believes’
does not occur in the account of the indirect speech act, CONVEYING FLOATS. So, there
is no easy amendment. Perhaps we could say that the same indirect speech act is performed
by ‘might’-belief and ‘might’-knowledge reports:
CONVEYING FLOATSKNOWLEDGE: A speaker who utters a sentence of the form
┌
A
knows that it might be that S
┐
, where p is the content of S in the context and o is the
person denoted by A, typically indirectly conveys, as a standardized speech act, that o
floats p.
But this cannot be the whole story. Consider the scenario surrounding (Belief Cookies):
(Knows Cookies) Susie and Lil: Betty, we’re interviewing people to get information
about who stole the cookies. Does Phil know who stole them?
Betty: No. But he knows that Angelica might be the thief.
Lil: # Yeah. But he’s wrong. She can’t be. I was with her all
afternoon.
Lil’s reply is inappropriate. But ContextualismKnowledge and CONVEYING
FLOATSKNOWLEDGE offer no explanation of this. According to those theories, Lil should be
able to respond as she does in (Knows Cookies). She should be able to say “Yeah” because
she agrees that Phil knows that it is consistent with his information that Angelica is the thief
125
I want to just mention cases in which a ‘might’-sentence that is embedded under a connective or quantifier is
then embedded under ‘believes’ or ‘thinks’. In such cases a speaker typically indirectly conveys that the
subject of the attitude report has the mental state that is typically indirectly expressed by a speaker who uses
the embedded ‘might’-sentence. For example, “Phil believes that if Angelica has crumbs in her pocket, then
she might be the thief” is typically used to indirectly convey that Phil conditionally floats that Angelica is the
thief if she has crumbs in her pocket.
114
and she agrees that he floats that Angelica is the thief (the two propositions that are conveyed
according to ContextualismKnowledge and CONVEYING FLOATSKNOWLEDGE, respectively).
And she should be able to say “He’s wrong” because she disagrees with Phil in virtue of her
sinking and his floating that Angelica is the thief. So, according to this combination of theses,
Lil’s response should be completely appropriate. Thus, this combination of theses can’t be
the whole story.
The problem, as we suspected, is that CONVEYING FLOATSKNOWLEDGE fails to
mark off what is special about knowledge ascriptions. In particular, knowledge ascriptions
have a presuppositional element. To ascribe knowledge, one must agree with the putative
knower – a knowledge ascription presupposes such agreement. For example, for my utterance
of “John knows that the Dodgers won last night” to be felicitous, all conversational
participants must agree that the Dodgers won last night (or be willing to accommodate this
fact). And to respond positively to a knowledge ascription (as Lil does), one must not just
think that the subject of the report has the attitude in question but one must have that attitude
as well. This applies not just to the direct speech act performed but also to the standardized,
indirect speech act. In order for Lil to respond positively to the ‘might’-knowledge ascription,
she must think that Phil floats that Angelica is the thief and she must, herself, float that
Angelica is the thief. This is because, in making her utterance, Betty not only asserts the
proposition that Phil knows that it is compatible with his information that Angelica is the thief
and indirectly conveys that Phil floats that Angelica is the thief. She also performs what we
can call a presuppositional expression of a float that Angelica is the thief. Because of this, for
the utterance to be acceptable, all parties must have the float. The reason that Lil’s response is
inappropriate is that by agreeing with the knowledge ascription (saying “Yeah”), she
presuppositionally expresses a float that Angelica is the thief and, by saying that Phil is wrong,
she communicates that she sinks this proposition. Such a combination of states is, as we saw
from the norm Singlemindedness in Chapter Three, irrational to have. The following is a
general statement of the improved version of CONVEYING FLOATSKNOWLEDGE:
CONVEYING + EXPRESSING FLOATSKNOWLEDGE: A speaker who utters a
sentence of the form
┌
A knows that it might be that S
┐
, where p is the content of S in
the context and o is the person denoted by A, typically indirectly conveys, as a
standardized speech act, that o floats p and typically presuppositionally expresses a
float in p.
115
This notion of presuppositional expression sounds like jargon, but it is natural once we
cash it out. Consider ascriptions of knowledge of non-modal claims like, “John knows that
the Dodgers won last night.” It is common to think that in order for an utterance of this
sentence to be appropriate, all parties to the conversation must accept or be willing to come to
accept that the Dodgers won last night. Because of this, we can say that a speaker who utters
this sentence conveys the proposition that the Dodgers won as a presupposition. We might
even say that the speaker presuppositionally conveys this. We can tell a precisely parallel
story for “Phil knows that Angelica might be the thief.” For an utterance of this to be
appropriate all parties must float, or be willing to come to float that Angelica is the thief.
Because of this, we can say that a speaker who utters this sentence expresses a float that
Angelica is the thief as a presupposition. And we could even say that the speaker
presuppositionally expresses this. This is the sense of the term that I am using above.
We might wonder what the notion of presupposition is doing here. Why don’t we just
talk about the speaker expressing this proposition, rather than doing so presuppositionally?
The reason is that the expression works like presuppositions when embedded in larger
constructions. For example, in ordinary cases of negation, we get presupposition phenomena,
where the negation does nothing to affect what it presupposed (though it does affect what is
asserted and indirectly conveyed). A speaker who utters “Phil doesn’t know that Angelica
might be the thief” typically presuppositionally expresses a float that Angelica is the thief.
Additionally, presuppositions don’t usually project out of disjunctions:
126
“Phil has a lot of
information about Angelica. Either he knows that Angelica might be the thief or he knows
that she isn’t.” An utterance of these sentences doesn’t presuppositionally express a float that
she is the thief. However, in the case of disjunctions, we must at least hold out the possibility
of what is known being the case.
127
If we know that Angelica is not the thief (and sink that she
is the thief), it is inappropriate to say, “Either Phil knows that Angelica might be the thief or
he knows that she isn’t.” Rather we would have to say “Either Phil thinks that Angelica might
be the thief or he knows that she isn’t.” I’ve not tested all of the common features of
presupposition projection, but based on these examples, it looks reasonable to conclude that
the presupposition phenomena for ‘might’-knowledge reports work exactly as one would
126
Here’s a non-modal case for comparison: “The great mathematician just answered the question of whether x
is prime. Can you believe he finally figured it out? He either knows that x is prime or knows that x is not
prime.”
127
The following is anomalous: “Either Phil knows that three is prime or he knows that three is not prime.”
This, presumably, is because we have ruled out that three is not prime (We have to say, “Either Phil knows
that three is prime or he believes that three is not prime”).
116
expect.
Let’s move on to consider sentences in which ‘might’ is embedded under attitude verbs
like ‘hope’ and ‘fear’ (Anand and Hacquard (2013) call these attitudes emotive doxastics).
Sentences like “You hope that it might rain” can seem anomalous. However, both of the
semantic theories that we’ve discussed in connection with Contextualism predict that
sentences in which ‘might’ is embedded under emotive doxastic verbs are syntactically and
semantically well-formed. So it would be puzzling, on my account, if no uses of such
sentences are felicitous.
Fortunately, some uses of that sentence are appropriate.
128
Consider the following case
(Mark Criley pc): Imagine that there is a family picnic planned for this afternoon. You would
rather play golf with your friends than go to the picnic. You know that if your family thinks it
might rain, the picnic will be canceled. And you want this to happen. However, you don’t
want it to rain because that would ruin your round of golf. In such a situation we can
appropriately utter, “You hope that it might rain.”
129
We can give the following account of the direct speech act performed by a speaker who
uses such a sentence (for the case of hope in particular):
ContextualismHope: The direct speech act performed by a speaker who assertively
utters
┌
A hopes that it might be that S
┐
, where p is the content of S in the context and
o is the person denoted by A, is an assertion of the proposition that o hopes that p is
consistent with the information of the context.
ContextualismHope does not, by itself, tell us anything about the situations in which an
utterance of a sentence like “You hope that it might rain” is felicitous or infelicitous. But I
think the following conjecture is plausible. When the information of the context for such a
report is the information of the subject, the use will generally be infelicitous, and when it is
some other body of information, it will be felicitous. This makes sense, as it is generally
problematic to utter “You hope that it is consistent with your information that it rains” or
“You hope that you don’t know (believe) that it doesn’t rain”. But it is completely fine to say,
“You hope that it is consistent with your family’s information that it rains” or “You hope that
your family doesn’t know (believe) that it doesn’t rain.”
I won’t give a full explanation of why this is so, but here is the sketch of such an
128
A speech act modifier view, like that in Schnieder 2010, would predict that all uses of this sentence are
infelicitous (at least where ‘might’ is epistemic).
129
Dorr and Hawthorne (forthcoming) and Marushak and Shaw (MS) give similar cases.
117
explanation, due to John Hawthorne.
130
The sketch relies on two ideas. First, hoping that
something obtains requires being ignorant about whether that thing obtains. Second, it is
strange to convey that someone is ignorant about her own state information or lack of
knowledge or belief. With these two ideas in hand, we can realize that in using “You hope
that it might rain”, we suggest that you are ignorant of whether it is consistent with the relevant
information that it rains (from the first idea). In the subject-centric use of ‘might’, this is
conveying that you are ignorant of whether it is consistent with your own information that it
rains. The second point predicts that this is a strange thing to convey – thus explaining the
anomaly. In the non-subject-centric use, what is conveyed is that you are ignorant of whether
it is consistent with some other body of information that it rains. But it is not strange to
convey that you are ignorant in this way. So, such utterances are predicted to be felicitous.
In the picnic case, the content of what we say about you, according to
ContextualismHope, is that you hope that it is consistent with your family’s information that it
rains. Given the way I just drew the line between felicity and infelicity, our utterance is
correctly predicted to be felicitous. Nonetheless, in ordinary scenarios, there is a presumption
that when ‘might’ is embedded under an attitude verb, the information of the subject of the
report is relevant. In such cases, as we’ve seen, an utterance of a sentence like “You hope that
it might rain” is infelicitous.
131, 132
There has been a lot of discussion of embedding ‘might’ under suppositional verbs, like
‘suppose’, ‘assume’, and ‘imagine’. Here is that natural extension of Contextualism to the case
of embedding under suppositional verbs:
ContextualismSupposition: The direct speech act performed by a speaker who
assertively utters
┌
Suppose that it might be that S
┐
, where p is the content of S in the
context, is a request that the addressee suppose that p is consistent with the
information of the context.
Again, we will see that speakers typically do more than just perform the direct speech act when
130
He gave this sort of explanation in a meeting of his seminar at USC in Fall 2013. He is more concerned with
these ideas as they apply to knowledge, rather than information or belief. Similar concerns are discussed in
Dorr and Hawthorne forthcoming (§9).
131
I won’t discuss whether a speaker who utters a sentence like “You hope that it might rain” typically performs
an indirect speech act. As we have seen, we can explain all of the data in terms of the direct speech act. It
may be that subject-centric uses of such sentences often constitute indirect speech acts about speakers hoping
that they take propositions as serious options. But conveying such hopes would be just as infelicitous as
conveying the corresponding hopes about consistency with one’s own information.
132
See Anand and Hacquard 2013 for a discussion of interesting data about the divergence in acceptability
between possibility and necessity modals under verbs like ‘hope’.
118
they utter sentences of this form.
Consider the sentence “Suppose it’s raining and it might not be.” Seth Yalcin (2007;
2011) contends that sentences of this form are marked. But ContextualismSupposition alone
predicts that we can coherently issue imperatives like these. We can see this by contrasting
“Suppose it’s raining and it might not be raining” and “Suppose it’s raining and it’s consistent
with your information that it is not raining.” As Yalcin notes, the latter seems like a fine thing
to say in many circumstances – those where we want our addressee to take on a supposition
involving her own ignorance. The former seems infelicitous in at least almost all situations.
But, according to ContextualismSupposition these sentences are, in speaker-centric cases, used to
perform the same direct speech act. So, if ContextualismSupposition were the whole story about
what is communicated, we would expect these two sentences to be on a par, as far as felicity
goes.
It is important to note that embedding under ‘suppose’ plays a pivotal role in this
puzzle. There is no problem for Contextualism from the fact that it is problematic to
assertively utter “It’s raining and it might not be.” This is because if one is in a position to
narrow down the open possibilities to those where it is raining (as one must be to utter the
first conjunct), one is also in a position to rule out that it is consistent with the information of
the context (in a speaker-centric context) that it is not raining (i.e. rule out that it might not be
raining). This is a Moore-paradox-style explanation of the seeming incoherence of assertively
uttering this sentence. And this explanation carries over to utterances of “It’s raining and it’s
consistent with my information that it isn’t.” Only when we embed the sentences under
‘suppose’ do we see the problematic contrast. Moore-paradox-style explanations rely on a
speaker assertively uttering a sentence, and they don’t go through when the sentence is
embedded under ‘suppose’. The felicity of “Suppose it’s raining and it’s consistent with your
information that it isn’t” witnesses this. The problem that remains, however, is an explanation
of the infelicity of “Suppose it’s raining and it might not be.” Yalcin 2007 contains extensive
discussion of the relationship between the puzzle in the text and Moore’s paradox.
133
Unsurprisingly, there is more going on when speakers utter suppositional ‘might’-
sentences. As we’ve seen throughout this dissertation, the most important feature of ‘might’-
sentences is that they are often used to perform a standardized indirect speech act:
133
See Schnieder 2010 for the position that it is never felicitous to utter a sentence where an epistemic modal is
embedded under ‘suppose’. He is wrong about this. Consider, “Suppose that it might rain. Should we bring
an umbrella?” His mistake seems to stem from focusing on ‘perhaps’ rather than ‘might’ or ‘may’. For more
on ‘perhaps’ see note 140.
119
REQUESTING SUPPOSITIONAL FLOATS: A speaker who utters a sentence of
the form
┌
Suppose that it might be that S
┐
, where p is the content of S in the context,
typically indirectly requests, as a standardized speech act, that the addressee be in a
suppositional state of floating p.
The request is that the addressee adopt the float as part of a suppositional state. It is
important, in understanding my solution to Yalcin’s puzzle, to understand what a
suppositional state of floating a proposition is. In general, supposing involves taking on some
attitudes for the purposes of hypothetical reasoning rather than genuinely adopting those
attitudes. Think of a suppositional state as one you enter when reasoning hypothetically. The
role that suppositional floats play in hypothetical reasoning with uncertainty is like the role
that ordinary floats play in ordinary reasoning with uncertainty. Suppositional floats are states
of taking a proposition as an option in hypothetical reasoning. They are not states of
supposing (in hypothetical reasoning) that you float some proposition. In general, being in a
suppositional state of floating p is different from supposing that you float p in the same way
that being in a suppositional state of accepting q (sometimes just called supposing q) is
different from supposing that you accept q.
The request to be in a state of suppositionally floating p is a standardized indirect
request. It is often the speech of primary importance, is not always performed by a speaker
who uses the expression,
134
and it can be figured out by a hearer in a manner that has a
Gricean flavor. It need not be explicitly calculated, is detachable, and is at-issue.
With REQUESTING SUPPOSITIONAL FLOATS in hand, we have the tools to
explain Yalcin’s puzzle. Since my solution is similar to Yalcin’s own solution, it will be useful
to understand his explanation first. According to Yalcin, “Suppose it might not be raining” is
used to request that the addressee be in a suppositional state that is compatible with the
possibility that it is not raining. So “Suppose it’s raining and it might not be” is used to
request that the addressee be in a suppositional state that accepts that it is raining and is
compatible with the possibility that it is not raining. This is a state in which all open
possibilities are ones in which it is raining and some possibility is one in which it is not
134
Here is an example of such a case from Dorr and Hawthorne forthcoming (38). Suppose we are overseeing a
hospital whose policy is to not release patients until HIV tests come back negative. In explaining this policy
to you, I can felicitously say, “Suppose Fred is HIV negative but might be HIV positive. Then he has to stay
in the hospital until his negative test results come through.” In such a case, I do not express a suppositional
float that Fred is HIV positive. Rather I just request you to take on the supposition that Fred is HIV
negative though the tests have not yet proved this. That such an utterance is felicitous suggests that the
expression of the suppositional float is particularly important for generating the infelicity in Yalcin’s cases.
120
raining. Since being in such a state is impossible, issuing such a request is incoherent and
using the sentence which encodes this request is inappropriate.
My account offers a similar explanation, which relies on REQUESTING
SUPPOSITIONAL FLOATS. Given REQUESTING SUPPOSITIONAL FLOATS, a
speaker who utters, “Suppose it might not be raining”, indirectly requests in a standardized
way that the addressee be in a suppositional state of floating that it is not raining. So,
according to my view, “Suppose it’s raining and it might not be” is typically used to perform
two requests: (i) a request that the agent be in a suppositional state that accepts that it is
raining and accepts that it is consistent with her information that it is not raining (from
ContextualismSupposition) and (ii) an indirect request that she be in a suppositional state of
floating that it is not raining (from REQUESTING SUPPOSITIONAL FLOATS). Neither
request alone is problematic, but together they are. For, accepting that it is raining and
floating that it is not raining is irrational (by the Singlemindedness norm from Chapter
Three). It is, likewise, irrational to be in suppositional versions of these states. And, since, by
The Direct/Standardized Indirect Approach, the requests to be in suppositional versions of
these states are at-issue speech acts, it is problematic to make them. So, it is inappropriate to
utter the sentence that is used to make them – “Suppose it’s raining and it might not be.”
Note, furthermore, that the crucial indirect request is standardized. So, “Suppose it is
consistent with your information that it is not raining” is not used perform the same indirect
request as “Suppose it might not be raining.” This explains the contrast, which Yalcin is right
to stress, between “Suppose it is raining and it is consistent with your information that it is not
raining” and “Suppose it is raining and it might not be.”
135, 136
In this section, I discussed a number of ways in which ‘might’-sentences embed under
attitude verbs. I gave an account of each distinctive sort of embedding. And I showed how
these accounts yielded plausible predictions about ‘might’-communication.
7. Antecedents of Conditionals
We are now in a position to see what happens when ‘might’ is embedded in the
antecedent of a conditional. Remember that in §2 that we accepted a view of conditionals
according to which they are used to make conditional assertions. That is, uses of conditionals
135
It also explains the felicity of “Suppose it is raining and, for all you know, it might not be.”
136
Other contextualist replies to Yalcin’s Puzzle are given by Barnett (2009), Sorenson (2009), and Dorr and
Hawthorne (forthcoming). Crabill (2013) gives a reply based on the Bach-inspired proto-propositional
account.
121
are assertions of the consequent on the supposition of the antecedent. I will retain this
position in this section. It, together with our account of how ‘might’ embeds under a verb like
‘suppose’, yields an account of how ‘might’ embeds in the antecedent of a conditional.
Consider the sentence, “If it might rain, then you should bring an umbrella.” A use of
this sentence can, pace Schnieder (2010), be felicitous (though there is a tendency for the
consequent of such conditionals to either be imperatives or contain a deontic or epistemic
modal). We can give the following account of ‘might’ in the antecedents of conditionals:
ContextualismAntecedent: The direct speech act performed by a speaker who
assertively utters
┌
If it might be that S
┐
, where p is the content of S in the context, is a
request that the addressee suppose that p is consistent with the information of the
context.
But, as in the ‘suppose’ case, the speaker also performs a standardized indirect request. He
requests that the addressee be in a suppositional state of floating p:
REQUESTING SUPPOSITIONAL FLOATSANTECEDENT: A speaker who utters an
expression of the form
┌
If it might be that S
┐
, where p is the content of S in the
context, typically indirectly requests, as a standardized speech act, that the addressee be
in a suppositional state of floating p.
This request has the properties of a standardized indirect speech act. It is often the speech act
of primary importance, is not always performed by a speaker who utters an antecedent
containing ‘might’, and can be reasoned to using relevance considerations with a Gricean
flavor. However, it can be grasped without such reasoning, is detachable, and is at-issue.
These properties all obtain in the same way that they did in the cases where ‘might’ was
embedded under ‘suppose’.
Yalcin (2007) realizes that the sort of problem we discussed in the last section arises not
just when ‘might’ is embedded under ‘suppose’, but also when it is embedded in the
antecedent of a conditional. The suppositional view of the antecedents of conditionals that
we’ve just developed allows us to offer the same explanation in each case. For example, we
can explain the infelicity of “If it’s raining and it might not be” in the same way that we
explained the infelicity of “Suppose it’s raining and it might not be” (and we can do so while
correctly predicting that “If it’s raining and it is consistent with my information that it is not
raining” is felicitous). “If it’s raining and it might not be” is used to request that the addressee
be in an irrational state (by the combination of direct and standardized, indirect requests
122
performed). So uttering this sentence is infelicitous.
Here I have given an account of ‘might’-sentences embedded in the antecedent of a
conditional. Above I gave an account of ‘might’-sentences embedded in the consequent of a
conditional. What about the case where a ‘might’-sentence is embedded in both the
antecedent and the consequent of a conditional?
Let’s start with the direct speech act performed by a speaker who uses such a sentence.
The direct speech act performed in uttering the antecedent is a request to be in a
suppositional state and the direct speech act performed in uttering the consequent is an
assertion. When we put these two acts together, we get a conditional assertion of the content
of the consequent, on the supposition requested by the antecedent:
ContextualismAnt+Cond: The direct speech act performed by a speaker who
assertively utters
┌
If it might be that S, then it might be that T
┐
, where p is the content
of S and q is the content of T in the context, is a conditional assertion of the
proposition that q is consistent with the information of the context on the supposition
that p is consistent with the information of the context.
There is nothing special about this account, or how the requisite speech acts compose to yield
the conditional assertion. It works just like the ordinary, non-modal case.
Again, it is reasonable that a speaker who utters such a sentence does more than simply
perform the direct speech act. Such a speaker also typically performs an indirect speech act.
This indirect speech act is related to the indirect speech acts performed by speakers who utter
the antecedent and the consequent – the first of which is a request to take on a suppositional
float and the second of which is an expression of a float. Together, these constitute an
expression of an attitude very much like a conditional float (what is expressed by a speaker
who uses a conditional ‘might’-sentence with a non-modal antecedent). But the attitude is a
bit different. A conditional float involves a commitment to float some proposition if the agent
learns some other proposition – where learning involves coming to have an attitude like
knowledge or acceptance. The attitude we are after also involves a commitment to float a
proposition. But this commitment comes into effect not if an agent learns some other
proposition, but rather if an agent floats some proposition. I’m going to call this sort of
attitude a specialized conditional float. Likewise, I will assume that there are specialized
conditional sinks (which are commitments to sink a proposition on floating some other
propositions).
123
Specialized conditional floats are different from all of the sorts of attitudes that were
discussed in Chapter Three. They are part of a more general category of attitudes that include
ordinary conditional floats. The idea is that there is a general category of conditional float-like
attitudes that breed commitments to float a proposition on having some other attitude. In the
case of ordinary conditional floats, this latter attitude is acceptance. In the case of specialized
conditional floats, it is a float (it could even be a float to a particular degree).
137
Like I said,
this picture of conditional attitudes requires commitments beyond those taken in Chapter
Three. I won’t try to fully to defend them here. Rather I will show the work that specialized
conditional floats can do in the sort of explanations of conversational data and inferences that
we’ve seen throughout this chapter – if we assume specialized conditional floats are similar to
ordinary conditional floats in terms of norms on reasoning and disagreement.
In particular, I’m going to assume that specialized conditional floats are governed by
rational rules which are, mutatis mutandis, like those that govern ordinary conditional floats.
For instance, something like Conditional Detachment holds: For any inquiry, if an agent, in
that inquiry, has a specialized conditional float (sink) in p given a float in q and floats q, then
she should, in that inquiry, float (sink) p. Finally, I’m going to assume that disagreement
works in a similar way with specialized conditional floats and sinks. For instance, an agent
who has a specialized conditional sink in p given a float in q will be disposed to negatively
evaluate a specialized conditional float in p given a float in q. So, she will, by Evaluationism,
disagree with an agent who has such a specialized conditional float. In making these claims
I’m assuming that specialized conditional floats and sinks follow ordinary conditional floats
and sinks in these areas.
Let’s return to the issue of communication. A speaker who uses a conditional with a
‘might’-sentence in both the antecedent and the consequent expresses a specialized
conditional float. It is not surprising that this is the effect that changing from a non-modal to
a modal antecedent has, because it makes sense that the supposition would correspondingly
change from requiring acceptance to floating:
EXPRESSING SPECIALIZED CONDITIONAL FLOATS: A speaker who utters a
sentence of the form
┌
If it might be that S, then it might be that T
┐
, where p is the
content of S and q is the content of T in the context, typically indirectly expresses, as a
standardized speech act, a specialized conditional float in q given a float in p.
137
See Lennertz MS b for a general discussion of this picture of generalized conditional attitudes.
124
Such a speech act is a standardized, indirect one as follows. It is often the speech act of
primary importance, though it is not always performed by speakers who use this sentence (in
particular, when they use ‘might’ exo-centrically). In cases where it is performed, a hearer can
figure this out using Gricean-style relevance reasoning – using the fact that expressing a
specialized conditional float is much more relevant than conditionally asserting a proposition
about what is consistent with one’s information on the supposition that some other
proposition is consistent with that information. The expression is also standardized. So, a
hearer need not go through the sort of reasoning just alluded to in order to figure out which
speech act was performed. Furthermore this speech act is detachable (compare with “If it is
consistent with my information that Angelica is the thief, then it is consistent with my
information that she is the glutton”). And, as we will see in the next paragraph, this act of
expression is at-issue.
EXPRESSING SPECIALIZED CONDITIONAL FLOATS has a number of
explanatory virtues. Consider, the following conversation:
(Glutton) Phil: If Angelica might be the thief, then she might be the glutton.
Lil: No. If she might be the thief, then she can’t be the glutton. The
cookies were sold to and eaten by someone far away – who couldn’t have
swiped them.
In making his utterance, Phil indirectly expresses a specialized conditional float that Angelica
is the glutton given a float that she is the thief (from EXPRESSING SPECIALIZED
CONDITIONAL FLOATS). Lil can respond in a straightforward manner by The
Direct/Standardized Indirect Approach. She has a specialized conditional sink that Angelica
is the glutton given a float that she is the thief, and she will disagree with Phil according to
Evaluationism. Thus, she can appropriately respond by saying “No.” Just like (Cookies)
couldn’t be explained without EXPRESSING FLOATS, (Glutton) can’t be explained without
EXPRESSING SPECIALIZED CONDITIONAL FLOATS.
Furthermore, EXPRESSING SPECIALIZED CONDITIONAL FLOATS allows us
to explain inferences like the following:
(22) If Angelica might be the thief, then she might be the glutton.
(23) Angelica might be the thief.
(24) So, Angelica might be the glutton.
Given a plausible account of the rational relations involving specialized conditional floats – in
125
particular, the analogue of Conditional Detachment that I mentioned above – an agent who
has the attitudes standardly expressed by a speaker who uses (22) and (23) is rationally
committed to having the attitude standardly expressed by a speaker who uses (24).
In this section I presented the account of ‘might’-sentences in the antecedents of
conditionals. The similarities between it and the account of ‘might’-sentences embedded
under suppositional verbs allowed us to explain the appearance of Yalcin’s puzzle in this
context. I also showed how the account of ‘might’-sentences in the antecedents of conditionals
combines with the account of ‘might’-sentences in the consequents of conditionals to give us
an account of sentences where ‘might’ occurs in both the antecedent and the consequent. This
joint theory produced nice explanations of why certain sorts of conversational responses are
appropriate and why certain sorts of inferences are compelling.
138
8. Tense
What happens when ‘might’ is embedded under a tense operator, like in: “Angelica
might have been the thief”? It is important to realize that such sentences are ambiguous
between talking about what Willer (2013) calls a past possibility and a possible past. We
might use this sentence to talk about what, in the past, was a possibility at that time (even
though it may no longer be a possibility now). Or we may use this sentence to talk about our
present uncertainty about a past state of affairs. We could disambiguate these with language
as follows: “It was the case that Angelica might be the thief” and “It might (now) be the case
that Angelica was the thief.” I will only be interested in the former reading here, since that is
the one that involves embedding a ‘might’-sentence. The latter is really just a simple ‘might’-
sentence.
So, here is the account of ‘might’-sentences which are embedded under a past tense
operator:
ContextualismPast: The direct speech act performed by a speaker who assertively
utters
┌
It was the case that it might be that S
┐
, where p is the content of S in the
context, is an assertion of the proposition that p was consistent with the information
relevant at some time before the time of the context.
Past tensed ‘might’-sentences are also typically used to perform an indirect speech act:
138
A similar sort of joint theory must be given for quantificational sentences like, “Each person who might be
the thief might be the glutton.” The amendments needed to go from ordinary to what I would call
specialized quantificational floats and sinks will be very similar to the amendments we saw in the conditional
case. I leave the details to the reader.
126
CONVEYING PAST FLOATS: A speaker who utters a sentence of the form
┌
It was
the case that it might be that S
┐
, where p is the content of S in the context, typically
indirectly conveys, as a standardized speech act, that in the past, the relevant person or
group floated p.
Let’s consider an example. Suppose that Phil learns that Angelica isn’t the thief
through questioning her. Lil always knew that she wasn’t the thief. Consider:
(Past Cookies) Lil: Angelica’s not the thief. Why did you bring her in for
questioning?
Phil: Well, she might have been the thief.
Lil: You were wrong. She couldn’t have been the thief.
According to ContextualismPast, Phil asserts that in the past, it was consistent with his
information then that Angelica was the thief. As usual, we can’t make sense of Lil’s negative
response as a response to what Phil asserted. For all Lil knows, it was consistent with Phil’s
information that Angelica was the thief. So she shouldn’t disagree with him about that.
According to CONVEYING PAST FLOATS, Phil also indirectly conveys that in the past, he
floated that Angelica was the thief. This indirect speech act is what Lil responds to. She
always sank that Angelica is the thief.
It makes sense that in (Past Cookies), Phil indirectly conveys that his former-self
floated that Angelica is the thief. The reason for this is that he is attempting to explain his
action. There is an analogy between present and past floats: present floats are to reasoning
about future action as past floats are to explaining past action. Phil’s past float was what
guided his reasoning and eventually led him to act in a certain way. And he conveys that he
had this float in order to explain his action. In uttering a sentence which embeds a ‘might’-
sentence under a past tense operator, a speaker typically indirectly conveys that his former self
floated the prejacent. The point of conveying this is to explain why the speaker’s former self
acted in the way he did.
Phil’s conveying that in the past, he floated that Angelica was the thief has all of the
hallmarks of a standardized, indirect speech act. It is the speech act of primary importance,
but is not always performed by a speaker who uses the sentence. A hearer can figure out that
it is performed in a way that has a Gricean flavor. Actions are explained by possibilities that
were taken seriously. So, in order for Phil’s utterance to be relevant for an explanation, he
must be talking about what he floated – not simply what was consistent with his information.
127
Despite such reasoning being available, a hearer can grasp that this speech act was performed
without going through some reasoning. And this speech act is detachable (compare with an
explicit consistency sentence). Finally, Phil’s indirect speech act can be at-issue, allowing for a
straightforward response, as we saw in (Past Cookies).
So, in this section, we’ve seen that a ‘might’-sentence embedded under past tense
operator functions in communication very much like a simple ‘might’-sentence. The main
difference is that past ‘might’-sentences report past floats for the purpose of explaining past
reasoning and action, while present ‘might’-sentences express floats for the purpose of guiding
present and future reasoning.
139
In this chapter, I’ve presented my picture of communication involving sentences in
which ‘might’ is embedded in larger constructions. I’ve not dealt with every way ‘might’ could
embed, but I’ve discussed a variety of examples that are particularly prevalent or interesting. I
take this chapter and the last to comprise my account of ‘might’-communication. My general
strategy has been to combine an account of the direct speech act performed by a speaker who
uses a particular type of ‘might’-sentence with an account of the indirect speech act performed
by such a speaker. In many cases the indirect speech act is most important – explaining data
about appropriateness of utterances and responses, as well as reasonable inferences. As I have
stressed, the indirect speech act is not determined compositionally – nor should we expect it to
be. This is because it is a pragmatic phenomenon, which is partially determined by the
purposes of the conversation in which the ‘might’-sentence is used.
In all of the cases we’ve seen in this chapter, the indirect speech act is either an
expression of or a conveying about a float-like attitude. This makes sense given that ‘might’-
sentences are used to coordinate reasoning with uncertainty and, as we discussed in Chapter
Three, float-like attitudes are what agents use in explicit reasoning with uncertainty. We did
see, in this chapter, that we needed to liberalize the picture from the earlier chapter to account
for uses of ‘might’ in the antecedents of conditionals. Though I didn’t fully argue that such a
liberalization was benign, I did suggest a general picture of conditional attitudes into which
ordinary and specialized conditional floats fit. Furthermore, as we saw, this liberalization
139
’Might’-sentences do not embed under the future tense very easily. Sentences like “It will be the case that
Angelica might be the thief” are pretty stilted. Uses of this sentence, while they are grammatical, seem like
they’ll rarely, if ever, be useful. Whether there are uses of this sentence that are acceptable will depend on
whether it makes sense to talk about future possibilities that are not present possibilities. Either way, the
explanation will be pragmatic and will be consistent with my general account.
128
allowed for a unified explanation of conversational data involving ‘might’-sentences –
including ‘might’-sentences in the antecedents of conditionals. Because of this unity and the
prima facie plausibility of the general picture of conditional attitudes, working on the
assumption of the existence of specialized conditional floats is a reasonable.
In the next chapter I will give an account of communication involving epistemic
‘must’-sentences and sentences which are used to talk about epistemic probability. This
account will use some of the resources developed in this chapter – in particular, in
understanding negated ‘might’-sentences – to explain conversational data involving epistemic
‘must’- and ‘likely’-sentences. Then, in Chapter Seven, I will return to explore issues in the
philosophy of mind about what float-like attitudes, at bottom, are.
129
Chapter Six: Other Strengths of Modality
The discussion in the preceding chapters has dealt almost exclusively with ‘might’-
sentences. But ‘might’ is part of a family of similar words. There are other terms that are
used to convey epistemic possibility and express floats. Some of these, like ‘might’, are modal
auxiliary verbs – eg. ‘may’ and ‘could’. Others are adjectives, like ‘possible’, and adverbs, like
‘possibly’. Finally, there are sentence modifiers, like ‘perhaps’ and ‘maybe’. I’m going to
assume that the treatment I’ve given for ‘might’ either applies to these terms directly or with
some obvious appropriate changes.
140
More interesting are modal terms that are associated with other strengths of modality.
The most obvious is ‘must’. ‘Must’ is a necessity modal.
141
In between the notions of
possibility and necessity is the more fine-grained notion of probability. This notion is often
associated with adjectives like ‘likely’ and ‘probable’ and – though this is more controversial –
with the auxiliary verbs ‘should’ and ‘ought’.
Just as in the case of ‘might’, it is difficult to explain all the uses of and responses to
uses of sentences containing these terms. The goal of this chapter is to extend the account of
communication that I have given for ‘might’ to these other terms. To do this, I will pair a
contextualist account of the direct speech act performed by a speaker who uses these terms
with an account of the standardized, indirect speech act typically performed by a speaker who
uses these terms. Just as in the case of ‘might’-sentences, the indirect speech act will often be
the expression of a float-like attitude. This makes sense since ‘must’- and ‘likely’-sentences
are, like ‘might’-sentences, used to coordinate reasoning.
In the first section of this chapter, I’ll give an account of ‘must’. Then, I will extend the
story to the more complicated case of words that denote probability. I’ll next discuss the
somewhat special case of the epistemic ‘should’. And I’ll conclude by showing that uses of
embedded ‘must’-, ‘likely’- and ‘should’-sentences should receive a treatment similar to the
one given in the last chapter for uses of embedded ‘might’-sentences. These sections combine
with what has come in the last two chapters to yield a unified picture of epistemic modal
communication.
1. ‘Must’
140
I am least sure of this in the case of the sentence modifiers. They sound much more strained in embedded
contexts than the rest of the terms. They may, in fact, function as speech act modifiers and have this written
into their semantics, in the way that Schneider (2010) suggests.
141
There are other necessity modals, like ‘have to’ and ‘got to’. The adjective, ‘necessary’, and adverb,
‘necessarily’, may get epistemic readings as well.
130
In this section I will give a contextualist account of the direct speech act performed by
a speaker who uses an epistemic ‘must’-sentence. When we pair this account with something
very much like our old nemesis from Chapter One – The Direct Approach – we will encounter
familiar problems in explaining conversations. We can avoid these problems by realizing that
‘must’-sentences, like ‘might’-sentences, are used to coordinate reasoning – and, so, are used to
indirectly express float-like attitudes. Realizing that these expressions are at-issue speech acts
allows us to explain the otherwise problematic conversations.
It’s natural to think that the notions of possibility and necessity are duals. That is, if
something is possible, then its negation isn’t necessary and if something is necessary, then its
negation isn’t possible (i.e. its negation is impossible). Since the natural language modals
‘might’ and ‘must’ are presumed to correspond to possibility and necessity, it is natural to
think that they stand in this relationship, as well. In particular, it is natural to think that what
is asserted by a speaker who uses
┌
It must be that S
┐
in a context is true just in case what is
asserted by a speaker who uses
┌
It’s not the case that it might not be that S
┐
in that context is
true. I’m going to be taking this assumption on board. This will provide us with a map for
giving an account of the direct speech act performed by a speaker who uses a ‘must’-sentence.
Let’s review the account of the direct speech act performed by a speaker who uses a
negated ‘might’-sentence (substituting the more cumbersome locution of “It’s not the case that
it might . . .” for “It can’t . . .” in ContextualismNegation):
The direct speech act performed by a speaker who assertively utters
┌
It’s not the case
that it might not be that S
┐
, where p is the content of S in the context, is an assertion
of the proposition that the negation of p is not consistent with the information of the
context.
From the assumption that ‘might’ and ‘must’ are duals, we are led to the following (notice the
weakening, given that the duality thesis is one about truth conditions of uses of these
sentences and not the sameness of the proposition asserted):
The direct speech act performed by a speaker who assertively utters
┌
It must be that
S
┐
, where p is the content of S in the context, is an assertion of a proposition that is true
just in case the negation of p is not consistent with the information of the context.
I’m going to tidy this up a bit, assuming two things (though, for our purposes, nothing
substantive hangs on them): first, that we’ve actually gotten at the proposition in the above
statement, not just one with equivalent truth conditions and, second, that a proposition
follows from a body of information if and only if its negation is not consistent with the
information. So we get:
131
Must Contextualism: The direct speech act performed by a speaker who assertively
utters
┌
It must be that S
┐
, where p is the content of S in the context, is an assertion of
the proposition that p follows from the information of the context.
One nice thing about Must Contextualism is that we were led to it from the account of
negated ‘might’-sentences together with the relationship between ‘might’ and ‘must’. But it is
exactly the sort of account we would have tried out if we had just been tempted to give a
contextualist story from scratch. We say something must be so if that thing follows from our
information. Our theoretical assumptions, then, match up very nicely with our pre-theoretic
judgments here.
Let’s consider a conversation. Suppose Susie and Lil are investigating the cookie theft.
Phil has incorrectly concluded that Angelica is the thief on the basis of misleading evidence.
Lil knows that the evidence that Angelica is the thief is misleading. They have the following
conversation:
(Must Cookies) Susie and Lil: Phil, we’re interviewing people to get information
about who stole the cookies. Do you know who stole them?
Phil: Angelica must be the thief.
Lil: No. She might not be. Some of the evidence against her is
misleading.
We can consider two bodies of information that could be the information of the context –
Phil’s and the group’s.
142
Suppose that the direct speech act is the only at-issue speech act.
Then, on the speaker-centric reading, Lil’s reply is predicted to be inappropriate and on the
group reading, Phil’s utterance is. This is the same sort of result we saw in the case of ‘might’-
sentences. So, just as in the ‘might’ case, we can’t explain some conversations on the
assumption that the direct speech act is the only at-issue speech act.
143
The obvious solution is to take a page out of our playbook for ‘might’. We need to
allow indirect, as well as direct, speech acts performed in uttering ‘must’-sentences to be at-
issue. We can do so by generalizing The Direct/Standardized Indirect Approach from a
theory about ‘might’-sentences to a theory about epistemic sentences, in general:
The Generalized Direct/Standardized Indirect Approach: An assertive utterance of an
epistemic sentence gives rise to the following at-issue speech acts: the object(s) of any
142
Here I assume that a body of information need not be restricted to a set of propositions that are known. It
can also include propositions that are merely believed. If information is restricted to what is known, then the
speaker-centric reading allows for a complete explanation of the case. However, I aim to remain neutral on
this issue and not help myself to this explanation here – relying, instead, on a similar explanation to the one I
gave in the case of ‘might’.
143
Analogues of the other theories of at-issue speech acts that we surveyed in Chapter Two will fail for the case
of ‘must’ for reasons similar to why they did for the case of ‘might’.
132
direct or standardized indirect speech act(s) performed by the speaker in making the
utterance.
Now that we have a more liberal account of at-issue speech acts, we need to see if Phil does
perform an indirect speech act and, if so, which one.
Let’s start from the duality of ‘might’ and ‘must’ –
┌
It must be that S
┐
is true just in
case
┌
It’s not the case that it might not be that S
┐
is true. Perhaps this relationship extends
beyond these sentences’ truth conditions in a context. Perhaps they also are typically used to
indirectly express the same attitude. A sentence of the form,
┌
It’s not the case that it might
not be that S
┐
, is typically used to express a sink in the negation of p (where p is the content
of S in the context). So, we might think that a sentence of the form,
┌
It must be that S
┐
, is
also typically used to express a sink in the negation of p.
This position not only follows from duality-like concerns, it also makes sense given the
purposes for which speakers use ‘must’-sentences. ‘Must’-sentences are used in order to
coordinate reasoning. They are used to mark off what propositions are taken as given, so that
the negation of these propositions are ruled out in reasoning. That is, a ‘must’-sentence is
used to express a sink in the negation of the prejacent:
EXPRESSING SINKSMUST: A speaker who utters a sentence of the form
┌
It must be
that S
┐
, where p is the content of S in the context, typically indirectly expresses, as a
standardized speech act, a sink in p’s negation.
When we say something must be so, we rule out its negation as an option in reasoning.
EXPRESSING SINKSMUST allows us to make good sense of (Must Cookies). In
making his utterance, Phil asserts that it follows from his information that Angelica is the thief
(according to Must Contextualism) and he indirectly expresses his sink that she is not the thief
as a standardized speech act (according to EXPRESSING SINKSMUST). Lil, on the other
hand, floats that Angelica is not the thief. Because she thinks that Phil came to his conclusion
based on misleading evidence, she gives a negative epistemic evaluation to Phil’s sink – and, by
Evaluationism, disagrees with him. Since Phil has indirectly expressed his sink as a
standardized speech act, Lil can, according to The Generalized Direct/Standardized Indirect
Approach, straightforwardly respond to it as she does in (Must Cookies).
We can verify that Phil’s expression of his sink has the six properties of a standardized,
indirect speech act. First, Phil’s expression is the speech act of primary importance – since
floats and sinks are what help agents in reasoning. Second, not every utterance of the sentence
“Angelica must be the thief” is an expression of a sink that she isn’t the thief. Exo-centric
133
uses, for example, are not.
144
Third, Lil can figure out what Phil expressed using reasoning
with a Gricean flavor. Since the content of Phil’s assertion is truth-conditionally equivalent to
an assertion of “It’s not the case that Angelica might not be the thief”, we can carry over the
derivation from the negation section of the last chapter. Fourth, this sort of reasoning is, in
actuality, bypassed; Lil can grasp Phil’s expression without going through any sort of
reasoning. Fifth, Phil’s act of expression is detachable (contrast the sentence he utters with “It
follows from my information that Angelica is the thief”). Finally, as seen in (Must Cookies),
Lil can straightforwardly respond to what Phil expressed. So, Phil’s expression of his sink has
all of the hallmarks of a standardized, indirect speech act.
In this section I gave an account of ‘must’-sentences that fits nicely with the account of
‘might’-sentences that was developed in the last two chapters. When a speaker utters a ‘must’-
sentence he asserts a proposition about what follows from his information and typically
indirectly expresses a sink in the negation of the prejacent. This account allows us to explain
otherwise inexplicable uses of and responses to uses of ‘must’-sentences.
2. Probability
Epistemic possibility and necessity are closely related to epistemic probability. I
advocate a contextualist account of ‘likely’-sentences, making use of a credence function.
145

This is a function which, for each proposition that some agent or group lends a particular
credence to, returns as its value the amount of confidence the agent or group has in that
proposition – measured on a real-number scale from zero to one. Likewise, the value the
credence function assigns to a proposition, given some other information, is the amount of
conditional confidence the agent or group has in the proposition given that information. Just
like there is a contextually determined body of information (which is often, but need not
always be, the information of the speaker), there is a contextually determined credence
144
Suppose again that Phil and Lil are playing battleship. Lil has 2 boats afloat – a battleship and a PT boat.
Phil knows that there is a ship on C4. He falsely believes that Lil always puts her PT boats against the side
of the board. And he knows that a PT boat cannot be both against the side of the board and on C4. Lil
knows all of this, though, in addition, she knows that there is a PT boat on C4. The following conversation is
appropriate:
(Must Battleship) Phil: You always put your PT boat against the side of the board. So there
must be a battleship on C4.
Lil: Yeah. That’s right. There must be.
Lil’s use of ‘must’ is an exo-centric use. It cannot be one that takes into account her own information.
Furthermore, she is not expressing a sink that there is not a battleship on C4. She does not rule out this
possibility as an option in reasoning, since she knows that the battleship is somewhere else.
145
I assume ‘likely’ takes, at the relevant level of abstraction, a proposition and a credal space (an information,
credence function pair) as arguments. I also suppose that, for a given context,
┌
It is likely that S
┐
is true just
in case
┌
It is probable that S
┐
is true and
┌
Probably S
┐
is true. These assumptions largely follow Yalcin
2010 – the difference being that I use credence functions, not probability functions (as I discuss in what
follows).
134
function (which is often, but need not always be, the credence function of the speaker).
This credence function resembles a probability function in a number of ways – it is tied
in some way to degrees of confidence, it assigns values on a scale between zero and one, and it
allows for conditionalization. But a credence function need not be, as is sometimes assumed,
a probability function. This is because the credence function need not satisfy the probability
axioms.
146
Furthermore, this credence function might not, and likely will not, be defined over
all propositions (or some complete algebra).
With this theoretical tool in hand, we can give the following account of the direct
speech act performed by a speaker who uses a ‘likely’-sentence:
Likely Contextualism: The direct speech act performed by a speaker who
assertively utters
┌
It is likely that S
┐
, where p is the content of S in the context, is an
assertion of the proposition that the value that the credence function of the context
assigns to p conditional on the information of the context is greater than the value it
assigns to not-p conditional on the information of the context.
147
Though we are enriching our notion of a context to include credence functions, we can leave
our accounts of ‘might’ and ‘must’ the same. Since a context includes a body of information,
we can simply say that ‘might’ is still concerned with consistency with the information of the
context and ‘must’ is still concerned with following from it.
148
Like in the case of ‘might’ and ‘must’, the direct speech act performed by a speaker
who uses a ‘likely’-sentence cannot explain all of the ways in which ‘likely’-sentences are used
and responded to in communication. We can see this by looking at the following
conversation, which we originally encountered in Chapter Two. Suppose that Phil’s credence
function assigns a high value to the proposition that Angelica is the thief, given his
information. But Lil’s credence function assigns a low – but nonzero – value to the
proposition that Angelica is the thief, given her information. Consider:
(Likely Cookies) Susie and Lil: Phil, we’re interviewing people to get information
146
The probability axioms may apply to a credence function normatively and not descriptively. For example,
though one could have credence .5 in p and .7 in not-p (so that the credences in a partition don’t sum to
one), one shouldn’t.
147
This may be an oversimplification. It may be that the correct account requires the value for p to be over
some contextually determined threshold, not over the value the function assigns to p’s negation. See Yalcin
2010 for a motivating example, where some people judge that a proposition is likely even though the credence
function assigns a greater value to its negation.
148
We could, on the other hand, change our accounts of ‘might’ and ‘must’ in the following way: what is
asserted by a speaker who utters
┌
It might be that S
┐
in a context is true just in case the credence function of
the context assigns a positive value to p given the information of the context; and what is asserted by a
speaker who utters
┌
It must be that S
┐
in a context is true just in case the credence function of the context
assigns the value 1 to p given the information of the context. This probabilistic account of ‘might’ and ‘must’
is similar to the one pursued in Schulz 2010 and criticized in Willer 2011. I’ll put this proposal to the side
without further discussion.
135
about who stole the cookies. Do you know who stole them?
Phil: No, but it’s likely that Angelica is the thief.
Lil: No. It’s not likely that she is. I was with her almost all
afternoon.
Phil asserts that the value his credence function assigns to the proposition that Angelica is the
thief, given his information, is higher than the value it assigns to the negation of that
proposition, given his information. But we cannot make sense of Lil’s response as a response
to his assertion, because, for all Lil knows, what Phil asserted is true. So, if this assertion were
the only at-issue speech act, it would not be appropriate for Lil to respond negatively. But it is
appropriate for Lil to respond in this way. So, there must be some other at-issue speech act.
149
The solution depends on realizing that Phil does more than just assert that proposition
in making his utterance. He also indirectly expresses an attitude. Remember that our
working hypothesis is that epistemic sentences are used to help coordinate reasoning with
uncertainty. Simple ‘might’-sentences are used to express bare floats, conditional ‘might’-
sentences are used to express conditional floats, ‘must’-sentences are used to express sinks, etc.
It makes sense that ‘likely’-sentences fit into this pattern. The sort of float-like attitude that it
makes sense for a ‘likely’-sentence to express is a degreed float. By saying that something is
likely, agents typically indirectly express an attitude of taking that proposition as a likely
option in reasoning. Agents do this because they aim to help the group in reasoning:
EXPRESSING LIKELY FLOATS: A speaker who utters a sentence of the form
┌
It is
likely that S
┐
, where p is the content of S in the context, typically indirectly expresses,
as a standardized speech act, a likely-degreed float in p.
150
These acts of expressing likely floats are standardized, indirect speech acts. They share
the three properties of indirect speech acts. First, for the reason presented in the last
paragraph, the expression of a likely float is often the speech act of primary importance.
Second, there are situations in which a speaker can utter a ‘likely’-sentence without expressing
a likely float.
151
Third, that a float is expressed can be figured out based on the direct speech
149
As in the other cases, Phil does not assert a proposition about the value of the group’s credence function
conditional on the group’s information. He is not in a position to make such an utterance since he doesn’t
know what information they have or what their credence function is like.
150
More generally:
EXPRESSING DEGREED FLOATS: A speaker who utters a sentence of the form
┌
It is D that
S
┐
, where p is the content of S in the context and D picks out degree of likelihood d, typically indirectly
expresses, as a standardized speech act, a d-degreed float in p.
151
The following is an example of an exo-centric use. Suppose again that Phil and Lil are playing battleship.
Lil has 2 boats afloat – a battleship and a PT boat. Phil knows that there is a ship on C4. He falsely believes
that Lil usually puts her PT boats against the side of the board. And he knows that a PT boat cannot be
both against the side of the board and on C4. Lil knows all of this, though, in addition, she knows that there
is a PT boat on C4. Still the following conversation is appropriate:
(Likely Battleship) Phil: You usually put your PT boat against the side of the board. So it is
136
act and the features of the context. This will be discussed in the next paragraph. The act of
expressing a likely float also has the properties of standardized speech acts. The hearer need
not go through any sort of reasoning to figure out what the speaker communicated. The
expression of a float is detachable from the content of the direct speech act (contrast with “My
credence function assigns a higher value to Angelica’s being the thief conditional on my
information than to her not being the thief conditional on my information”). Finally, it is
acceptable to straightforwardly respond to Phil’s expression, as shown in (Likely Cookies).
Thus, it looks like Phil’s expression of a likely float in (Likely Cookies) is a standardized,
indirect speech act.
As I said, a hearer can figure out that a speaker who utters a ‘likely’-sentence expresses
a likely float based on the direct speech act performed and the features of the context. The
feel of this reasoning is the same as for ‘must’- and negated ‘might’-sentences. Facts about the
value a credence function assigns to a proposition conditional on the information of the
context are often not transparent to an agent. So, in uttering a sentence to assert some
proposition about the value of his credence function, a speaker signals that he takes himself to
know the value (or at least the approximate value) that his credence function assigns to that
proposition. In Phil’s case he signals that he notices that his credence in the proposition that
Angelica is the thief given his information is higher than his credence in that proposition’s
negation given his information. When one knows one’s credence in a proposition is high and
one is reasoning about a question to which that proposition is relevant, one will typically likely
float that proposition. Furthermore, figuring out one’s credence in a proposition is not
something an agent will try to do unless he is aiming to take that proposition as an option in
reasoning to some degree. When he figures out his credence in that proposition, as long as it
is great enough, he typically takes that proposition as an option to the corresponding degree.
Since Lil knows all of these facts about Phil, she can conclude that he expressed his likely float
that Angelica is the thief. In general, a hearer can figure out that a speaker has expressed a
likely float by reasoning based on what he asserts in uttering a ‘likely’-sentence together with
the assumption that he is aiming to help in reasoning.
The fact that Phil indirectly expresses a likely float as a standardized speech act in
(Likely Cookies) allows us to explain Lil’s response. Phil takes it as a likely option in
reasoning that Angelica is the thief. Lil takes it as an unlikely option in reasoning that
likely that there is a battleship on C4.
Lil: Yeah. That’s right. It’s likely that there is.
Lil’s use of ‘likely’ is an exo-centric use. It cannot be one that takes into account her own information.
Furthermore, she is not expressing a likely float that there is a battleship on C4. She does not take this
possibility as a likely option in reasoning, since she knows that the battleship is somewhere else.
137
Angelica is the thief. Lil thinks that her unlikely float better respects the evidence, so she
negatively evaluates Phil’s likely float. So, by Evaluationism, she disagrees with Phil. As we
said, Phil indirectly expresses his likely float as a standardized speech act. Given The
Generalized Direct/Standardized Indirect Approach, Lil can appropriately respond to it by
saying “No.”
‘Likely’ is gradable; a proposition can be very likely. The direct speech acts performed
by speakers who use sentences involving ‘very likely’ leave us searching for other at-issue
speech acts – and similarly for expressions of quantitative uncertainty, like ‘three-quarters
likely’. We can extend the account just given in a straightforward manner to these expressions.
As we saw in Chapter Three, people can have very likely floats and three-quarters likely floats.
‘Very likely’-sentences and ‘three-quarters likely’-sentences are used to indirectly express these
attitudes. In general, a ‘likely’-sentence of some degree is typically used to indirectly express a
float of that degree.
152
The illustration that such speech acts are indirect and standardized
proceeds in the same manner as the likely case. So, I won’t take the space to explicitly do that
for each case.
In this section I’ve extended my account from coarse-grained epistemic terms – like
‘might’ and ‘must’ – to fine-grained ones which are concerned with probability. Pairing a
contextualist account of the direct speech act performed by a speaker who uses a ‘likely’-
sentence with an account of how a speaker uses such a sentence to indirectly express a likely
float allowed us to explain data about uses of and responses to uses of sentences of this sort.
3. How ‘should’ Differs from ‘likely’
‘Should’ and ‘ought’ can be used as epistemic modals. They are often called weak
necessity modals. In what follows I’ll talk solely about ‘should’, though the account I give will
apply to ‘ought’ as well.
153
It seems that ‘should’ is stronger than ‘might’ but weaker than
‘must’. That is (talking about sentence truth for ease of exposition), it seems that if
┌
It should
be that S
┐
is true at a context, then
┌
It might be that S
┐
is true at that context, but not vice
versa. And it seems that if
┌
It must be that S
┐
is true at a context, then
┌
It should be that S
┐
is true at that context, but not vice versa.
152
See note 150 for a precise statement.
153
‘Should’ and ‘ought to’ seem, on first look, to have exactly the same distribution. But von Fintel and Iatridou
(2008) give the following example to show that there is at least one sort of use of ‘should’ that can’t be
mirrored by a use of ‘ought to’:
(25) It’s strange that he should do that.
(26) # It’s strange that he ought to do that.
This is a strange example, however, because ‘should’ in (25) seems to have a vacuous reading – i.e. the
sentence means that it is strange that he does that.
138
We just dealt with an epistemic term that is stronger than ‘might’ and weaker than
‘must’ – ‘likely’. Perhaps we can give the same account of ‘should’ as we gave of ‘likely’. Such
an account is inspired by Finlay’s (2009; 2010) account of ‘ought’.
154
This seems promising
given the following case. Imagine that Angelica mailed a package four days ago. When she
mailed it, she learned from UPS that 80% of packages arrive at their destination within four
days. Now (four days after mailing the package), she is talking to Susie. Both of the following
conversations are appropriate:
(Should Package) Susie: Did your package get there yet?
Angelica: I’m not sure, but it should be there by now.
(Likely Package) Susie: Did your package get there yet?
Angelica: I’m not sure, but it it is likely that it is there by now.
But, treating ‘should’ and ‘likely’ the same isn’t quite right, as the following contrast
shows.
155
Consider the package scenario. But this time suppose that Angelica also knows that
the package did not, in fact, arrive yet. Consider the following conversations:
(Should Package') Susie: Did your package get there yet?
Angelica: It’s not there yet, but it should be.
(Likely Package') Susie: Did your package get there yet?
Angelica: # It’s not there yet, but it’s likely that it is.
Since Angelica’s response in (Should Package') is appropriate and her response in (Likely
Package') isn’t, ‘should’ and ‘likely’ can’t have the same communicative impact. In fact,
‘should’ is different from most other epistemic terms in this way. Consider scenarios in which
Angelica utters, “It’s not there yet, but it might be” or “It’s not there yet, but it must be”. Both
are inappropriate.
What is different about ‘should’? Here is my hypothesis. It places restrictions on the
information that is taken into account. Roughly, in an utterance of a sentence of the form
┌
It
should be that S
┐
where p is the content of S in the context, ‘should’ takes into account the
maximal subset of information of the context that does not entail p or its negation. To be
more precise, we need to realize that there is often no unique maximal subset of the
154
A related account is Horn’s (1989) treatment of ‘ought’ as quantifying in the way that ‘most’ does. Von Fintel
and Iatridou (2008) object to this sort of account. They contend that “You should do the dishes” doesn’t
mean that in most of the relevant worlds you do the dishes. It means, they think, that in the best of the
relevant worlds, you do the dishes. Von Fintel and Iatridou’s way of treating the epistemic ‘should’ involves
using an ordering which ranks possibilities by normality. Propositions that should be so are those that are
true in all the possibilities that are ranked most highly according to the ordering. The requirement of truth
in all of some set of possibilities is why ‘should’ is called a weak necessity modal. And the restriction to only
the most normal – rather than all the open – possibilities is why people call ‘should’ a weak necessity modal.
155
Data like these are discussed in Copley 2006 and Swanson forthcoming. My explanation is significantly
different from theirs. Thanks to Justin Snedegar for extensive discussions about these issues.
139
information of the context that does not entail p or its negation. Consider the set {a, b, b →
p}. This has two largest subsets which do not entail p: {a, b} and {a, b → p}. How are we to
choose? I don’t think this issue matters that much, but for precision let’s settle on the
following set: the set that results from intersecting all maximal subsets of the information of
the context which do not entail p or it’s negation. Let’s call this set the p-restricted
information of the context.
156
For the example just given, the p-restricted information of the
context is the set {a}.
Here is the complete statement of the view:
Should Contextualism: The direct speech act performed by a speaker who
assertively utters
┌
It should be that S
┐
, where p is the content of S in the context, is an
assertion of the proposition that the value that the credence function of the context
assigns to p conditional on the p-restricted information of the context is greater than
the value it assigns to p’s negation conditional on the p-restricted information of the
context.
This account allows the speaker to talk about his credence in a proposition putting aside his
knowledge that it is either true or false. This is useful in talking about what would probably
have happened if the normal course of events progressed rather than some unlikely event
occurring.
This is exactly what happens in (Should Package'). Though Angelica knows the
package isn’t there, she conveys, roughly, that absent that fact she would have thought it likely
that the package is there. This is as opposed to what she says in (Likely Package'). Here she
conveys both that the package is not there and that it is likely, given her information, that the
package is there. This lands her in a sort of Moorean inconsistency, since in order to
appropriately assert that the package is not there she must, at least, not think that it is likely
that it is there.
157
Since, in (Should Package'), Angelica only conveys a proposition about what
is likely given the p-restricted information of the context, there is no Moorean inconsistency.
So, it falls nicely out of Should Contextualism why (Should Package') is appropriate (and it
falls nicely out of Likely Contextualism, why (Likely Package') is not appropriate).
So far, we’ve just been discussing the direct speech act performed by a speaker who
156
In general the p-restricted information of the context is given by a two place function, f, that takes a context,
C, and a ‘should’-sentence:
f (C,
┌
It should be that S
┐
) = the intersection all of the maximal subsets of the information of C
which do not entail p or its negation (where p is the content of S in C).
157
There is similar Moorean inconsistency in the ‘might’ and ‘must’ cases mentioned above. This issue is
related to Yalcin’s puzzle. The case of ‘should’ offers a seeming counterexample to Yalcin’s account, which
implies that all epistemics yield the data from Yalcin’s puzzle (though Yalcin (pc) expressed doubt that
‘should’ is a pure epistemic). Thanks to Justin Snedegar for discussion.
140
uses an epistemic ‘should’-sentence. But, a complete account of ‘should’-communication will
also include the way in which utterances of ‘should’-sentences are typically used to indirectly
express attitudes. To motivate this account, let’s look at a familiar sort of conversation.
Suppose Susie and Angelica are wondering if the package has arrived and Lil is listening in:
(Package) Susie: Did your package get there yet?
Angelica: I’m not sure, but it should be there by now.
Lil: No. It shouldn’t be. I checked the tracking device and it was still
across the continent this morning.
We cannot explain Lil’s response as targeting the direct speech act performed by Angelica’s
utterance. Given the package-is-there-by-now-restricted information of the context – which is
just Angelica’s information – the value that the credence function assigns to the proposition
that the package is there is greater than the value it assigns to the proposition that it is not.
And, for all Lil knows, this is the case. So, it doesn’t make sense why she can respond
negatively if she is responding to the direct speech act.
As usual, the way to make sense of Lil’s response is as to an indirect speech act that
Angelica performs. Angelica indirectly expresses a likely float that the package is there by now:
EXPRESSING LIKELY FLOATSSHOULD: A speaker who utters a sentence of the
form
┌
It should be that S
┐
, where p is the content of S in the context, often indirectly
expresses, as a standardized speech act, a likely-degreed float in p.
In our case, Lil unlikely floats that the package is there by now. So she disagrees with
Angelica, given Evaluationism. And, given The Generalized Direct/Standardized Indirect
Approach, Lil can respond negatively.
As usual, we can see that Angelica’s expression of her likely float is a standardized,
indirect speech act. It is the speech act of primary importance. And not all uses of the
sentence she utters are expressions of likely floats. Consider, for example, if she had said, “It’s
not there yet, but it should be.” In such a case, Angelica does not express a likely float.
158

Finally, as we will see below, a hearer can figure out that Angelica performs this act of
expression based on the direct speech act performed and the features of the context. This
expression of a likely float is also standardized. The hearer can recover it without going
through any reasoning. And the act of expression is detachable. For example, suppose
instead that Angelica said, “My credence in the package being there, given the set of
information that includes my information except for the propositions that entail whether the
158
In general, these sorts of cases are ones where the speaker makes it clear that she is making the epistemic
claim relative to some body of information that does not include facts that entail whether or not the prejacent
is true. This relies on the specific features of ‘should’, as opposed to, say, ‘likely’, as will be discussed below.
141
package is there, is greater than my credence in the package not being there, given that set of
information.” Then she would not have expressed a likely float. Finally, as seen in (Package),
her speech act allows Lil to reply in a straightforward manner.
Hearers can figure out the indirect speech act performed by a speaker who uses a
‘should’-sentence in almost the same way they can for a ‘likely’-sentence. I’ll say a bit here
about how this could be, given that a ‘should’-sentence and the corresponding ‘likely’-sentence
are not, in general, used to assert the same proposition in the same context. The difference is
that ‘likely’-sentences are used to say something about what is likely given the information of
the context while ‘should’-sentences are used to say something about what is likely given the
prejacent-restricted information of the context. However, in the cases where a speaker
expresses a likely float by using a ‘should’-sentence, this difference is irrelevant, as we will see.
Consider Angelica’s utterance in (Package): “I’m not sure, but it should be there by
now.” With her use of the ‘should’-sentence, Angelica asserts that the value that the credence
function of the context assigns to the proposition that the package is there conditional on the
package-is-there-by-now-restricted information of the context is greater than the value it
assigns to the proposition that the package is not there conditional on the package-is-there-by-
now-restricted information of the context. But, in this situation, we can assume that the
package-is-there-by-now-restricted information of the context is identical to the information of
the context. If these were not identical, Angelica would be violating the maxim of quantity.
That is, if she knew either that the package was or wasn’t there, she should say so in order to
be cooperative. By using the weaker, ‘should’-sentence instead, she would not be making her
contribution as informative as required (as the maxim of quantity says).
159
So, we can
conclude that her information does not entail that the package is there nor does it entail that
the package isn’t there, and so that the package-is-there-by-now-restricted information of the
context just is her information. Because of this, Angelica’s utterance of the ‘should’-sentence
in this context is an assertion of the same proposition the corresponding ‘likely’-sentence
would be used to assert. Once a hearer realizes this, she can go through the same reasoning as
she would for an utterance of a ‘likely’-sentence.
160
This account explains why there isn’t a likely float expressed in (Should Package').
159
This is similar to the quantity implicature that is used to derive the some-but-not-all and most-but-not-all
readings of ‘some’ and ‘most’.
160
One issue is why we can’t also use something like quantity concerns to push away from a ‘should’-sentence
getting the same reading as a ‘likely’-sentence. For instance, why wouldn’t we think that if the package-is-
there-by-now-restricted information were just the information of the context, it would be more cooperative to
simply utter a ‘likely’-sentence. I think the answer to this issue is that ‘should’ is part of a modal auxiliary
scale with ‘might’ and ‘must’ (just like ‘most’ is on a scale with ‘some’ and ‘all’). And these quantity concerns
are much stronger in relation to linguistic items on the same scale.
142
Here, the speaker isn’t using a ‘should’-sentence in an attempt to help in reasoning. So, she
doesn’t express a likely float. This is good because we can’t give the same sort of derivation for
these cases that we just gave above. In cases where the speaker says that the package is not
there, the quantity concerns that make us think her ‘should’-sentence will get the same reading
as the corresponding ‘likely’-sentence are canceled. Without this reading, the reasoning that
allows a hearer to figure out that a float is expressed cannot be carried out.
In this section, I gave an account of the epistemic modal, ‘should’. This account
explains the ways in which ‘should’ is like ‘likely’ and the ways in which it isn’t. It also
explains how ‘should’ sentences are sometimes used to indirectly express likely floats, while
making sense of the interesting cases in which they aren’t used in this way.
4. Embedding
So far, I’ve only given accounts of uses of simple ‘must’-, ‘likely’-, and ‘should’-
sentences. For a complete account of epistemic modal communication, we need to extend the
accounts to uses of complex ‘must’-, ‘likely’-, and ‘should’-sentences. The strategy for doing
this is the same one followed in the last chapter. I’ll accept the natural account of the direct
speech act performed by a speaker who uses each type of complex epistemic sentence. Such
sentences are also standardly used to perform indirect speech acts – the ones that it makes
sense for speakers to perform given the direct speech act performed and the purposes of the
conversation.
I won’t try to give an account of the direct speech act performed by a speaker who uses
each type of complex epistemic sentence. Rather, I’ll just give the strategy for figuring out how
to form such an account and then illustrate how to carry out this strategy with a single
example. The strategy is to take the account of the direct speech act performed by a speaker
who uses a complex ‘might’-sentence and make the appropriate changes to reflect the different
strength of modality. So, instead of talking about consistency with a body of information, the
account for ‘must’ will talk about following from a body of information and the account for
‘likely’ will talk about being assigned a high value by the credence function given a body of
information.
Let’s figure out the account for the example of belief reports with ‘must’ in the
complement. Start with the account for belief reports with ‘might’ in the complement:
ContextualismBelief: The direct speech act performed by a speaker who assertively
utters
┌
A believes that it might be that S
┐
, where p is the content of S in the context
and o is the person denoted by A, is an assertion of the proposition that o believes that
143
p is consistent with the information of the context.
Now we make the appropriate changes so that we are dealing with following from a body of
information rather than being consistent with it.
Must ContextualismBelief: The direct speech act performed by a speaker who
assertively utters
┌
A believes that it must be that S
┐
, where p is the content of S in the
context and o is the person denoted by A, is an assertion of the proposition that o
believes that p follows from the information of the context.
This strategy will allow the reader to figure out the natural accounts of the direct speech act
performed by a speaker in uttering any complex epistemic sentence.
But giving an account of the direct speech act performed by complex epistemic
sentences does not constitute a complete account of communication involving uses of these
sentences. We also need to give accounts of the standardized, indirect speech acts that these
sentences are typically used to perform. This part of the account will also be similar to the
cases involving complex ‘might’-sentences. Again, we will alter the relevant attitudes in
accordance with the epistemic term used in the sentence.
161
For instance, we had:
EXPRESSING CONDITIONAL FLOATS: A speaker who utters a sentence of the
form
┌
If S then it might be that T
┐
, where p is the content of S and q is the content of
T in the context, typically indirectly expresses, as a standardized speech act, a
conditional float in q given p.
So, for a conditional ‘must’-sentence, we have:
EXPRESSING CONDITIONAL SINKS: A speaker who utters a sentence of the form
┌
If S then it must be that T
┐
, where p is the content of S and q is the content of T in
the context, typically indirectly expresses, as a standardized speech act, a conditional
sink in q’s negation given p.
This account can be used in explaining the same sorts of conversations and inferences that we
used the account involving conditional floats to explain. And the correct account of
embedding under ‘suppose’ and in the antecedent of a conditional will allow us the same sort
of explanation of Yalcin-style data involving ‘likely’ and ‘must’.
So, no complications are added by combining the sort of account I gave of complex
161
Here is a very brief survey of the prominent cases (without arguments or examples). In uttering negated
‘must’- or ‘likely’-sentences, a speaker typically indirectly expresses the contrary attitude – a float in the
negation of the prejacent of a ‘must’-sentence, an unlikely float in the prejacent of a ‘likely’-sentence, etc. For
conditionals, disjunctions, and quantifiers, a speaker typically indirectly expresses conditional or
quantificational sinks or likely floats. When a speaker reports an epistemic belief, he also typically indirectly
conveys that the subject sinks the proposition or floats it to the appropriate degree. Likewise, when these
terms are embedded under past operators, they are typically used to indirectly convey the subject’s prior sink
or float to the appropriate degree.
144
sentences in the last chapter with the account of other strengths of modality that I gave in this
chapter. This combination allows us to explain how complex sentences containing these other
epistemics are used in conversations and pieces of reasoning.
In the previous two chapters I gave an account of communication involving uses of
epistemic ‘might’-sentences. In this chapter, I have extended this account to explain
communication involving uses of other epistemic sentences – with a focus on ‘must’-, ‘likely’-,
and ‘should’-sentences. Together, these chapters constitute my theory of epistemic
communication. In the next, and final, chapter my primary concern is not communication.
Rather, I will fill out the picture of the mind in order to accompany this account of ‘might’-
communication. The chapter will be an investigation into what float-like attitudes are.
145
Chapter Seven: Floats, Credences, and Acceptance in an Inquiry
In this Chapter, I investigate whether we can reduce float-like attitudes to other sorts of
attitudes that are already part of our ontology of the mind. As we saw in Chapter Three,
floating a proposition is not a matter of having a belief in that proposition or a belief in its
negation. Floating is characteristic of situations of uncertainty. Many accounts of the mind
already countenance attitudes that are intimately connected to uncertainty. These accounts
realize that an agent can be more or less confident of a proposition. We can call attitudes of
confidence credences. Credences are partial, belief-like attitudes whose strengths are
traditionally represented as real numbers in the interval [0,1]. Since both floats and credences
involve partial, but (typically) not total confidence in the truth of a proposition, we might
think they are related. So, one central project of this chapter is to investigate the relationship
between floats and credences.
I’ll first investigate whether an agent’s floats are reducible to some subset of her
credences. After finding problems for each salient way of cashing this out, I will conclude that
it’s not the case that an agent’s floats are reducible to her credences or some subset of her
credences. Nonetheless, I will argue that floats and credences are closely related – that floating
is related to having high enough credence in the way that the attitude of accepting in an
inquiry is related to having full credence. This leads to a characterization of floats as partial
versions of acceptance in an inquiry. In the literature, acceptance is thought to be a practical
attitude – in the sense that some reasons of the right kind for accepting a proposition are
practical rather than evidential. So, I will characterize the sense in which floating, as a partial
version of acceptance, is practical by showing which sorts of non-evidential reasons are of the
right kind to count in favor of floating and which aren’t. Finally, I will revisit our hypotheses
from Chapter Three, about evaluation and disagreement, in light of the new conclusion that
float-like attitudes are practical.
1. Reduction to Credences
In this section, I will entertain and reject a number of views for reducing floats to
credences, progressing from simple to more plausible proposals. I will evaluate the proposals
based on whether their predictions match our judgments about how agents reason in
particular examples – specifically, what they take as options in reasoning. This method will
146
show that the salient promising candidates for a reduction fail.
The simplest account of the relationship between floats and credences is that floats are
positive credences:
REDUCTION TO CREDENCES: Floating p to degree x is having a positive credence of
degree x in p.
162
REDUCTION TO CREDENCES is pretty obviously incorrect. When an agent reasons, she doesn’t
take as options all of the propositions in which she has positive credence. Consider our old
example of Phil reasoning about who stole the cookies. Phil does not take it as an option that
Brad Pitt is the thief. He does not reflect on this proposition in his reasoning or attempt to
find evidence regarding its truth. It simply doesn’t play a role when he is thinking about who
is the thief. This is in contrast to the propositions that Tommy is the thief, that Chuckie is
the thief, and that Angelica is the thief. Nonetheless, Phil does have a positive (albeit very
low) credence that Brad Pitt is the thief. So, an agent’s floats can’t just be her positive
credences.
163
We can avoid this problem while retaining the spirit of REDUCTION TO CREDENCES by
putting a restriction on the positive credences that count as floats. In our case, it seems that
the propositions that Phil floats are those in which he has a relatively high credence. He has
positive credence that Brad Pitt is the thief, but his credence in this proposition is relatively
low. As it happens, he does not float this proposition. This suggests that, as Malte Willer
(2013, 52) says:
[I]n at least many cases agents decide which possibilities to take seriously on the basis
of general considerations of plausibility. Some possibilities are really far-fetched, while
others seem to be more reasonable, and in general only possibilities that meet a certain
standard of plausibility play a significant role in one’s practical and theoretical
deliberation.
So, floating a proposition seems to be correlated with having sufficiently high credence in a
proposition (rather than just having any positive credence).
We can formalize this by introducing a threshold that an agent’s credences must pass
162
The following suggestion (whatever other theoretical virtues and vices it has) fails for the same reason as
REDUCTION TO CREDENCES:
BELIEF SUPPORT REDUCTION: Floating p to degree x is believing that p is supported by one’s
information to degree x.
163
Richard Holton (2008) gives a similar example involving practical, rather than theoretical, reasoning. I will
focus, here, on theoretical reasoning, though I hope something like the view I advocate is true of the practical
case as well.
147
in order to count as floats:
REDUCTION TO CREDENCES OVER THRESHOLD: Floating p to degree x is having a
credence of degree x in p, where x is above the threshold.
The threshold need not be fixed. It is plausible that it will vary across different inquiries.
Sometimes unlikely propositions are taken as options in reasoning while other times they are
not. This is not a problem for REDUCTION TO CREDENCES OVER THRESHOLD.
REDUCTION TO CREDENCES OVER THRESHOLD says that, for some inquiry, an agent’s
floats are a subset of her credences. An initially attractive feature of the view is that this subset
is formally characterizable. Which credences are in the subset depends solely on the strength
of those credences. But, while this makes for a theoretically neat package, it gets the facts
wrong regarding which propositions an agent takes as options in reasoning. Factors other than
confidence can come into conflict with and trump confidence in determining what an agent
floats.
164
By showing this, I will show that REDUCTION TO CREDENCES OVER THRESHOLD –
and any threshold-based reduction – is false. It is important to note that our original thesis,
REDUCTION TO CREDENCES, is a threshold-based reduction as well – where the threshold that
must be surpassed is 0. So, the following scenarios serve as counterexamples to REDUCTION
TO CREDENCES, as well.
The first example shows that it is possible for an agent to not float a proposition about
which she has positive credence because of distraction or impairment. Suppose Phil has a .1
credence that Tommy is the thief. But suppose that he has to leave in 15 seconds to pick up
his nephew from preschool. In thinking about who stole the cookies, the possibility that
Tommy is the thief simply does not – given his hurry and other distractions – occur to Phil.
In such a case, he has some confidence in the proposition that Tommy is the thief, but he
does not float that proposition. This case seems possible, but REDUCTION TO CREDENCES says
it is impossible. To be a counterexample to REDUCTION TO CREDENCES OVER THRESHOLD, we
just need to slightly amend the case so that Phil’s credence that Tommy is the thief is strong
enough to surpass the threshold. This sort of scenario serves as a counterexample to
164
We might think that an agent doesn’t take as options all propositions in which she has a credence over the
threshold, but only those that meet that condition and are relevant to the inquiry. This could lead us to the
following amendment to REDUCTION TO CREDENCES OVER THRESHOLD:
REDUCTION TO RELEV ANT CREDENCES OVER THRESHOLD: Floating p to degree x is having a credence
of degree x in p, where p is relevant and x is above the threshold.
Given the right characterization of relevance, REDUCTION TO RELEV ANT CREDENCES OVER THRESHOLD will
avoid the issue just mentioned. However, this move is not particularly interesting for our purposes, so I will
ignore it in the text. The counterexamples that I present in the text apply just as well to REDUCTION TO
RELEV ANT CREDENCES OVER THRESHOLD.
148
REDUCTION TO CREDENCES and REDUCTION TO CREDENCES OVER THRESHOLD.
We could have a variation on this scenario, which shows that it is possible for an agent
to choose not to float a proposition in which she is somewhat confident. Suppose that Phil is
thinking about the cookie theft and the proposition that Tommy is the thief does come to
mind. But, in this variation, Phil simply chooses to deliberate without taking it as an option
in his reasoning. He may be irrational in such a case. Whether or not he is irrational, such a
scenario seems possible, but, again, both REDUCTION TO CREDENCES and REDUCTION TO
CREDENCES OVER THRESHOLD say this is impossible.
Each of these examples suffices to show that REDUCTION TO CREDENCES and
REDUCTION TO CREDENCES OVER THRESHOLD are false. But, the first involves an agent who is
not totally reflective and the second involves an agent who is irrational. We might wonder
whether there are examples that witness the failure of our theses in which agents are both
rational and fully reflective. Indeed, there are cases where a rational, reflective agent’s floats
come apart from her positive credences (which fall over the threshold). I will give two
examples, both of which show that, for certain sorts of practical reasons, an agent may float
propositions in which she has very little or no confidence. It’s worth keeping in mind,
however, that the cases just given are enough to show that REDUCTION TO CREDENCES and
REDUCTION TO CREDENCES OVER THRESHOLD, which are metaphysical theses and do not
depend on the reflectiveness or rationality of the agent, are false.
Suppose Phil is inquiring with his friends into who is the thief. The case is difficult, so
none of them think they can solve it on their own, though they hope that they can together.
All of the members other than Phil think it is somewhat likely that Hillary Clinton is the thief.
Phil thinks it is not at all likely that she is. He has zero credence (which is, of course, below
whatever threshold might be suggested for floating) that Clinton is the thief. Nonetheless, the
group is adamant that Phil take this proposition as an option in reasoning, so that he can be
on the same page as the group in reasoning. For the sake of the inquiry, Phil floats that
Clinton is the thief. Both REDUCTION TO CREDENCES and REDUCTION TO CREDENCES OVER
THRESHOLD say this is impossible. The strength of Phil’s credence is not even partially
determining his float. Furthermore, this is a scenario where Phil is rational and can
adequately reflect on the proposition in question.
165
165
This case is not uncontroversial. One might respond that Phil, himself, doesn’t float that Clinton is the thief,
but the group does. However, the onus here is on the objector to say more about how the group float relates
to the individual floats and to show that there is a corresponding group credence (without which, this move
would not count as a defense of REDUCTION TO CREDENCES OVER THRESHOLD). Such a move would have to
149
Here is a different sort of case that suggests the same conclusion – that a rational,
reflective agent’s floats can come apart from her positive credences (which fall over a
threshold). Suppose that Lil is doing her astronomy homework. She looks at the first
problem and notices that she has seen this sort of problem before. She remembers some
information about the sort of problem – in particular, that some of the astronomical theories
studied in class get the right answer to this problem and some don’t; however, all of the ones
that do not yield distinct answers. She hasn’t studied hard enough, so she only has the math
skills to do the calculations with two theories – Galilean astronomy and Copernican
astronomy (Einsteinian astronomy, for example, would be too hard for her). So, in trying to
solve the problem, Lil takes Galilean astronomy as an option and takes Copernican astronomy
as an option. She solves the problem with the assumptions of the former and then with the
assumptions of the latter. If the solutions match up, she will know that she has the right
answer. Otherwise, she will need to do some more investigating. Despite this, Lil is certain
that both Galilean astronomy and Copernican astronomy are incorrect. She has no credence
in the truth of either theory. She merely takes them as options in solving the problem. So,
this example shows that REDUCTION TO CREDENCES and REDUCTION TO CREDENCES OVER
THRESHOLD are false. And this case, like the last, seems like one in which the agent is rational
and properly reflective.
166
All of these examples show that an agent’s positive credences (which exceed the
threshold) can come apart from her floats. Other concerns can outweigh strength of
confidence in determining which propositions she takes or rules out as options in reasoning.
For this reason, both REDUCTION TO CREDENCES and REDUCTION TO CREDENCES OVER
THRESHOLD are false. It seems that the only way to see what an agent floats is directly – by
seeing what propositions she takes as options in her reasoning.
We might think that this shows that an agent’s floats are not just some subset of her
credences. This, however, is a bit too hasty. We have shown that an agent’s floats are not a
subset which is determined by the strength of the credences. But this doesn’t mean that her
respect the many constraints for aggregating individual credences into group credences (see Genest and Zidek
1986 and Brössel and Eder forthcoming for a survey of these issues). Alternatively, one might respond that
Phil doesn’t actually float the proposition, but merely pretends to. But it is difficult to square the fact that
the conclusion the group ends up reaching is a genuine answer in the inquiry with the suggestion that the
reasoners just pretended to have the attitudes used in reasoning. For more on this idea in the context of
acceptance, see note 170.
166
Thanks to Shyam Nair for suggesting this sort of case. This case is, like the last, not uncontroversial, though
for different reasons. We might think that Lil floats not the Galilean theory of astronomy but, rather, that
the Galilean theory yields the right answer to this sort of problem. Thanks to Mark Schroeder for discussion.
150
floats aren’t some other subset of her credences. Which one? Well, the subset that includes
the credences whose objects are propositions that the agent takes seriously in reasoning:
REDUCTION TO CREDENCES TAKEN SERIOUSLY: Floating p to degree x is having a
credence of degree x in p, where that credence plays a role in the agent’s inquiry.
167
The notion of playing a role in inquiry that we are concerned with is one that includes
influencing how an agent structures her reasoning, searches for evidence, weighs how to
allocate resources, etc. It isn’t important for our purposes to say much more than this. What
is important is that REDUCTION TO CREDENCES TAKEN SERIOUSLY is still a view on which
floats are credences.
One might complain that REDUCTION TO CREDENCES TAKEN SERIOUSLY is an
uninformative analysis of floats. It merely tells us that an agent’s floats are the credences that
play a role in her inquiry. On this view, we can’t tell which propositions an agent floats by
looking at the structural properties (like the strengths) of her credences. However, the
problems for REDUCTION TO CREDENCES and REDUCTION TO CREDENCES OVER THRESHOLD
arose because those theories gave too central a place to these structural features. So, it is a
virtue that REDUCTION TO CREDENCES TAKEN SERIOUSLY does not focus merely on the
strength of the credence.
Though REDUCTION TO CREDENCES TAKEN SERIOUSLY is not informative to a
practitioner trying to figure out what propositions an agent floats from information about her
credences, it is informative in another sense. It is very informative to the theorist of the mind
who is constructing a theory of the sorts of attitudes agents have. It tells us that we don’t need
to posit floats as basic attitudes. They can still be credences. So, while REDUCTION TO
CREDENCES TAKEN SERIOUSLY offers us no algorithmic way of determining whether, for a
particular inquiry, an agent floats some proposition, it does tell us something interesting about
the sort of attitude a float is – namely, a credence.
REDUCTION TO CREDENCES TAKEN SERIOUSLY seems to avoid the problems for the
views discussed so far, since floats are defined to be the credences that play a role in reasoning.
However, REDUCTION TO CREDENCES TAKEN SERIOUSLY is incorrect for the following reason:
It does not leave room for the possibility of floating a proposition to one degree while having a
credence in it to a different degree. This is because REDUCTION TO CREDENCES TAKEN
SERIOUSLY says that to float a proposition to a particular degree is to have a credence in that
167
Thanks to Julia Staffel (pc) for suggesting this view.
151
proposition to that degree, which plays a role in reasoning. But, agents can float a proposition
to one degree while having a credence of some different degree in that proposition.
There are versions many of the scenarios from above where this happens. For
instance, suppose that Phil is reasoning about who stole the cookies, but is in a hurry. Above
we supposed that Phil, in his hurry, did not take it as an option that Tommy is the thief, even
though Phil had a positive credence that he is. Likewise, it seems that Phil could, in his hurry,
take it as a 25% likely option that Tommy is the thief, even though Phil is really only 10%
confident that he is. The distraction could cause him to ignore the proposition altogether, as
in the above case, or, as I have suggested here, it could lead him to treat the proposition as
likely to a different degree than his credence. REDUCTION TO CREDENCES TAKEN SERIOUSLY
says this latter scenario is impossible, but it seems possible.
The group reasoning case from above requires no amendment at all to be a
counterexample to REDUCTION TO CREDENCES TAKEN SERIOUSLY. Recall that in that
scenario, Phil has zero credence that Hillary Clinton is the thief, but the members of his
group want him to float that she is, so that they can all be on the same page in reasoning. Phil
cooperates, making him have a credence of zero strength and a float of positive strength in the
same proposition – that Clinton is the thief. But, again, REDUCTION TO CREDENCES TAKEN
SERIOUSLY does not allow for this.
In general, the degree to which an agent floats a proposition can come apart from the
agent’s degree of credence in that proposition. But REDUCTION TO CREDENCES TAKEN
SERIOUSLY requires that, if an agent floats a proposition, she floats it to the degree of her
credence in that proposition. So, REDUCTION TO CREDENCES TAKEN SERIOUSLY is false.
There are two natural reactions for someone with a reductionist mindset to have to the
problems for REDUCTION TO CREDENCES TAKEN SERIOUSLY. The first is to wonder whether
we could have a view according to which a credence of one strength is a float of a different
strength. However, such a view is implausible, since the float and the credence have different
modal profiles – an agent could have that float without having that credence and vice-versa.
Given this divergence, it is very difficult to see in what sense the credence is the float on the
suggested account. So, I think this suggestion is a non-starter.
The other natural reaction involves realizing that all of the reductions considered so far
are ones according to which each individual float can be reduced to some individual credence.
But maybe the problems we raised for this sort of view don’t arise when we look at the
152
reduction as one of an entire float state (all of an agent’s float-like attitudes) to an entire credal
state. Such a reduction could be successful even if we don’t have a reduction of each particular
float-like attitude to a particular credal one:
REDUCTION TO ENTIRE CREDAL STATE: An agent’s entire float state is reducible to her
entire credal state.
This view looks promising because it does not impose requirements on the strength of an
agent’s credence in a particular proposition in order for that agent to float that proposition to
some degree. For that reason, none of the counterexamples that we have seen so far can, in
their current form, be pressed against REDUCTION TO ENTIRE CREDAL STATE, since they
merely cause trouble for views that place restrictions on what strengths of credences count as
what strengths of floats.
Unfortunately, amending the above examples to be about diachronic situations does
cause problems for this view. And the reason is much the same as above. Float states, just
like individual floats, can vary despite a credal state remaining constant. Consider, for
example, the case of distraction or impairment. This was a case where Phil did not float that
Tommy is the thief because that proposition didn’t come to his mind. Suppose now,
however, that the proposition that Tommy is the thief comes to Phil’s mind a few seconds
later, and he then floats that Tommy is the thief. This can happen even if all of his credences
remain the same. So, it is a counterexample to REDUCTION TO ENTIRE CREDAL STATE.
168

There are similar scenarios involving the other sorts of cases that were problems – where we
look at evolutions of float states that happen in the presence of unchanging credal states.
Thus, float states do not supervene on and are not reducible to credal states. An agent’s floats
can vary even when her credences do not.
2. An Analogy to Acceptance and Belief
Our results to this point suggest that there is no plausible way to maintain that floats
are reducible credences while allowing floats to fulfill their key feature – being the attitudes
168
We might think that Phil’s credal state has changed. He has a higher credence now that the proposition that
Tommy is the thief has come to mind. So, we might think that the change in float state is accompanied by a
change in credal state. However, we should realize that it need not be. Phil could, even when the
proposition that Tommy is the thief comes to mind, continue to not float that proposition. Since what
agents float is up to them in this sense, this scenario is possible. And it constitutes a counterexample to
REDUCTION TO ENTIRE CREDAL STATE – one which persists in the face of this suggested reply to my original
counterexample.
153
that agents have toward propositions that they take as options in reasoning. Nonetheless,
there is an interesting and highly informative relationship between the attitudes, which we can
understand by analogy. The analogy involves the attitude of acceptance in an inquiry.
Before getting to the analogy, let me say a few words about the attitude of acceptance in
an inquiry. The notion of acceptance that I am concerned with is the notion that authors like
Michael Bratman (1992) and L. Jonathan Cohen (1989) are after. To accept a proposition in
an inquiry is to take it for granted (Bratman 1992) or to take it as a premise (Cohen 1989) in an
inquiry. Later, I’ll say more about its properties, but it may be helpful to look at a couple of
simple examples. A common case of acceptance is when I believe a proposition, say, that the
meeting starts at noon on Thursday, and also accept it in my reasoning about where I should
be on Thursday afternoon. In such a case, I both believe this proposition and accept it in my
inquiry. But, if acceptance always coincided with belief, it would be eliminable in a picture of
the mind. The reason acceptance garners any attention at all is that it can come apart from
belief. For instance, Cohen (1989, 369) mentions the following example: “for professional
purposes a lawyer might accept that his client is not guilty even though he does not believe it.”
Both Cohen (1989, 369) and Bratman (1992, 8) suggest that an agent might accept a
proposition because of one’s close friendship with a person who claims that the proposition is
true, even though the evidence (including that provided by the friend’s testimony) does not
favor believing such a proposition. We will see further examples of acceptance throughout
this section.
I want to contrast what theorists concerned with reasoning in a context, like Bratman
and Cohen, say about acceptance in an inquiry with what is sometimes said about theory
acceptance in the philosophy of science literature. For example, van Fraassen (1980, 12) says
that acceptance of a theory involves a belief “that it is empirically adequate.” And he (4) says
that theory acceptance also involves, “a commitment to a research programme”.
169
I think this
characterization of acceptance as involving this belief and this commitment is too narrow. As
shown in the examples above, we can accept single propositions – not just entire theories.
Furthermore, as we will see, what an agent accepts can vary across inquiries – something that
is not allowed given van Fraassen’s claim that acceptance involves a commitment to a research
program (a commitment he means to apply to future encounters with new data). Nonetheless,
theory acceptance may just be a type of acceptance in the wider sense that is of interest to us.
169
Maher (1990) gives a competing account that, while recognizing that acceptance of a theory is not the same as
belief in it, rejects both of van Fraassen’s necessary conditions on theory acceptance.
154
Indeed, one good reason for accepting a scientific theory for the purpose of answering some
question is the belief that it will explain the phenomena required for answering that question.
So, I want to leave it open that my account applies to theory acceptance, as discussed by van
Fraassen, though I don’t want to limit my discussion of acceptance to his circumscribed
notion.
What I want to do now is compare the relationship between floats and credences to
that between acceptance and belief. I’ll start by comparing the problem cases for REDUCTION
TO CREDENCES and REDUCTION TO CREDENCES OVER THRESHOLD with four cases, which are
counterexamples to an analogous hypothesis to the ones we’ve been considering – that
accepting a proposition in an inquiry is believing that proposition.
First, consider a scenario where you are reasoning about what ingredients you need to
buy for the dinner you will make tonight, but you have to leave in 15 seconds to pick your
nephew up from preschool. Your partner told you that she bought some tofu yesterday, so
you believe that you have tofu. However, given your time constraints, this doesn’t come to
mind. So you quickly jot down ‘tofu’ on your shopping list and run out to pick up your
nephew. This is a case where you believe that you have tofu, but you do not accept it in your
inquiry. It looks quite a lot like the case where Phil has positive credence that Tommy is the
thief, but Phil doesn’t float that he is, because it doesn’t come to mind at the time. Both are
examples where, because of temporal or cognitive limitations, an agent doesn’t float or accept
some proposition that she has positive credence in or believes.
We could also amend the scenario so that you do remember that there is tofu in the
fridge, but you simply reason about what is needed for the recipe without accepting this fact.
This may be irrational, but it seems possible. This case looks quite a lot like the one where
Phil has positive credence that Tommy is the thief and this proposition comes to mind, but
he chooses not to take it as an option in reasoning.
Just as in the case of floats and credences, these first two cases are enough to show that
acceptance and belief come apart. But, again, we can see that there are examples where the set
of propositions that a rational, reflective agent accepts are not identical to the set of
propositions that she believes. As Stalnaker (1984, 93) says, these are cases where “[a]ccepting
a certain false proposition may greatly simplify an inquiry, or even make possible an inquiry
not otherwise possible, while at the same time it is known that the difference between what is
accepted and the truth will have no significant effect on the answer to the particular question
155
being asked.” These cases also mirror our cases from above.
Consider a scenario where you and your partner are on the way home from work. You
are talking about what you need to buy at the grocery store. She thinks that you have tofu in
the fridge, but you think that you saw her accidentally buy tempeh instead of tofu. She insists
that it is tofu. You concede for reasons of expediency and accept in the inquiry that you have
tofu, though you do not believe it. In such a case, you seem neither irrational nor non-
reflective. This case looks a lot like the one where Phil has zero credence that Hillary Clinton
is the thief, but the group requires that he take that proposition as an option in reasoning and,
for reasons of practical expediency, he does.
170
Finally, consider the following scenario. You are figuring out the proper height for the
suspension towers for a bridge that you are building. You do not believe that Newtonian
physics is correct, but you accept Newtonian physics in the context, because you know that it
will yield the right answer about tower height. This, too, is a case where you are rational and
are reflecting properly. Furthermore, it looks like the case from above where Lil floats some
astronomical theories in which she has zero credence, in order to efficiently get the right
answer to her homework problem.
So, there are belief/acceptance versions of the problem cases for REDUCTION TO
CREDENCES and REDUCTION TO CREDENCES OVER THRESHOLD. The role played in those cases
by belief is analogous to the role played in the original cases by positive credence, and the role
played in those cases by acceptance in an inquiry is analogous to the role played in the
original cases by floats. Generally, the propositions an agent accepts in some inquiry are
propositions that the agent believes. But sometimes she accepts other propositions.
171

170
In note 165, I raised the possible worry that Phil does not float that Clinton is the thief, but merely pretends
to. The same worry may arise here. We may think that you merely pretend to accept that you have tofu.
Bratman (1992, 9) echoes my response above, noting the connection between accepting and genuine answers
in inquiry:
Is such context-relative acceptance mere pretence? I do not think it is. In accepting that p I do not
simply behave as if I think that p: I also reason on the assumption that p. So there is not the kind
of indirect, circuitous connection between reasoning and action that is characteristic of pretence.
He continues to address the related hypothesis that accepting is just supposing:
Nor is context-relative acceptance mere supposition. “Suppose I had a million dollars”, I ask myself.
“What should I do with it?” Such a question may trigger contingency planning based on the mere
supposition that I have such wealth. But this planning will not directly shape my action. If I
conclude, for example, that with such wealth I should invest in General Motors my conclusion will
not lead directly to my calling up my broker. In contrast, [what] I accept (in the context) . . . directly
shapes my subsequent conduct. Context-relative acceptance is tied more directly to action than is
mere supposition; and it is tied more directly to practical reasoning than is mere pretence.
171
See Bratman 1992 for a number of scenarios in which this is the case. Clarke (1994) thinks that acceptance
can come apart from belief only in the other direction – where an agent believes something but does not
accept it. This position seems to spring from his using ‘acceptance’ in a slightly different way than people
156
Likewise, generally, the propositions that an agent floats to some degree are those that she has
a credence in to that degree. But sometimes she floats propositions to different degrees. This
suggests the following analogy:
Analogy
(first pass)
: Floating is related to having positive credence in the way that accepting
in an inquiry is related to believing.
The plausibility of the analogy is even easier to see if we use the ordinary language expression
to talk about floating – “taking as an option” – and Bratman’s (1992) common expression for
talking about accepting – “taking for granted”: Taking as an option is to having positive
credence as taking for granted is to believing. This strikes me as extremely natural.
The analogical view allows us to explain the sort of case that was problematic for
REDUCTION TO CREDENCES TAKEN SERIOUSLY – where an agent floats a proposition to some
degree in an inquiry even though she has a credence of a different degree in that proposition.
We do so in the same way that we explain an agent who accepts in some inquiry a proposition
that she does not believe, since for each pair, the attitudes, though related, are distinct.
Analogy
(first pass)
is supported by the fact that floating has many of the characteristic
features of accepting in an inquiry and having a positive credence has many of the
characteristic features of believing. Bratman (1992, 9) contrasts believing and accepting as
follows:
Belief has four characteristic features: (a) it is . . . context-independent; (b) it aims at the
truth of what is believed; (c) it is not normally in our direct voluntary control; and (d) it
is subject to an ideal of agglomeration. In contrast, what one accepts/takes for granted
(a) can reasonably vary . . . across contexts; (b) can be influenced by practical
considerations that are not themselves evidence for the truth of what is accepted; (c)
can be subject to our direct voluntary control; and (d) is not subject to the same ideal of
agglomeration across contexts. So acceptance in a context is not belief.
172
Our examples from section 3 show that floats and credences share the same contrasts on
features (a), (b), and (c) (note that (d) doesn’t apply since even credences aren’t subject to an
like Cohen (1989) and Bratman (1992). For Clarke, acceptance is like a mental analogue of assent –
something closer to judging than taking for granted in an inquiry.
172
Cohen agrees with Bratman on (a) – (c). Cohen (1989, 368) talks, regarding (a), about accepting a proposition
“either for the long term or for immediate purposes only”. And he (1989, 369) agrees with Bratman on (b):
“the reasons for accepting that p need not always be epistemic ones: they might be ethical or prudential.”
Regarding (c), Cohen (1989, 369-370) says, “But acceptance that p, in the relevant sense of ‘acceptance’ is
voluntary. It is decidable at will, while belief that p is not”. See also Alston 1996, Engel 1999, Stalnaker 1984
(Ch. 5), and Clarke 1994 for agreement on some of these points, though there is variation in how, exactly,
these authors conceive of acceptance.
157
ideal of agglomeration). First, what an agent floats can vary with the context while her
credences cannot. Second, floats are properly answerable to some practical, as well as
evidential, concerns, while credences are answerable only to evidential ones. Third, floats are
semi-voluntary (we can decide what to take as an option in reasoning) while credences are
not.
173
These similarities are quite telling, but Analogy
(first pass)
, as we have stated it, isn’t quite
right. Looked at in one way, Analogy
(first pass)
tells us about the relationship between floating
and having positive credence. But, it also makes predictions about the relationship between
floating and accepting in an inquiry. In particular, it predicts that floating bears the same
relationship to accepting in an inquiry that having positive credence bears to believing.
Unfortunately, this prediction is false.
The simplest way to see this is by looking at the rational co-tenability of various
combinations of attitudes. The following is a natural and very popular (though not universally
accepted) position: an agent can rationally believe a proposition without being subjectively
certain of that proposition. That is, an agent can rationally believe a proposition while still
having a nonzero credence in its negation. For instance, I can rationally believe that it will be
sunny tomorrow in Phoenix while still having a positive (if small) credence in the proposition
that it will not be sunny tomorrow in Phoenix. Unfortunately, this same relationship does not
hold of acceptance and floating. If an agent accepts a proposition, it is irrational for her to
take its negation as an option in reasoning. For instance, if I accept in an inquiry that there is
tofu in the fridge, then I am irrational if I also float in that inquiry that there is not tofu in the
fridge. So, the relationship between acceptance and floating is different than the relationship
between belief and credence, and Analogy
(first pass)
is false.
Analogy
(first pass)
is not, however, far off the mark. Often, what I have been calling
credence is called partial belief and what I have been calling belief is called full belief. Such
talk is misleading because it makes it look like one is just a partial version of the other. Given
the point just made, it is clear that our ordinary notion of belief is not just full, maximal, or
total credence. But it is this attitude – full credence – that has the properties that we are
looking for. Focusing on the relationship between partial and full credence (rather than partial
173
Floats are in some ways similar to and in some ways different from what Wedgwood (2012) calls ‘practical
credences’. Though both floats and practical credences share the property of varying across contexts,
Wedgwood’s practical credences are, as he (2012, 325) says, “rationally required to be guided purely by
certain distinctively epistemic values” (which I take not to include considerations of advancing the inquiry).
He contrasts practical credences with states of acceptance (drawing on Bratman’s characterization) with
respect to this feature.
158
credence and belief) remedies the disanalogy from above. For it is irrational for an agent to
both have full credence in a proposition and have partial credence in its negation.
Nonetheless, full credence still contrasts with acceptance in exactly the ways that Bratman
mentions in the above quotation (Bratman (1992, 2-4) explicitly notes that these properties
apply to credence as well). Full credence is context-independent, properly responsive only to
evidence, and non-voluntary, while acceptance has none of those features.
174
So, the correct
analogical characterization is the following:
Analogy: Floating is related to having positive, partial credence in the way that
accepting in an inquiry is related to having full credence.
3. Floats as Partial Versions of Acceptance
Analogy paints a pretty specific picture of floats. The picture looks like this: Just as
there is a scale of credence and an agent can have partial or full credence, there is a scale of
acceptance and an agent can partially or fully accept a proposition. Floats, then, are just
partial versions of acceptance.
Here’s a more detailed way to think about the picture. Agents have attitudes of treating
propositions as likely to particular degrees in reasoning (as was shown in §1, these attitudes do
not just amount to credences). Treating a proposition as fully, maximally, or totally likely in
reasoning is what is often called accepting that proposition (what I sometimes call fully
accepting that proposition). But agents can treat propositions as less than fully likely in their
reasoning. Treating a proposition as less than fully likely is what I have called floating that
proposition. So, accepting and floating are states of the same general kind – the former being a
full strength version and the latter being a partial version. This is just as full and partial
credence are states of the same general kind – again, the former being a full strength version
and the latter being a partial version.
175
The idea that acceptance admits of degrees flies in the face of much of the literature on
the relationship between acceptance and belief.
176
For instance, D.S. Clarke (1994, 146) says, “a
belief state admits of degrees, while acceptance is all-or-nothing.” And Pascal Engel (1999,
174
Indeed, it does better than belief with respect to the second feature. The issue of pragmatic encroachment
suggests that qualitative belief – as opposed to credence – is properly answerable to some pragmatic
considerations. See Hawthorne 2004, Stanley 2005b, Fantl and McGrath 2009, and Schroeder 2012a.
175
To extend the analogy to bare floats, we need a picture of credences that don’t have real-value strengths.
Such a picture of imprecise credences is fairly popular. See Levi 1980, Jeffrey 1983, and Sturgeon 2008,
among others.
176
Bratman (1992), however, is silent on this issue. And it is often explicitly allowed that acceptance comes in
degrees in the philosophy of science literature – see, for instance, van Fraassen 1980 (9) and Maher 1990.
159
219-20) echoes, “Acceptance is qualitative or categorical, whereas belief is graded and subject
to degrees.” Cohen (1989, 374) agrees, saying, “acceptance that p (or that q) is not itself a
matter of degree in the way that a person’s belief that p may be stronger than his belief that q.
To accept that p is to adopt the policy of taking the proposition that p as a premiss in
appropriate circumstances, and you either adopt that policy or you do not.” These authors
seem to run together the notions of qualitative belief and credence when they talk about belief
being graded (as we’ve seen, it is the latter, not the former that is graded). But what is
important is that they deny that acceptance is graded in any sense (and, therefore, in the way
that credence is). But, the literature simply got this wrong. Just as an agent can take a
proposition for granted (or, in Cohen’s words, take it as a premise) in an inquiry, she can take
a proposition as an option in an inquiry. That is, just as there are attitudes of full acceptance,
there are attitudes of partial acceptance, or floats.
We might wonder why we have two similar-looking mental systems – a credal system
and what I will call an acceptance system (which is the system that includes floats and attitudes
of full acceptance). Here is an idea that I have adapted slightly from Bratman 1992 (10-11) to
apply to our present scenario. An agent’s credal state is what Bratman calls the default
cognitive background against which reasoning takes place. But in particular circumstances, the
agent may adjust her default background in various ways. The most typical, in our case, is to
only reason using a subset of the propositions in which she has positive credence – the
relevant propositions in which she has relatively high credence. But it may also happen that
she adjusts this background for other reasons. She might do so for practical reasons – as in the
group reasoning and astronomy problem cases – for no good reason at all – as in the choice
case – or inadvertently – as in the distraction case. However she comes to have it, this
adjusted state, which she actually puts to work in her reasoning in some context is her
acceptance state. So, the credal system encodes an agent’s confidences about various
propositions and provides the background against which adjustments are made for reasoning
in particular circumstances. And the state that results from the adjustment, which she uses in
reasoning in those circumstances, is the acceptance state.
There is, however, a rather different way to look at the relationship between credal
states and acceptance states. On Bratman’s picture, acceptance states come about from
starting with the credal state and amending it for particular purposes in response to various
concerns. We might, on the other hand, think of the acceptance state as prior. Stephen
160
Finlay (pc) has suggested the following sort of account. As agents, we treat propositions in
particular ways in reasoning; we accept propositions to different degrees in reasoning. Our
credal state arises from our dispositions to accept (partially or fully) propositions in ordinary
circumstances.
177
According to a simple version of this view, a credence of .5 in the
proposition that Angelica is the thief is a disposition to float (i.e. partially accept) that
proposition to degree .5.
I won’t decide between these two attractive pictures here. Both carve out an important
role for float/acceptance attitudes and present a plausible account of their relationship to
credal attitudes. And both respect our main conclusion – that floating a proposition is
partially accepting that proposition in an inquiry.
4. Floats as Practical Attitudes
Acceptance is thought to be a practical attitude used in inquiry. The characterization
of floats as partial versions of acceptance implies that floats are practical attitudes. More
precisely, some of the reasons that support floating a proposition are practical reasons, as
opposed to evidential ones. In what follows I will argue that there is an interesting sense in
which this is true.
Before doing that, I want to present and put aside an uninteresting sense in which the
reasons that support floating a proposition are practical. I take it that the facts that are reasons
for floating a proposition are those that count in favor of floating that proposition. And I take
it that being offered a million dollars to float a proposition counts in favor of floating that
proposition. Finally, the fact that you are offered a million dollars to float a proposition is not
(at least in an ordinary case) evidence in favor of the truth of that proposition. So, I take it
that there are practical, non-evidential reasons for floating a proposition. This is a rather
uninteresting conclusion, since it does not capture the distinctively practical nature of floats.
We can see this by noticing that the argument just given succeeds for credences as well. Being
offered a million dollars counts in favor of having a credence in a proposition, as well.
But, there is an interesting distinction in the ways that a fact can be a reason for or
count in favor of floating a proposition. I will suggest that there are some reasons that are
177
Finlay sees this picture as one where floats just are occurrent credences. I think this is doubtful, since Phil
can be completely aware of his very low credence that Brad Pitt is the thief without taking this proposition as
an option in his reasoning. Awareness may not be the same as occurrence, but I find it hard to see how, in
this case, one could deny that Phil’s credence is occurrent without simply defining one’s occurrent credences
as the credences one is using in reasoning. A similar sort of account of credences is suggested in Holton
2008 (39).
161
what are called the right kind of reasons for floating a proposition and some that are what are
called the wrong kind of reasons for floating a proposition.
178
The interesting thesis that I will
argue for is that there are practical reasons that are the right kind of reasons for floating.
179
Let me try to illustrate the right/wrong kind distinction by way of an example involving
belief. There seems to be a stark difference between the sorts of reasons at play in the
following two cases. Imagine that you are standing by an urn filled with balls. In the first
case, you draw and replace a ball from the urn fifty times. Each time, the ball drawn is red.
You are thinking about whether you will draw a red ball on your next attempt. The fact that
you drew a red ball each of the first fifty attempts is a reason to believe that you will draw a red
ball on your next attempt (it may or may not be a sufficient reason to form this belief, but it
counts in favor of forming it). Contrast this with the following case, in which you draw a black
ball on each of your first fifty attempts. In this case, someone offers you a million dollars to
believe that you will draw a red ball on your next draw. There is a sense in which the fact that
someone offered you a million dollars to have the belief counts in favor of believing that you
will draw a red ball on your next attempt. But it is easy to feel that there is something fishy
going on.
180
People generally have the following reaction to these cases. They think that the
reason to form the belief in the first case is the right kind of reason to form a belief, while the
reason in the second is not the right kind of reason to form a belief. I want to stay officially
neutral on how, exactly, to delimit the right from wrong kinds of reasons for beliefs.
181
I just
wanted to use the familiar case of belief to exemplify the distinction.
Now that we have a feeling for the general difference between the right and wrong kinds
of reasons, let’s move on. We’ve seen that floating a proposition is just partially accepting that
proposition in an inquiry. This suggests that if there is a right kind/wrong kind of reasons
178
See, for instance, Rabinowicz and Rønnow-Rasmussen 2004, Hieronymi 2005, Schroeder 2012b; 2013a, Piller
2006, and Reisner 2009. Sometimes this distinction is framed as one between whether an attitude is good to
have and whether it is fitting (D’Arms and Jacobson 2000).
179
There are theorists who would take issue with the two preceding paragraphs because they think that there are
no reasons of the wrong kind. What I will end up classifying as reasons of the wrong kind are not, according
to them, reasons at all. See Way 2012, Kelly 2002, and Skorupski 2007. Kolodny (2005) and Shah (2006)
take this position about reasons for belief. While I don’t advocate this position, it still leaves interesting
room to be explored. Theorists of this bent can take the following discussion as an attempt to delineate
reasons from non-reasons for floating.
180
Pascal’s wager is a classic example involving a wrong kind of reason for belief. See Shah 2006 for discussion.
181
A natural first stab at the distinction is that the sorts of things that are reasons of the the right kind for beliefs
are evidence. See Kelly 2002, Shah 2006, Moran 1988 (who all also think that there are no reasons of the
wrong kind for belief). Some think that there can be non-evidential considerations that are the right kind of
reasons for belief. For views in this neighborhood, see Foley 1993, Fantl and McGrath 2009, and Schroeder
2012a; 2012b (who thinks that there can be non-evidential reasons against belief). But it seems, at least in our
case, that being offered a million dollars to believe something isn’t one of them.
162
distinction for accepting, there will be such a distinction for floating, as well. It also suggests
that, if there is such a distinction for accepting, the line between right and wrong kinds of
reasons will likely be in the same place for floating. So, investigating the case of acceptance
can serve as a useful step toward getting answers in the case of floating.
182
We might think that, since acceptance is a practical attitude, there simply is no right
kind/wrong kind of reasons distinction. All reasons for acceptance are reasons of the right
kind. But, it seems that this isn’t so. Suppose you are inquiring into the proper dimensions
to build the towers of a suspension bridge. And suppose that you are offered a billion dollars
to accept Aristotelian physics in your inquiry. One can suppose, further, that Aristotelian
physics would lead you to create a bridge that collapses (this supposition is not necessary, but
makes judgments more vivid). The fact that you are offered a billion dollars to accept
Aristotelian physics is a reason to do so. But, it seems, it is not a reason of the right kind.
183

So, not all considerations that favor accepting a proposition are reasons of the right kind for
accepting that proposition. There is some line between the right and wrong kinds of reasons
for acceptance.
Let’s try to get at the line by investigating a couple more examples. I take it that
evidential reasons are of the right kind for acceptance. For instance, in the first sort of urn
case, the fact that you have drawn red balls in each of your first fifty attempts is a reason of the
right kind to accept that you will draw a red ball on your next attempt. This may make us
wonder whether the line between the right and wrong kind of reasons for acceptance is just the
line between evidential and non-evidential reasons.
It isn’t. There are non-evidential reasons that are of the right kind for acceptance.
Though we’ve seen that not all practical reasons are of the right kind for acceptance, there are
practical reasons of the right kind for acceptance. This idea is common in the literature. I
take it that in the following quotes the authors are talking not about reasons in general, but
about reasons of the right kind. As Bratman (1992, 9) says, “acceptance can be driven by a
wide range of practical considerations, considerations that provide practical reasons for
acceptance rather than evidence for the truth of what is accepted.” And Cohen (1989, 369)
agrees, “the reasons for accepting that p need not always be epistemic ones.”
182
Schroeder (2012b) gives four earmarks of the right kind/wrong kind distinction. In what follows, I will focus
on one earmark – that the wrong kind of reasons have a distinctive flavor (the one we feel in the wrong kind
of reason urn case above). Whether the other three earmarks that Schroeder gives apply to acceptance is not
straightforward. I’ll leave a discussion of this for another time.
183
Thanks to Shyam Nair for suggesting this case and to Mark Schroeder for discussion.
163
In the last section we saw an example where a practical, non-evidential consideration is
a reason of the right kind for accepting a proposition. Suppose you are, again, inquiring into
the proper dimensions to build the towers of a suspension bridge. Though you believe that
Newtonian physics is false, you know that using Newtonian physics will allow you to figure out
the proper dimensions. The fact that accepting Newtonian physics will lead to the right
answer is a practical, not an evidential, reason for accepting the theory. It supports accepting
the theory to achieve your goal – figuring out the proper dimensions for your bridge – but it
does not weigh in favor of the truth of the theory. Be that as it may, this reason seems like the
right kind of reason for accepting this theory in your context. So, it seems that a practical
reason can be a reason of the right kind for accepting a proposition.
We’ve said that floating is partially accepting. So, it is reasonable to think that there
will also be a distinction between the right and wrong kind of reasons for floating. And it is
reasonable to think that the line between the right and wrong kinds of reasons for floating
should be in the same place as for accepting.
Indeed, there do seem to be reasons of the wrong kind for floating, just as for
acceptance. Imagine that Lil is doing her astronomy homework which contains some difficult
calculations. An eccentric billionaire promises to give her a million dollars if she floats
Ptolemaic astronomy in answering the problem. Imagine that Lil knows that doing so will
prevent her from getting the right answer. Nonetheless, the fact that she will get a million
dollars for doing so is a reason to float Ptolemaic astronomy. However, it is not a reason of
the right kind for doing so. Just as the offer of money for building a bridge based on
Aristotelian physics, this reason is of the wrong kind for having the attitude.
So, just as for acceptance, there are reasons of the wrong kind for floating. But it also
seems that the same sorts of reasons that were reasons of the right kind for acceptance are
reasons of the right kind for floating. Evidential reasons are of the right kind for floating (the
fact, say, that you’ve drawn a red ball on half of your first fifty attempts is a reason to float that
you will draw a red ball on your next attempt). But, practical reasons can also be of the right
kind for floating. Let’s look at some examples that show this.
Consider, again, two cases from §1. First is the case where Phil is inquiring with his
friends into who is the thief. He has zero credence in the proposition that Hillary Clinton is
the thief. Nonetheless, the other members of the group take it as an option that she is the
thief. The inquiry would be greatly facilitated if all the participants floated the same
164
proposition. This fact is a reason of the right kind for Phil to float that Clinton is the thief –
even though it is not evidence that she is. This is a case where Phil has a reason for floating
which is of the right kind but is not evidence about the truth (or falsity) of the proposition
floated.
Second, consider the case where Lil is doing an astronomy problem. She only is able
to do the calculations according to the assumptions of Galilean astronomy and Copernican
astronomy. And she remembers that if two theories yield the same answer, that answer is the
correct one. These facts seem to constitute a reason of the right kind to float Galilean
astronomy and float Copernican astronomy. Nonetheless, they are not evidence in favor of
either theory. Again, we have a case where Lil has a reason for floating, which is of the right
kind but is not evidence for the truth (or falsity) of the proposition floated.
So, there are reasons of the right kind for floating that are not evidence. It’s worth
noting that this appears to contrast with reasons for credence, which all seem to be evidential.
I won’t argue for this here, but I will remind the reader that in the two cases just surveyed,
reasons which are of the right kind for floating are of the wrong kind for changing one’s
credence. So, if there are non-evidential reasons for changing one’s credence, they must be of
a different sort.
So far, I’ve used some examples to make vivid the distinction between the right kind
and the wrong kind of reasons for floating, and I’ve suggested that the right kind/wrong kind
line for floating is not the same as the line between evidential and practical concerns (though I
have hypothesized that it is for credences). Now, I want to say a bit more about what makes
something a reason of the right kind for floating. Floats are fundamentally attitudes that
agents adopt in an inquiry. The goal of the inquiry is to find the answer to some question.
The purpose of floating a proposition is to help the agent answer the question she is inquiring
about. Since this is the purpose of the attitude, it makes sense that reasons of the right kind
for having the attitude are those that promote, in the right way, achieving that goal, while
reasons of the wrong kind for having the attitude are those that don’t. That is, reasons of the
right kind for floating are those that promote, in the right way, answering the question while
reasons of the wrong kind for floating are those that do not.
184
This is not, by any means, an
184
This characterization is quite similar to the general outline suggested in Schroeder 2012b. He (2012b, 483)
says, “the class of right-kind reasons with respect to any activity will need to depend on the nature of that
activity, in some way.” And later, Schroeder (2013b, 485) continues: “the right kinds of reason for and against
intention and belief are clearly associated with benefits or costs of intention and belief that are much more
closely connected to the nature of intention and belief than the benefits and costs are in cases involving
wrong-kind reasons.” I also take my account to be somewhat similar to the one in Hieronymi 2005 in that
165
informative statement of necessary and sufficient conditions for being a reason of the right
kind. In this characterization I have referred to promoting the goal in the right way. The
mystery of solving the wrong kind of reasons problem lies in understanding what the right way
is.
Nonetheless, our discussion does reveal an important general lesson about how to draw
the line between the right and wrong kinds of reasons for an attitude. It is quite common to
believe that the distinction between the right kind and the wrong kind of reasons for an
attitude is just the distinction between what are called content-related and attitude-related
reasons for that attitude. Let me first say what this distinction is. Then I will state the theory
built on it. Finally, I will show how our examples show this theory to be false.
In general, we can draw a distinction between what Piller (2001; 2006) calls content-
related and attitude-related reasons for an attitude (Parfit (2001) draws the same distinction –
calling some reasons object-given and some state-given). As Schroeder (2012b, 462) puts the
distinction, content-related reasons bear on the content of the attitude, while attitude-related
reasons “just bear on the benefits of being in the state of having that attitude.” It turns out
that there is significant difficulty in cashing out the property of bearing on the content of an
attitude in a way general enough to apply to all sorts of attitudes toward that content.
185
But, I
think we can safely sidestep this issue by following Olson (2004) and Piller (2006) in starting
with a characterization of attitude-related reasons and allowing content-related ones to be the
remainder. They say that a reason for attitude, A, is attitude-related just in case it cannot be
described without reference to A, itself. I will take it that this self-reference is the mark of an
attitude-related reason. And we can then define content-related reasons the be all reasons for
an attitude that are not attitude-related.
With this distinction in hand, we can evaluate the following natural thesis about the
distinction between the right kind and the wrong kind of reasons for an attitude: The right
kind of reasons for an attitude are all and only the content-related reasons for the attitude and
the wrong kind of reasons for an attitude are all and only the attitude-related reasons for the
attitude. A number of theorists – Olson (2004), Parfit (2001), and Stratton-Lake (2005) – have
taken this position or something very much like it.
186
For example, we can contrast the
we both think that the nature of the attitude in question determines what sorts of reasons count as being of
the right kind. However, her account does not seem to extend to attitudes of acceptance (including floats).
Further comparison lies beyond the scope of this paper.
185
See Schroeder 2012b for discussion.
186
Kelly (2002) focuses on this sort of position for belief, but also thinks it applies to desire, regret, fear, and,
perhaps, other attitudes. However, he doesn’t think this carves the general distinction between right and
166
reasons in favor of a belief in p – that the belief in p will lead to the happiness of the agent and
that there is overwhelming evidence for p. The first sort of reason is an attitude-related
reason, while the second is content-related. Content-related reasons are supposed to be of the
right kind while attitude-related ones are supposed to be of the wrong kind.
Consideration of floating, however, suggests that the content/attitude distinction isn’t
the right way to draw the line between the right and wrong kinds of reasons, in general.
187
This
is because some reasons of the right kind for floating are what would be classified as attitude-
related. The reason for floating that Hillary Clinton is the thief is that floating this will
advance Phil’s inquiry into who is the thief. This is an attitude-related reason. So, this
example shows that the right kind of reasons can, for certain attitudes, include those that are
attitude-related. Thus, a general account of the right/wrong kind of reason distinction cannot
assimilate it to the distinction as between content- and attitude-related reasons.
In this section, I’ve given an account of reasons for floating according which reasons of
the right kind for floating include not just evidential reasons but also practical ones. And I’ve
suggested an upshot for the debate about how to draw the line between the right and wrong
kinds of reasons, in general.
5. Other Float-like Attitudes
In this chapter, the discussion has centered around floats (mostly bare and sometimes
degreed). We’ve not yet talked about other sorts of float-like attitudes – sinks, conditional
floats, and quantificational floats. Can the picture of floats developed to this point account for
these other attitudes?
This picture can be extended to other float-like attitudes. Let’s start by considering
sinks. A sink, remember, is an attitude of ruling out a proposition as an option in reasoning.
It’s natural to think that ruling out a proposition as an option in reasoning is just taking its
negation for granted in reasoning. That is, it is natural to think that sinking a proposition is
just accepting its negation. Given this, we have a picture according to which simple (non-
conditional, non-quantificational) float-like attitudes are just attitudes of (partial and full)
acceptance. This is a very tidy picture.
wrong kinds of reasons.
187
Kelly (2002) makes a point about practical considerations rationalizing the attitude of supposing which is
similar. And a similar conclusion would follow from some remarks about assuming in Piller 2001. See
Rabinowicz and Rønnow-Rasmussen 2004 and Schroeder 2012b; 2013 for different arguments for this
conclusion.
167
Let’s move on to complex float-like attitudes. The obvious position about conditional
floats and sinks is that they are conditional versions of partial and full acceptance, respectively.
It is natural to think that there are conditional versions of acceptance, just as there are
conditional versions of belief, credence, desire, intention, etc. On the other hand, it is not
common to think that there are quantificational versions of acceptance (or quantificational
versions of any attitude). But, I take it that this is just an oversight in the literature about the
sorts of states of mind there are – not a reasoned position that blocks our attempt to cash out
float-like attitudes in terms of acceptance-like ones. Indeed, I think there are good reasons for
allowing quantificational mental states into our ontology (see Chapter Three and especially
Lennertz forthcoming a for more on this). So, I think we should allow that there are
quantificational versions of partial and full acceptance and that they are the same thing as
quantificational floats and sinks.
In this short section, we’ve seen that there are no obvious obstacles to extending the
account of floats as attitudes of partial acceptance to other float-like attitudes. Thus, it seems
that we can make sense of float-like attitudes, in general, as acceptance-like attitudes.
6. Epistemic Evaluation and Disagreement Revisited
In Chapter Three I presented a sufficient condition for disagreement, Evaluationism,
which was underpinned by agents’ evaluations of others’ attitudes. I suggested that
disagreement between agents in virtue of floats and sinks depends on a disposition to give a
negative epistemic evaluation – i.e. an evaluation of the accuracy of an attitude and its
responsiveness to the evidence. However, in our current context this may seem a bit strange.
I just spent a great deal of time discussing how some reasons of the right kind for floating a
proposition are not evidence in favor of its truth. We might wonder, then, whether
evaluations of floats really do go by way of the sorts of evaluations of accuracy and evidence
that I discussed in Chapter Three or if they go by way of all of the reasons of the right kind for
having the attitude – evidential or otherwise. In this section I will suggest that we need to
amend the account from Chapter Three because evaluations of an attitude are sensitive to all
reasons of the right kind for having the attitude. I will also show how such evaluations can
underpin a disagreement condition that allows us to give similar explanations to the ones we
gave using Evaluationism.
Before undertaking this discussion, we should pause to keep the following in mind.
168
The account of evaluation of attitudes and disagreement that I gave in Chapter Three had a
lot of explanatory power. And, intuitively, it should still apply in a large swath of cases – those
where agents are guided in forming and evaluating float-like attitudes toward propositions
solely according to concerns of accuracy and evidence. Though I have been stressing, in this
chapter, the fact that other reasons may properly support floating a proposition, we should
remember that most of the cases that led to this conclusion were not the central cases of
reasoning with floats. In the central cases – like the original incarnation of the case of the
missing cookies – we want the account of evaluation and disagreement that I gave in Chapter
Three to still apply. What this means is that we should try to form a broader picture of
evaluation and disagreement which has our old picture of epistemic evaluation and
disagreement as a subpicture.
Let’s start to build this broader picture by returning to think about the dimensions
along which agents can evaluate others’ attitudes. We saw in Chapter Three that attitudes can
be evaluated with respect to their accuracy and how they respect the evidence. Here I want to
focus on a different dimension along which an attitude can be evaluated. Let’s think about the
group reasoning case, where Phil has zero credence in the proposition that Hillary Clinton is
the thief, but he floats it nonetheless – because he knows that doing so will allow the inquiry to
progress. In the context of this group reasoning (where Phil has put his thoughts about
likelihood to the side and floats that Clinton is the thief for the purpose of the inquiry), he will
positively evaluate his interlocutors’ floats that Clinton is the thief. But he does not do so
because he thinks such floats are accurate or properly responsive to the evidence. Rather he
positively evaluates such floats because they are conducive to figuring out the answer to the
guiding question of the inquiry. So, it seems that there can sometimes be a dimension of
evaluation that takes precedence over an evaluation of accuracy and evidential virtue.
There is a simple explanation for what is going on in this case. It is that such an
evaluation is sensitive to reasons which, though not evidential reasons, are reasons of the right
kind for having the float. As we saw in §4, the following fact is a reason of the right kind for
floating that Clinton is the thief in the scenario of the group reasoning: that floating that
Clinton is the thief is conducive to figuring out the answer to the guiding question of the
inquiry. And it seems plausible, in general, that agents evaluate floats in virtue of reasons of
the right kind for having the states. An agent will positively evaluate a float on a practical
dimension if she takes it to promote, in the right way, the current inquiry.
169
So we’ve seen that the non-evidential aspect of an evaluation can take precedence over
concerns of accuracy and respecting the evidence, since non-evidential reasons can be stronger
than, and trump, the other reasons of the right kind. Nonetheless, if such non-evidential
reasons are not present, then the evaluation won’t be made on those grounds and the
mechanism of evaluation that we discussed in Chapter Three kicks in. This is what happens
in many cases – those where what it is rational to float and sink (in virtue of reasons of the
right kind) depends only on considerations of accuracy and evidence – like in the original case
of Phil reasoning about who stole the cookies.
The general picture of evaluation that emerges is quite tidy. An agent evaluates an
attitude based on her estimation of the reasons of the right kind for having that attitude. If an
agent finds that the reasons of the right kind support having an attitude, then she will
positively evaluate that attitude. If she thinks that they support not having the attitude, then
she will negatively evaluate that attitude. If she does not have an opinion about whether they
support having the attitude or not, then she will not evaluate that attitude. In the case of floats
this plays out as follows. Non-evidential, practical reasons, which are of the right kind, can
trump considerations of accuracy and respecting the evidence. Agents will, in such cases,
make an evaluation based on what they take such reasons to support. I also think it is
possible that these sorts of reasons can be rather weak and can be outweighed by
considerations of accuracy and respecting the evidence. For instance, consider a variation of
the group reasoning case where, if Phil floats that Clinton is the thief, a minor distracting
discussion may be avoided – saving the group five seconds in figuring out the answer. It is
plausible that this is not a strong enough reason in favor of floating that Clinton is the thief to
outweigh the reasons against having that float. So, it is plausible that Phil will not have that
float in that situation and will negatively evaluate such a float. What is important to note
about this picture is that in cases where there are no practical reasons of the right kind for
having a float, the evaluation will proceed in the manner discussed in Chapter Three.
188
Thus,
we have a nice picture according to which evaluations of attitudes follow an agent’s take on the
reasons of the right kind in favor of that attitude. And, as we have seen, this picture preserves
the explanatory power of the account of evaluation that I developed in Chapter Three.
Now that I’ve discussed the nature of evaluations, we should see how this bears on
188
This is so under the assumption that the perceived accuracy or inaccuracy of an attitude always offers a
stronger reason in favor of an attitude than all of the evidential considerations pulling in the other direction.
Such an assumption is plausible, but we need it in order to ensure that evaluations of accuracy always take
precedence over evaluations of evidential virtues.
170
disagreement. In Chapter Three we gave the following disagreement condition:
Evaluationism: Suppose that agent, A, has an epistemic attitude, x. Another agent, B,
disagrees with A if B is disposed to give a negative epistemic evaluation of x.
We now possess a more general notion of evaluation of attitudes – where it depends on the
agent’s take on the balance of reasons of the right kind for or against the attitude. So, we
should generalize the disagreement condition accordingly:
Evaluationism*: Suppose that agent, A, has an attitude, x. Another agent, B, disagrees
with A if B is disposed to give a negative evaluation of x based on the reasons of the
right kind for having x.
Evaluationism* extends the notion of disagreement characterized by Evaluationism to all
attitudes for which there are reasons of the right kind.
189
Evaluationism* predicts that whether one agent disagrees with another will depend on
the situation in which that agent finds herself. This follows from two general facts about
dispositions. First, an agent can have a disposition to do some action in some circumstances
and do a contrary action in other circumstances. I take this to be an obvious fact about
dispositions. The second fact is less obvious. What an agent is disposed to do (absent a
specification of specific circumstances) depends on the actual situation she is in. A more
philosophical way of saying this is that dispositions track what obtains in closer possible
worlds where specified triggering conditions are met. For instance, suppose we are wondering
what I am disposed to do on being given a parka. The answer to this question is different if
we ask it about me while I am in one hundred degree heat in the desert than it is if we ask it
about me while I am in freezing temperatures in just a t-shirt. If we ask this question while I
am in the desert, the answer is that I am disposed to not put it on (and perhaps to leave it in
the car or make a sarcastic joke about being given a parka in such conditions). In nearby
worlds where I am given a parka, it is still very hot and this affects what I am disposed to do.
If we ask this question while I am in freezing temperatures, the answer is that I am disposed
to put it on right away. In this scenario, the nearby worlds are freezing worlds. So, in general,
what an object is disposed to do can vary with the actual circumstances in which it finds itself.
Let’s look at two versions of reasoning about the cookie theft to see what impact this
189
We might welcome this result, for example, in the case of intention – where it seems that I can disagree with
you if you intend that we go to the movies tonight and I negatively evaluate your intention based on the right
kinds of reasons for intending (perhaps I intend that we go bowling instead, since I think the reasons of the
right kind support bowling rather than seeing a movie). I won’t discuss all of the sorts of attitudes for which
Evaluationism * will predict disagreement, though I think assessing Evaluationism * in general is an interesting
project.
171
has on Evaluationism*. Suppose, first, that Phil is reasoning with the group. In this case there
are strong non-evidential reasons in favor of floating that Clinton is the thief (that doing so is
necessary for proceeding in the inquiry). In this case, we can ask if Phil is disposed to give a
negative evaluation of a float that Clinton is the thief. As we said above, he is not (in fact, he
is disposed to give a positive evaluation of such a float). So, according to Evaluationism * he
does not disagree with his interlocutors in virtue of their floats that Clinton is the thief (in fact,
he agrees with them, as the to-be-presented Evaluationism Agree * will bear out). Now picture
an alternative scenario in which Phil is reasoning on his own about who stole the cookies.
The social pressures and corresponding non-evidential reasons in favor of floating that Clinton
is the thief are no longer present. In this case, we can ask if Phil is disposed to give a negative
evaluation of a float that Clinton is the thief. Here the answer is yes. This is because he
thinks that Clinton is not the thief and would give a negative evaluation of the accuracy of
such a float. So, according to Evaluationism * he does disagree with someone who floats that
Clinton is the thief.
Thus, Evaluationism* predicts that disagreement varies based on the situation the
potential disagree-er is in. This means that disagreement is, in a sense, context-dependent.
And, I think, this is the correct prediction. If we are thinking about Phil, in the context of
engaging in group inquiry, then we will say that he does not disagree with his interlocutors in
virtue of their float that Hillary Clinton is the thief. After all, he is part of a group which takes
this possibility as a serious option in reasoning.
190
However, if we are thinking about Phil as
reasoning on his own about who stole the cookies, we will think that he does disagree with
someone who floats that Clinton is the thief. It is a virtue of Evaluationism * that it predicts
this variation in disagreement.
We can also generalize the agreement condition given in Chapter Three,
Evaluationism Agree, as follows:
Evaluationism Agree *: Suppose that agent, A, has an attitude, x. Another agent, B,
agrees with A if B is disposed to give a positive evaluation of x based on the reasons of
the right kind for having x.
As with disagreement, agreement will vary based on the context in which the agent finds
herself. For example, Evaluationism Agree * will correctly predict that, when we are thinking
190
We need to be careful not to be misled into rejecting this conclusion for the wrong reason. Just because Phil
doesn’t disagree with the members of the group in virtue of their floats does not mean he doesn’t disagree
with them in virtue of some other attitudes. Indeed, it is plausible that in this case, he disagrees with them in
virtue of their credences about Clinton being the thief.
172
about Phil in the context of engaging in the above group inquiry with all of its social pressures,
Phil agrees with the bearer of a float that Hillary Clinton is the thief (because he is disposed to
give a positive evaluation of the float in such circumstances). But it will also predict that,
when we are thinking about Phil on his own, he does not agree with the bearer of that float
(because he is not disposed to give a positive evaluation of the float in such circumstances).
These results about agreement and disagreement are exactly the ones that we want.
They allow us to properly extend the account given in Chapter Three to accommodate the fact
that floats are responsive to non-evidential reasons, while retaining the ability to use the
disagreement and agreement conditions in explanations of the range of data about epistemic
modal communication.
In this chapter I explored what float-like attitudes are and how they are sensitive to
more than just the evidence in favor of a proposition. I argued that floats cannot be reduced
to credences. But I used the relationship between floats and credences to come to the
conclusion that floats are partial versions of the attitude of accepting in an inquiry. We then
saw that there are right and wrong kinds of reasons for floating, just like accepting, and that the
line between the right and wrong kinds of reasons allows some non-evidential reasons to count
in favor, in the right way, of having these attitudes. I then extended this account to other
float-like attitudes, suggesting that they are all just versions of acceptance-like attitudes.
Finally, I showed how to broaden the notions of evaluation and disagreement introduced in
Chapter Three to account for the practical nature of float-like attitudes, while retaining the
explanatory power of the account given there.
Together with Chapter Three, this chapter comprises my account of float-like attitudes,
which, I have claimed, are what agents use in explicit reasoning with uncertainty. This picture
of the mind dovetails nicely with the account epistemic modal communication developed in
Chapters Four, Five, and Six. In this dissertation, I have developed an overall theory of
reasoning and communication in situations of uncertainty that is quite attractive. It is, I
think, plausible on its face, and it explains a wide range of data about epistemic
communication that, as we saw in Chapters One and Two, spelled trouble for the existing
theories in the literature.
173
References
Alston, William. 1996. “Belief, Acceptance, and Religious Faith.” In Faith, Freedom, and
Rationality, ed. Jeff Jordan and Daniel Howard-Synder, 3-27. Lanham, Maryland:
Rowman & Littlefield Publishers.
Anand, Pranav, and Valentine Hacquard. 2013. “Epistemics and Attitudes.” Semantics and
Pragmatics 6, no. 8: 1-59.
Asher, Nicholas, and Alex Lascarides. 2011. “Indirect Speech Acts.” Synthese 128: 183-228.
Austin, J.L. 1962. How to Do Things with Words. Cambridge, MA: Harvard University Press.
Austin, J.L. 1979. “Performative Utterances.” In Philosophical Papers, ed. J. Urmson & G.
Warnock, 233-52. Oxford: Oxford University Press.
Bach, Kent. 1975. “Performatives are Statements Too.” Philosophical Studies 28: 229-36.
Bach, Kent. 1994. “Conversational Impliciture.” Mind and Language 9: 124-62.
Bach, Kent. 2011. “Perspectives on Possibilities: Contextualism, Relativism, or What?” In
Epistemic Modality, ed. Andy Egan and Brian Weatherson, 19-59. Oxford: Oxford
University Press.
Bach, Kent, and Robert Harnish. 1979. Linguistic Communication and Speech Acts.
Cambridge: MIT Press.
Bach, Kent, and Robert Harnish. 1992. “How Performatives Really Work: A Reply to
Searle.” Linguistics and Philosophy: 93-110.
Barnett, David. 2006. “Zif is if.” Mind 115: 519-66.
Barnett, David. 2009. “Yalcin on ‘Might’.” Mind 118: 771-75.
Bennett, Jonathan. 2003. A Philosophical Guide to Conditionals. Oxford: Oxford University
Press.
Björnsson, Gunnar, and Alexander Almér. 2010. “The Pragmatics of Insensitive Assessments:
Understanding the Relativity of Assessments of Judgments of Personal Taste,
Epistemic Modals, and More.” In The Baltic International Yearbook of Cognition,
Logic, and Communication 6: 1-45.
Björnsson, Gunnar, and Stephen Finlay. 2010. “Metaethical Contextualism Defended.” Ethics
121: 7-36.
Blackburn, Simon. 1988. “Attitudes and Contents.” Ethics 98: 501-17.
Boisvert, Daniel. 2008. “Expressive-Assertivism.” Pacific Philosophical Quarterly 89: 169-203.
Bratman, Michael. 1987. Intentions, Plans, and Practical Reason. Cambridge, MA: Harvard
174
University Press.
Bratman, Michael. 1992. “Practical Reasoning and Acceptance in a Context.” Mind, 101: 1-15.
Braun, David. 2012. “An Invariantist Theory of ‘Might’ Might be Right.” Linguistics and
Philosophy 35: 461-89.
Brössel, Peter, and Anna-Maria Eder. Forthcoming. “How to Resolve Doxastic
Disagreement.” Synthese.
Cappelen, Herman, and John Hawthorne. 2009. Relativism and Monadic Truth. Oxford:
Oxford University Press.
Clark, Herbert. 1979. “Responding to Indirect Speech Acts.” Cognitive Psychology 11: 430-77.
Clarke, D.S. 1994. “Does Acceptance Entail Belief?” American Philosophical Quarterly 31: 145-
55.
Cohen, L. Jonathan. 1989. “Belief and Acceptance.” Mind 98: 367-89.
Cook, Ezra. 2013. “Epistemic Modals and Common Ground.” Inquiry 56: 179-209.
Crabill, Joshua. 2013. “Suppose Yalcin is Wrong About Epistemic Modals.” Philosophical
Studies 162: 625-35.
D’Arms, Justin. and Daniel Jacobson. 2000. “The Moralistic Fallacy: On the
‘Appropriateness’ of Emotions.” Philosophy and Phenomenological Research 61: 65-
90.
Davis, Wayne. 2003. Meaning, Expression, and Thought. Cambridge: Cambridge University
Press.
Davison, Alice. 1975. “Indirect Speech Acts and What to Do with Them.” In Syntax and
Semantics vol. 3: Speech Acts, ed. Peter Cole and Jerry L. Morgan, 143-85. New York:
Academic Press.
DeRose, Keith. 1991. “Epistemic Possibilities.” The Philosophical Review 100: 581-605.
DeRose, Keith, and Richard Grandy. 1999. “Conditional Assertions and ‘Biscuit’
Conditionals.” Noûs 33: 405-20.
Dietz, Richard. 2008. “Epistemic Modals and Correct Disagreement.” In Relative Truth, ed.
Manuel Garcia-Carpintero and Max Kölbel, 239-62. Oxford: Oxford University Press.
Dorr, Cian, and John Hawthorne. Forthcoming. “Embedding Epistemic Modals.” Mind.
Dowell, Janice. 2011. “A Flexible Contextualist Account of Epistemic Modals.” Philosophers’
Imprint 11, no. 14: 1-25.
Edgington, Dorothy. 1995. “On Conditionals.” Mind 104: 235-329.
175
Egan, Andy. 2007. “Epistemic Modals, Relativism, and Assertion.” Philosophical Studies 133:
1-22.
Egan, Andy. 2010. “Disputing about Taste.” In Disagreement, ed. Richard Feldman and Ted
A. Warfield, 247-86. Oxford: Oxford University Press.
Egan, Andy, John Hawthorne, and Brian Weatherson. 2005. “Epistemic Modals in
Context.” In Contextualism in Philosophy: Knowledge, Meaning, and Truth, ed.
Gerhard Preyer and Georg Peter, 131-68. Oxford: Oxford University Press.
Engel, Pascal. 1999. “Dispositional Belief, Assent, and Acceptance.” Dialectica 53: 211-26.
Fantl, Jeremy, and Matthew McGrath. 2009. Knowledge in an Uncertain World. Oxford:
Oxford University Press.
Finlay, Stephen. 2005. “Value and Implicature.” Philosophers’ Imprint 5, no. 4: 1-20.
Finlay, Stephen. 2009. “Oughts and Ends.” Philosophical Studies 143: 315-40.
Finlay, Stephen. 2010. “What Ought Probably Means, and Why You Can’t Detach It.”
Synthese 177: 67-89.
Finlay, Stephen. 2014. Confusion of Tongues: A Theory of Normative Language. Oxford:
Oxford University Press.
Finlay, Stephen, and Benjamin Lennertz. MS. “What Might but Must not Be.”
von Fintel, Kai, and Anthony S. Gillies. 2007. “An Opinionated Guide to Epistemic
Modality.” In Oxford Studies in Epistemology 2, ed. Tamar Szabó Gendler and John
Hawthorne, 32-62. Oxford: Oxford University Press.
von Fintel, Kai, and Anthony S. Gillies. 2008. “CIA Leaks.” The Philosophical Review 117: 77-
98.
von Fintel, Kai, and Anthony S. Gillies. 2011. “Might Made Right.” In Epistemic Modality, ed.
Andy Egan and Brian Weatherson, 108-30. Oxford: Oxford University Press.
von Fintel, Kai, and Anthony S. Gillies. MS. “Would You Believe That von Fintel and Gillies
Might be Wrong? Comments on a paper by Ben Lennertz.” APA Pacific, April 20,
2011. http://mit.edu/fintel/fintel-gillies-2011-lennertz-apa.txt
von Fintel, Kai, and Sabine Iatridou. 2003. “Epistemic Containment.” Linguistic Inquiry 34:
173-98.
Foley, Richard. 1993. Working Without a Net. Oxford: Oxford University Press.
Genest, Christian, and James Zidek. 1986. “Combining Probability Distributions: A Critique
and an Annotated Bibliography.” Statistical Science 1: 114-35.
176
Geurts, Bert. 2005. “Disjunction.” Natural Language Semantics 13: 383-410.
Gibbard, Alan. 1990. Wise Choices, Apt Feelings: A Theory of Normative Judgment.
Cambridge: Harvard University Press.
Gibbard, Alan. 2003. Thinking How to Live. Cambridge: Harvard University Press.
Grice, H. Paul. 1989a. “Indicative Conditionals.” In Studies in the Ways of Words, 58-85.
Cambridge, MA: Harvard University Press.
Grice, H. Paul. 1989b. “Logic and Conversation.” In Studies in the Ways of Words, 22-40.
Cambridge, MA: Harvard University Press.
Hacking, Ian. 1967. “Possibility.” The Philosophical Review 76: 143-68.
Hale, Bob. 1993. “Can There be a Logic of the Attitudes?” In Reality, Representation, and
Projection, ed. John Haldane and Crispin Wright, 337-63. New York: Oxford
University Press.
Hawthorne, John. 2004. Knowledge and Lotteries. Oxford: Oxford University Press.
Hieronymi, Pamela. 2005. “The Wrong Kind of Reason.” Journal of Philosophy 102: 437-57.
Holton, Richard. 2008. “Partial Belief, Partial Intention.” Mind 117: 27-58.
Horn, Laurence. 1989. A Natural History of Negation. Chicago: University of Chicago Press.
Huemer, Michael. 2007. “Epistemic Possibility.” Synthese 156: 119-42.
Jackson, Frank. 1979. “On Assertion and Indicative Conditionals.” The Philosophical Review
88: 565-89.
Jeffrey, Richard. 1983. “Bayesianism with a Human Face.” In Minnesota Studies in the
Philosophy of Science Volume 10: Testing Scientific Theories, ed. John Earman, 133-
56. Minneapolis: Minnesota University Press.
Joyce, James. 1998. “A Nonpragmatic Vindication of Probabilism.” Philosophy of Science 65:
575-603.
Kelly, Thomas. 2002. “The Rationality of Belief and Some Other Propositional Attitudes.”
Philosophical Studies 110: 163-96.
Knobe, Joshua, and Seth Yalcin. MS. “Fat Tony Might be Dead: An Experimental Note on
Epistemic Modals.”
Kölbel, Max. 2004. “Faultless Disagreement.” Proceedings of the Aristotelian Society 104: 53-
73.
Kolodny, Niko. 2005. “Why be Rational?” Mind 114: 509-63.
Kratzer, Angelika. 1977. “What ‘Must’ and ‘Can’ Must and Can Mean.” Linguistics and
177
Philosophy 1: 337-55.
Kratzer, Angelika. 1981. “The Notional Category of Modality.” In Words, Worlds, and
Contexts: New Approaches to Word Semantics, ed. Hans-Jürgen Eickmeyer & Hannes
Rieser, 38-74. Berlin: Walter De Gruyter.
Kratzer, Angelika. 1986. “Conditionals.” In Papers from the Parasession on Pragmatics and
Grammatical Theory, ed. Anne M. Farley, Peter Farley, and Karl Eric McCollough,
115-35. Chicago: The Chicago Linguistic Society.
Kratzer, Angelika. 1991. “Modality.” In Semantics: An International Handbook of
Contemporary Research, ed. Armin von Stechow and Dieter Wunderlich, 639-50.
Walter De Gruyter.
Lennertz, Benjamin. Forthcoming a. “Quantificational Credences.” Philosophers’ Imprint.
Lennertz, Benjamin. Forthcoming b. “Taking ‘Might’-Communication Seriously.” Analytic
Philosophy.
Lennertz, Benjamin. MS a. “Asymmetric Disagreement and Epistemic Evaluation.”
Lennertz, Benjamin. MS b. “Generalizing Conditional Attitudes.”
Lennertz, Benjamin. MS c. “Simple Contextualism about Epistemic Modals is Incorrect.”
Levi, Isaac. 1980. The Enterprise of Knowledge. Cambridge: MIT Press.
Lewis, David. 1976. “Probabilities of Conditionals and Conditional Probabilities.” The
Philosophical Review 85: 297-315.
Lewis, David. 1979. “Attitudes De Dicto and De Se.” The Philosophical Review 88: 513-543.
MacFarlane, John. 2003. “Future Contingents and Relative Truth.” Philosophical Quarterly
53: 321-36.
MacFarlane, John. 2005a. “Making Sense of Relative Truth.” Proceedings of the Aristotelian
Society 105: 321-39.
MacFarlane, John. 2005b. “The Assessment Sensitivity of Knowledge Attributions.” In
Oxford Studies in Epistemology 1, ed. Tamar Szabó Gendler and John Hawthorne,
197-233. Oxford: Oxford University Press.
MacFarlane, John. 2008a. “Nonindexical Contextualism.” Synthese 166: 231-50.
MacFarlane, John. 2008b. “Truth in the Garden of Forking Paths.” In Relative Truth, ed.
Manuel Garcia-Carpintero and Max Kölbel, 81-102. Oxford: Oxford University Press.
MacFarlane, John. 2011a. “Epistemic Modals are Assessment Sensitive.” In Epistemic
Modality, ed. Andy Egan and Brian Weatherson, 144-78. Oxford: Oxford University
178
Press.
MacFarlane, John. 2011b. “Simplicity Made Difficult.” Philosophical Studies 156: 441-48.
MacFarlane, John. 2014. Assessment Sensitivity: Relative Truth and its Applications. Oxford:
Oxford University Press.
MacFarlane, John. MS. “Varieties of Disagreement.”
Maher, Patrick. 1990. “Acceptance without Belief.” PSA: Proceedings of the Biennial Meeting
of the Philosophy of Science Association, Volume One: 381-92.
Marushak, Adam, and James Shaw. MS. “Epistemic Expressivism, Attitudes, and Evidence.”
Montminy, Martin. 2012. “Epistemic Modals and Indirect Weak Suggestives.” Dialectica 66:
583-606.
Moran, Richard. 1988. “Making Up Your Mind: Self-Interpretation and Self-Constitution.”
Ratio 1: 135-51.
Morgan, Jerry L. 1975. “Some Interactions of Syntax and Pragmatics.” In Syntax and
Semantics vol. 3: Speech Acts, ed. Peter Cole and Jerry L. Morgan, 43-58. New York:
Academic Press.
Morgan, Jerry, L. 1978. “Two Types of Convention in Indirect Speech Acts.” In Syntax and
Semantics Volume 9: Pragmatics, ed. Peter Cole and Jerry L. Morgan, 261-80. New
York: Academic Press.
Olson, Jonas. 2004. “Buck-Passing and the Wrong Kind of Reason.” Philosophical Quarterly
54: 295-300.
Parfit, Derek. 2001. “Rationality and Reasons.” In Exploring Practical Philosophy. From
Action to Values, ed. D. Egonsson et al, 17-39. Aldershot: Ashgate Publishing.
Piller, Christian. 2001. “Normative Practical Reasoning.” The Aristotelian Society
Supplementary Volume 75: 195-216.
Piller, Christian. 2006. “Content-Related and Attitude-Related Reasons for Preferences.”
Royal Institute of Philosophy Supplement 59: 155-82.
Price, Huw. 1983. “Does ‘Probably’ Modify Sense?” Australasian Journal of Philosophy 61:
396-408.
Quine, W.V. 1969. “Propositional Objects.” In Ontological Relativity and Other Essays, 139-
60. New York: Columbia University Press.
Rabinowicz, Wlodek, and Toni Rønnow-Rasmussen. 2004. “The Strike of the Demon: On
Fitting Pro-attitudes and Value.” Ethics 114: 391-423.
179
Reisner, Andrew. 2009. “The Possibility of Pragmatic Reasons for Belief and the Wrong Kind
of Reasons Problem.” Philosophical Studies 145: 257-72.
Ridge, Michael. 2006. “Ecumenical Expressivism: Finessing Frege.” Ethics 116: 302-36.
Ross, Jacob, and Schroeder, Mark. 2013. “Reversibility or Disagreement.” Mind 122: 43-84.
Sadock, Jerrold. 1974. Towards a Linguistic Theory of Speech Acts. New York: Academic
Press.
Schaffer, Jonathan. 2011. “Perspective in Taste Predicates and Epistemic Modals.” In
Epistemic Modality, ed. Andy Egan and Brian Weatherson, 179-226. Oxford: Oxford
University Press.
Schroeder, Mark. 2010. Noncognitivism in Ethics. London: Routledge.
Schroeder, Mark. 2012a. “Stakes, Withholding, and Pragmatic Encroachment on Knowledge.”
Philosophical Studies 160: 265-85.
Schroeder, Mark. 2012b. “The Ubiquity of State-Given Reasons.” Ethics 122: 457-88.
Schroeder, Mark. 2013. “State-Given Reasons: Prevalent if not Ubiquitous.” Ethics 124: 128-40.
Schroeder, Mark. MS. “Attitudes and Epistemics.”
Schnieder, Benjamin. 2010. “Expressivism Concerning Epistemic Modals.” Philosophical
Quarterly 60: 601-15.
Schulz, Moritz. 2010. “Wondering What Might Be.” Philosophical Studies 149: 367-86.
Searle, John. 1975. “Indirect Speech Acts.” In Syntax and Semantics vol. 3: Speech Acts, ed.
Peter Cole and Jerry L. Morgan, 59-82. New York: Academic Press.
Shah, Nishi. 2006. “A New Argument for Evidentialism.” The Philosophical Quarterly 56:
481-98.
Simons, Mandy. 2010. “A Gricean View on Intrusive Implicatures.” In Meaning and Analysis:
New Essays on H. Paul Grice, ed. Klaus Petrus, 138-69. New York: Palgrave
Macmillan.
Skorupski, John. (2007). Buck-Passing about Goodness. In Hommage à Wlodek:
Philosophical Papers Dedicated to Wlodek Rabinowicz, ed. Toni Rønnow-Rasmussen
et al, 1-15.
Soames, Scott. 2006. “Understanding Assertion.” In Content and Modality: Themes from the
Philosophy of Robert Stalnaker, ed. Judith Thomson and Alex Byrne, 222-50 . Oxford:
Clarendon Press.
Soames, Scott. 2010. “True At.” Analysis 71: 124-33.
180
Sorenson, Roy. 2009. “Meta-agnosticism: Higher Order Epistemic Possibility.” Mind 118: 777-
84.
Stalnaker, Robert. 1975. “Indicative Conditionals.” Philosophia 5: 269-86.
Stalnaker, Robert. 1978. “Assertion.” Syntax and Semantics 9: 315-32.
Stalnaker, Robert. 1984. Inquiry. Cambridge: MIT Press.
Stanley, Jason. 2005a. “Fallibilism and Concessive Knowledge Attributions.” Analysis 65: 126-
31.
Stanley, Jason. 2005b. Knowledge and Practical Interests. Oxford: Oxford University Press.
Stephenson, Tamina. 2007. “Judge Dependence, Epistemic Modals, and Predicates of
Personal Taste.” Linguistics and Philosophy 30: 487-545.
Stevenson, Charles. 1937. “The Emotive Meaning of Ethical Terms.” Mind 46: 14-31.
Stratton-Lake, Phillip. 2005. “How to Deal with Evil Demons: Comment on Rabinowicz and
Rønnow‐Rasmussen.” Ethics 115: 788-98.
Sturgeon, Scott. 2008. “Reason and the Grain of Belief.” Noûs 42: 139-65.
Swanson, Eric. 2006. Interactions with Context. Ph.D. Dissertation, MIT.
Swanson, Eric. 2010. “On Scope Relations between Quantifiers and Epistemic Modals.”
Journal of Semantics 27: 529-40.
Swanson, Eric. 2011. “How Not to Theorize about the Language of Subjective Uncertainty.” In
Epistemic Modality, ed. Andy Egan and Brian Weatherson, 249-69. Oxford: Oxford
University Press.
Swanson, Eric. Forthcoming. “The Application of Constraint Semantics to the Language of
Subjective Uncertainty.” Journal of Philosophical Logic.
Toulmin, Stephen. 1958. The Uses of Argument. Cambridge: Cambridge University Press.
van Fraassen, Bas. 1980. The Scientific Image. Oxford: Clarendon Press.
van Roojen, Mark. 1996. “Expressivism and Irrationality.” The Philosophical Review 105: 311-
35.
Veltman, Frank. 1996. “Defaults in Update Semantics.” Journal of Philosophical Logic 25: 221-
61.
Way, Jonathan. 2012. “Transmission and the Wrong Kind of Reason.” Ethics 122: 489-515.
Wedgwood, Ralph. 2012. “Outright Belief.” Dialectica 66: 309-29.
White, Roger. 2010. “Evidential Symmetry and Mushy Credence.” In Oxford Studies in
Epistemology Vol. 3, ed. Tamar Szabó Gendler and John Hawthorne, 161-88. Oxford:
181
Oxford University Press.
Willer, Malte. 2011. “Realizing What Might Be.” Philosophical Studies 153: 365-75.
Willer, Malte. 2013. “Dynamics of Epistemic Modality.” The Philosophical Review 122: 45-92.
Yalcin, Seth. 2007. “Epistemic Modals.” Mind 116: 983-1026.
Yalcin, Seth. 2010. “Probability Operators.” Philosophy Compass 5: 916-37.
Yalcin, Seth. 2011. “Nonfactualism about Epistemic Modals.” In Epistemic Modality, ed. Andy
Egan and Brian Weatherson, 295-332. Oxford: Oxford University Press.
Zimmermann, Thomas Ede. 2000. “Free Choice Disjunction and Epistemic Possibility.”
Natural Language Semantics 8: 255-90.
182

Abstract (if available)

Abstract My dissertation presents complementary views of reasoning with uncertainty and communicating in situations of reasoning with uncertainty. Reasoning with uncertainty proceeds from the possibilities an agent takes as options, rather than those that are merely consistent with her information or that she has positive credence in. I develop a descriptive and normative picture of such reasoning. This view of reasoning informs my account of epistemic communication, according to which agents typically use epistemic 'might'-sentences to coordinate collective reasoning with uncertainty. They do so by indirectly expressing attitudes of taking a proposition as an option in reasoning. This view of the pragmatics of epistemic communication is compatible with the traditional theory of the semantics of such sentences. Furthermore, it is able to solve a number of challenges raised for that theory by focusing on what speakers typically use epistemic sentences to do—rather than focusing on the meanings of these sentences.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Telling each other what to do: on imperative language

PDF

Reasons, obligations, and the structure of good reasoning

PDF

Contrastive reasons

PDF

Positivist realism

PDF

Meaningfulness, rules, and use-conditional semantics

PDF

Constraining assertion: an account of context-sensitivity

PDF

Iffy confidence

PDF

Conceptually permissive attitudes

PDF

Reasoning with degrees of belief

PDF

Public justification beyond legitimacy

PDF

The mental states first theory of promising

PDF

What "ought" ought to mean

PDF

Linguistic understanding and semantic theory

PDF

A deontological explanation of accessibilism

PDF

The virtue of reasonableness: on the normative epistemic implications of public reason liberalism

PDF

Rationality and the primacy of the occurrent

PDF

Process-oriented rationality

PDF

A reduplicative analysis of sentence modal adverbs in Spanish

PDF

The minimal approval view of attributional-responsibility

PDF

Beliefs that wrong

Asset Metadata

Creator Lennertz, Benjamin (author)

Core Title Reasoning with uncertainty and epistemic modals

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Philosophy

Publication Date 06/23/2014

Defense Date 05/14/2014

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag contextualism,epistemic modals,OAI-PMH Harvest,pragmatics,uncertainty

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Schroeder, Mark (committee chair), Soames, Scott (committee chair), Finlay, Stephen (committee member), Schein, Barry (committee member)

Creator Email belennertz@gmail.com,lennertz@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-425853

Unique identifier UC11286675

Identifier etd-LennertzBe-2581.pdf (filename),usctheses-c3-425853 (legacy record id)

Legacy Identifier etd-LennertzBe-2581.pdf

Dmrecord 425853

Document Type Dissertation

Format application/pdf (imt)

Rights Lennertz, Benjamin

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA