i i

OUP CORRECTED PROOF – FINAL, 20/9/2017, SPi

i i

Mood

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 20/9/2017, SPi

i i

OXFORD SURVEYS IN SEMANTICS AND PRAGMATICS

general editors: Chris Barker, New York University, and Chris Kennedy,
University of Chicago
advisory editors: Kent Bach, San Francisco State University; Jack
Hoeksema, University of Groningen; Laurence R. Horn, Yale University; William
Ladusaw, University of California Santa Cruz; Richard Larson, Stony Brook
University; Beth Levin, Stanford University; Mark Steedman, University of
Edinburgh; Anna Szabolcsi, New York University; Gregory Ward, Northwestern
University

published
1 Modality
Paul Portner
2 Reference
Barbara Abbott
3 Intonation and Meaning
Daniel Büring
4 Questions
Veneeta Dayal
5 Mood
Paul Portner

in preparation
Aspect
Hana Filip
Lexical Pragmatics
Laurence R. Horn
Conversational Implicature
Yan Huang

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 20/9/2017, SPi

i i

Mood

PAUL PORTNER

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 20/9/2017, SPi

i i

3
Great Clarendon Street, Oxford, ox2 6dp,
United Kingdom
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide. Oxford is a registered trade mark of
Oxford University Press in the UK and in certain other countries
© Paul Portner 2018
The moral rights of the author have been asserted
First Edition published in 2018
Impression: 1
All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by licence or under terms agreed with the appropriate reprographics
rights organization. Enquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above
You must not circulate this work in any other form
and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available
Library of Congress Control Number: 2017939633
ISBN 978–0–19–954752–4 (hbk.)
978–0–19–954753–1 (pbk.)
Printed and bound by
CPI Group (UK) Ltd, Croydon, cr0 4yy
Links to third party websites are provided by Oxford in good faith and
for information only. Oxford disclaims any responsibility for the materials
contained in any third party website referenced in this work.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 20/9/2017, SPi

i i

Contents
General preface vii
Acknowledgments viii
List of figures and tables ix

 Introduction 
. What do we study when we study mood? 
.. Conceptual preliminaries 
.. The general concept of mood 
. Main findings about the nature of mood 
. Background on modality 
.. Classifications of modality 
.. Modality in possible worlds semantics 
. The flow of information in discourse 
.. The dynamic approach 
.. Speech act theory 
.. Update potential and illocutionary force 
. Looking ahead 
 Verbal mood 
. Subsentential modality 
. Indicative and subjunctive 
.. Ideas about the indicative/subjunctive contrast 
.. Semantic theories of verbal mood in complement clauses 
.. Clauses which are not complements to a selecting predicate 
. Beyond verbal core mood 
.. Other mood-indicating forms 
.. The roles of semantics, syntax, and non-grammatical factors 
 Sentence mood 
. Sentence mood, clause type, and sentential force 
.. Clause types as grammatical categories 
.. Sentential forces as pragmatic categories 
.. The syntax/sentence mood interface 
. Sentence mood in speech act theory 
.. The performative hypothesis 
.. Adjustments to classical speech act theory 
.. The dynamic force hypothesis in speech act theory 
. Sentence mood in the dynamic approach 
.. Declaratives in the dynamic approach 
.. Interrogatives in the dynamic approach 
.. Imperatives in the dynamic approach 
.. Minor types: optatives and exclamatives 

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 20/9/2017, SPi

i i

vi contents

. Theories of clause type systems and sentence mood 

.. Clause type systems in speech act theory 
.. Clause type systems in the dynamic approach 
.. Other properties of clause type systems 
. Looking ahead 
 Core mood, reality status, and evidentiality 
. Prospects for a unified theory of core mood 
. Reality status and evidentiality 
.. Reality status 
.. Evidentiality 
.. Final remarks 

References 
Index 

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 20/9/2017, SPi

i i

General preface
Oxford Surveys in Semantics and Pragmatics aims to convey to the reader the
life and spirit of the study of meaning in natural language. Its volumes provide
distillations of the central empirical questions driving research in contemporary
semantics and pragmatics, and distinguish the most important lines of inquiry
into these questions. Each volume offers the reader an overview of the topic at
hand, a critical survey of the major approaches to it, and an assessment of what
consensus (if any) exists. By putting empirical puzzles and theoretical debates into a
comprehensible perspective, each author seeks to provide orientation and direction
to the topic, thereby providing the context for a deeper understanding of both
the complexity of the phenomena and the crucial features of the semantic and
pragmatic theories designed to explain them. The books in the series offer research-
ers in linguistics and related areas—including syntax, cognitive science, computer
science, and philosophy—both a valuable resource for instruction and reference
and a state-of-the-art perspective on contemporary semantic and pragmatic theory
from the experts shaping the field.
Paul Portner’s survey on mood provides a welcome new platform for the work
on a major but understudied topic in the semantics of natural language. The vast
majority of modern semantic studies have focused on the ways in which the
morphosyntactic and semantic properties of the constituent elements of declarat-
ive sentences—and to a somewhat lesser extent, interrogatives and imperatives—
interact with each other and with aspects of the context of utterance to determine
truth-conditional content. However, as an increasingly large number of scholars
have begun to appreciate, the full set of linguistically marked distinctions in
clause type is finer-grained than the traditional three-way distinction between
declarative, interrogative, and imperative reflects, richly subtle in its detail, and
indicative of a systematic relation between clausal and verbal morphosyntax and
constraints on the ways that a sentence can be used to perform a speech act.
Comprehensive, careful, instructive, and insightful, Portner thoroughly explores
the intricate tension between clausal and verbal morphosyntax and illocutionary
content, covering performativity, speech act theory, dynamic update, and eviden-
tiality. This volume bridges traditional studies of clause typing and contemporary
semantic and pragmatic theory, and lays the foundation for future advances in our
study of the relation between sentential morphosyntax and illocutionary force.
Chris Barker
New York University
Christopher Kennedy
University of Chicago

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 20/9/2017, SPi

i i

Acknowledgments
The idea for this book goes back to a proposal I made in 2003 to Oxford University
Press for a volume on modality and mood, and from that time I have appreciated
the patience and support of both the editors of the Oxford Surveys series, Chris
Barker and Chris Kennedy, and the editors at the press, especially John Davey and
Julia Steer. It turned out that it was far from feasible to include my thoughts on
mood and modality, and the relation between them, in a single volume. I am very
grateful that the press allowed me to pursue this project in two parts. I revised the
portion of the manuscript on modality into Modality (Portner 2009), and then set
out to write a second volume on mood. This latter part of the project turned out
to be much more difficult, because the range of relevant ideas in the literature is so
much more varied and disconnected, but for me the process of engaging with that
literature has also led again and again to the feeling that a new insight is waiting to
be understood. I hope that readers will experience to some extent the benefits, as I
have, of finding new ideas and connections within the literature on mood.
As I wrote the book I continued my various research projects on modality and
mood, and my collaborators, discussants, and reviewers during that time have
helped me immensely in this project. I can mention especially my collaborat-
ors Graham Katz, Elena Herburger, Miok Pak, Aynat Rubinstein, and Raffaella
Zanuttini. I received helpful comments on Chapter 3 from Malte Willer and on the
entire manuscript from an anonymous reviewer. In presentations of ideas which
flow from the perspective on mood adopted in this book, I have received crucial
feedback from Maria Aloni, Gennaro Chierchia, Nate Charlow, Liz Coppock, Kai
von Fintel, Anastasia Giannakidou, Magda Kaufmann, Angelika Kratzer, Alda
Mari, Craige Roberts, and Steve Wechsler. This feedback has in many cases led
me to rethink my understanding of the literature and trends present in current
research. I thank my students for sharing their insights and for help with research
and editing, in particular Lissa Krawczyk, Hillary Harner, and Akitaka Yamada.
My family has remained confident in my ability to complete this project and,
amazingly, convinced of its value through the long years that it hung about the
house as a competitor for my attention. I love the fact that my kids Noah and
Ben believe in the importance of research and writing, and I deeply appreciate the
respect my wife Sylvia holds for any project which has value to me. The feeling that
what one of us cares about, we all do, has given me the confidence to move forward
whenever the project seemed too complex and difficult. I dedicate the book to my
family and especially to Sylvia.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 20/9/2017, SPi

i i

List of figures and tables

Figures
1.1 Strength via accessibility relation 17
1.2 Modal base and ordering source 19
2.1 Entailment pattern under strong modal analysis 53
3.1 The question as function from worlds to answers 191
3.2 The question as a partition 191
3.3 The context as a partition 195
3.4 Asking a question in the partition context 196
3.5 The inquisitive context 198

Tables
1.1 Semantic classifications for mood and modality 7
1.2 Versions of the dynamic approach 28
3.1 Terminology for sentence moods 124
3.2 Structured discourse context of Portner (2004) 181
3.3 Main contributions of dynamic theories of imperatives 215

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 20/9/2017, SPi

i i

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

1
Introduction

Mood is a fundamental and traditional concept used in the description of human

language and in theoretical inquiries into the nature of language. Because it is
fundamental and traditional, one might think that the theory of mood is an
important topic within linguistics. Yet mood as such, in its most general sense, is
not a topic of research, and there are several reasons for this. One is that the concept
of mood is so fundamental that different scholarly traditions have molded their
understanding of what mood is to fit their conception of the nature of language.
We see it used in different ways in philosophy, formal semantics and pragmatics,
and descriptive/typological linguistics. And even setting aside broad differences in
approach to the study of language, mood is a difficult topic because all linguists
agree that it relates to at least the two different grammatical phenomena: verbal
mood, the category which includes indicatives and subjunctives, and sentence
mood, the category which includes declaratives, interrogatives, and imperatives.
It has generally not seemed likely that a general theory of mood incorporating
both of these categories would be possible, and so, in many instances, linguists take
care to explain that they are talking about one kind of mood and not the other. It
will emerge as we go forward, however, that similar ideas and puzzles recur as we
try to understand the two kinds of mood. They might not be as different as we
often assume.
In writing this book, I have two primary goals. The first is to provide scholars
interested in the semantics and pragmatics of mood with a deeper source of
background than is found in any single previous work. In doing this, I draw on as
much of the literature on mood as I am able, but I think and write from within the
formal semantics tradition. And second, I seek to formulate and advance new ideas
about the semantics and pragmatics of mood. As I survey, organize, and explain
the literature, I will sometimes sketch out novel analyses. In some cases, I do this
as a model for discussion without entirely endorsing the analysis, while in others,
I advocate for a certain hypothesis.
It must not go without saying that there is a great deal of insightful research
on mood which will not be mentioned in this book. The fields of linguistics and
philosophy are too large and the concepts of mood are too useful—they lead to
such a vast literature that even the subset of it that is worth studying today is so
extensive that no single book can incorporate a responsible discussion of it all.
What is included here is the result of my best judgment about what combination
of literature, explanation, and new ideas can best further the goals of the book.

Mood. First edition. Paul Portner.

© Paul Portner 2018. First published 2018 by Oxford University Press.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

2 introduction

There is a great deal of additional important scholarship on mood which I would

recommend. I hope that this book can help readers make better use of it.

Some terms and conventions. Before beginning the substantive discussion,

I would like to specify some of my terminology and conventions.

1. Notational conventions. In mentioning a word, phrase, or sentence:

(a) The use of small caps indicates a technical term whose precise meaning
is being discussed in the immediate context; in many cases, the term is
actually defined there.
(b) The use of boldface merely serves to draw the reader’s attention to a
particular word or words.
(c) The use of italics indicates that the material is being referred to as a
linguistic object. Exception: Within numbered examples and figures,
such material will not be italicized, unless there is other material within
the numbered example which is not being referred to as a linguistic
object.
(d) The use of ‘single quotes’ indicates that the material stands for the
linguistic objects in a number of languages, under the assumption that
all of those objects are similar enough across the relevant languages to be
spoken of as if they are the same. Example: “In most Romance languages,
‘believe’ selects the indicative.” Single quotes are also used for glosses and
paraphrases.
(e) The use of “double quotes” indicates that the quoted material should be
understood as spoken or written by someone. Double quotes are also
used in the traditional way as “scare quotes,” indicating that the quoted
material should not necessarily be taken literally.
(f) [Brackets] indicate some aspect of the context in which an example is to
be understood.
2. I use the term logical form in two ways:
(a) When capitalized, as “Logical Form,” it refers to a level of linguistic repres-
entation at which certain semantic properties, such as scope, are explicitly
represented. Logical Form can be a level of syntax or an independent level
between syntax and semantics.
(b) When not capitalized, as “logical form,” it refers to a representation
which displays or elucidates important abstract properties of meaning.
The logical form usually has some logic-like properties, such as a simple
vocabulary and an unintuitive syntax. Unlike Logical Form, a logical form
is not understood to be a level of linguistic representation.
3. When using a pronoun anaphoric to an antecedent which could refer to
a male or female individual, I roughly alternate genders. Example: “If a
philosopher reads this, she might disagree” (next time “he”). Occasionally,
however, it will be better to use “he or she,” “s/he,” or singular “they.”

In difficult cases, I will apply these conventions in the way I feel is most helpful.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

what do we study when we study mood? 3

1.1 What do we study when we study mood?

1.1.1 Conceptual preliminaries
One of the most fundamental issues to be addressed by semantic theory concerns
the kind of meanings that simple declarative sentences have, and we can judge
any approach by how well it helps us understand such basic properties as: they
can be (and typically are) either true or false; they can be used to share factual
information; they are appropriate for representing the contents of thought and
some other cognitive states; and they stand in logical relations to one another like
entailment and synonymy. Consider the declarative sentence (1a). It has all of these
properties:

(1) a. Ben is holding a bird.

b. This is true because: Ben is holding John Fei.
c. Noah told me: “Ben is holding a bird.” And now I know what Ben is
holding.
d. Noah thinks that Ben is holding a bird.
e. Ben is holding a bird. Entails: Ben is in close proximity to a bird.

Moreover, semanticists think they have a good (though not yet perfect) under-
standing of how each of these properties should be explained, an understanding
which is based on the theoretical construct of a proposition.
In formal semantics, our theories of sentence meaning most commonly work
with the idea that propositions can be defined in terms of possible worlds. A
possible world is a way things could be, complete through space and time, an
alternative history of the universe. Our own universe-history can be referred to
as the “actual world” or “real world.” For the purposes of linguistic semantics,
let us assume that we have a set of possible worlds conceivable by humans. By
this I mean that any difference in how things could be which a human could
recognize, imagine, or describe corresponds to a difference between possible
worlds in the set of all worlds W. For example, if I tell you that Ben has a
cockatiel, you can imagine that it is grey, or that it is white. Therefore, W
should contain at least one possible world in which it is grey and at least one
in which it is white. The set of worlds conceivable by humans seems sure to be
adequate for doing natural language semantics, if any theory based on possible
worlds is.
A proposition in possible worlds semantics is a subset of W. For example, the
meaning of (1a) is, or can be characterized in terms of, the set (2):

(2) {w : Ben is holding a bird in w}

(= the set of worlds in which Ben is holding a bird.)

This definition of proposition helps to explain the properties summarized in (1).

For example, (1b) amounts to the claim that the actual world is one of the worlds
in the set. Though the conception of propositions as sets of possible worlds is not

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

4 introduction

without its problems, we can ignore them for now. Our goal is to use it to develop
a useful way of thinking about mood.
Another feature of language we need to understand if we are to get a handle
on the concept of mood is modality. Modality is “the linguistic phenomenon
whereby grammar allows one to say things about, or on the basis of, situations
which need not be real” (Portner 2009, p.1). The noun Ben in (1a) is not modal; it
is used to refer to a particular, actual person. In contrast, the auxiliary should in (3)
is modal:

(3) Ben should put down the bird.

This sentence says that situations in which Ben puts down the bird are in some
respect better than, or preferable to, situations in which he does not. Since he either
will or will not put down the bird, some of these situations will never be real, and
so the word should counts as modal.

1.1.2 The general concept of mood

With these preliminaries in place, we can attempt a definition of mood:

Mood is an aspect of linguistic form which indicates how a proposition is used

in the expression of modal meaning.

As we start out, it’s acceptable to be vague about what we mean by the “use” of a
proposition, because linguists have employed the term “mood” in many different
ways. Let us consider two examples:

1. In many languages, subordinate clauses (and sometimes root clauses as well)

come in different forms, known as verbal moods, such as indicative mood
and subjunctive mood. These clauses can be used to help represent various
cognitive states and mental events, such as beliefs, desires, and dreams. When
they are used in a subordinate clause to help talk about a desire (as in (4)),
they typically take the subjunctive form; in contrast, when they are used to
describe a dream ((5)), they typically take indicative form.

(4) Pierre veut que Marie soit heureuse. (French)

Pierre wants that Marie [Link] happy
‘Pierre wants Marie to be happy.’
(5) J’ai rêvé qu’il était président.
I dreamed that he [Link] president
‘I dreamed that he was president.’

 Under certain circumstances, they may take a different form, such as infinitive. For our purposes

here, it’s helpful to pretend there are only two possibilities.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

what do we study when we study mood? 5

We call this kind of mood “verbal mood” because it is frequently (though not
always) marked in a grammatical sense on the verbal head of the clause; for
example, the verb soit is a subjunctive form of ‘be’ in French.
2. In all languages, root sentences have various functions, including easy-to-
intuit ones like directing somebody to do something (as in (6a)), requesting
information ((6b)), and providing information ((6c)).

(6) a. Pick up a bird!

b. Is Ben holding a bird?
c. Ben is holding a bird.

Each of these functions is associated in an intuitively clear way (which is

nonetheless hard to define) with particular members of a paradigm of forms
known as sentence moods. The most prominent sentence moods are the
three just illustrated: the imperative, interrogative, and declarative.

Each of these two important concepts of mood realizes the general description of
mood indicating how a sentence’s proposition is to be used: verbal mood tells us
something about how it is to be used, within the compositional computation of
meaning, to describe an individual’s mental life, while sentence mood indicates
how it is to be used, in a multi-party exchange, to achieve specified communicative
functions.
Given an understanding of mood like the one just developed, what would a
linguistic theory of mood look like? We could spell out a too-simple theory of
verbal mood like this: Assume that verbs like want and dream, verbs which take
sentential complements and talk about some aspect of their subject’s mental life,
are modal words which express a relation between two arguments, an individual
and a proposition. Their logical form can be represented as V(x, p), for example
wants(Pierre, {w : Marie is happy w}). Mood choice is determined by the following
principle:

(7) If the relation expressed by the verb concerns a preference about how the
future will be, the clause which denotes the proposition argument of this
relation should be in the subjunctive mood. Otherwise, it should be in the
indicative mood.

Given this principle, if you want to say something which means wants(Pierre,
{w : Marie is happy w}), the verb ‘be’ will be subjunctive, but if you want to say
something which means dreams(I, {w : Pierre was president w}), it will be indicative.
Nothing semantic changes between the two cases, other than the main relation,
wants or dreams.
And we could spell out a too-simple theory of sentence mood like this: Assume
that the function of a root sentence used in dialogue is to adjust the speaker’s and
hearer’s shared assumptions. For example, sometimes the speaker may want to
create a shared assumption that one way the future could be is preferable to another,
perhaps threatening some sort of punishment upon the addressee if the preferred

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

6 introduction

future does not come about. We can describe this as the speaker directing the
addressee to do something. There is a particular grammatical form, the imperative,
for a sentence used with this purpose, as in (6a). This reasoning suggests the
following principle:

(8) If a sentence is used to create an assumption that it is preferable that the

proposition expressed by the sentence comes to be true (as opposed to false),
it should be expressed using imperative mood.

There would be similar principles for interrogative sentences (Will Ben put down
the bird? creates an assumption that the addressee will help the speaker know some-
thing) and indicative sentences (Ben will put down the bird creates an assumption
that a certain fact holds). Nothing would differ among these cases in terms of the
proposition involved, but there would be crucial differences in the goal which the
speaker aims to achieve by using a sentence which denotes that proposition.
These pictures of subjunctive and imperative clauses have been presented to
help convey an understanding of the idea behind the informal definition of mood.
This conception of mood will be very important in the book, as it serves as an
unarticulated intuition behind the actual practices linguists have in describing
phenomena as “mood,” and because it plays a role in many attempts to provide
concrete semantic or pragmatic analyses of mood forms. However, the specific
statements about the subjunctive and imperative above are not to be taken as
serious proposals. Besides oversimplifying the relevant phenomena, they make
many assumptions which could turn out to be wrong: for example, they assume that
all sentences have propositions as their basic meanings and that the various moods
have no effect themselves on the meaning of a given sentence. All such assumptions
must be carefully evaluated as part of any serious investigation of mood. Much
of the work in this book is to examine some of the phenomena which meet the
characterization of mood based on the concept of “modal use” and to consider
various theories of them. In the next subsection, I give a preview of where this way
of thinking about mood will lead us.

1.2 Main findings about the nature of mood

The literature on mood is extensive, contradictory, and at times confusing.
Nevertheless, I hope we won’t end up in a muddle. The discussion will lead not only
to lots of questions, but to a few linguistically significant conclusions as well. Here
at the outset, I would like to highlight the work’s most basic, broadly conceptual
findings:

1. While verbal mood and sentence mood are distinct both in terms of
morphosyntax and in terms of meaning, they are closely related. They
are related because of tight parallels between the modal semantics of
sentence-embedding constructions, which determine verbal moods, and the

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

main findings about the nature of mood 7

communicative functions served by root sentences in conversation, in terms

of which sentence moods are characterized.
2. At a formal level, many theories of verbal mood and sentence mood can be
expressed within a unified framework. This framework states that both kinds
of mood encode the interaction between a clause’s meaning and a contextually
given partially ordered set of worlds (the posw). The posw may be
derived from either the grammatical or the discourse context. Though the
posw-framework is clearly not the last word on the relation between verbal
mood and sentence mood, it can point the way towards a more general and
explanatory theory of the two kinds of mood and the relation between them.
On the assumption that some types of verbal mood and some types of
sentence mood can be analyzed in terms of a single theoretical system (be
it the posw or something else), I label the broader linguistic category which
includes them both core mood.
3. There is a range of forms and constructions across languages whose relation
to core mood is not currently well understood. Among these are evidentials,
the kinds of forms which Elliott (2000) describes as marking “reality status,”
and numerous other specific moods and modes used in the description
of languages typologically different from those which have been the focus
of research on verbal mood. We do not yet have a good enough under-
standing of all the forms with labels like “realis,” “irrealis,” “conjunct,” “in-
ferential,” “hearsay,” “conjecture,” “potential,” “desiderative,” “assertive,” and
“contingent” to say how they are related to core mood.

I think that core mood covers more than what linguists typically think of as mood,
rather than less, and I don’t know of any linguistic forms which at once should
clearly be classified as mood, yet also clearly not as core mood. However, some types
of elements, for example evidentials, which I think could reasonably be thought
of as core mood, are not typically thought about in that way, and so they may
exemplify non-core mood. More significantly, few phenomena which have been
described in terms of the concepts of reality status have been analyzed in a precise
enough way for it to be clear what their relation to core mood is. Table 1.1 outlines

Table . Semantic classifications for mood and modality

Modality
Core mood Non-core mood (The rest of modality)
Verbal Sentence Other core
mood mood mood

indicative, imperative, evidentials? evidentials? epistemic, priority,

subjunctive, declarative, deontic, dynamic
certain interrogative reality status? reality status? modals; modal
infinitives adjectives, adverbs

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

8 introduction

this way of thinking about the relationships among various kinds of mood and
between mood and modality.

1.3 Background on modality

As described above, for the purposes of this book I define modality as the linguistic
phenomenon whereby grammar allows one to say things about, or on the basis of,
situations which need not be real. Since, on this conception, mood fits within this
broader category of modality, it will be essential to have an understanding of the
semantics of modality. In this section I will provide an overview of some key ideas
in the semantic theory of modality, drawing on the more extensive discussion by
Portner (2009).

1.3.1 Classifications of modality

We begin with two ways of classifying modal expressions: by the linguistic level at
which they are expressed and by the meanings they convey.

Classifications of modality by level of linguistic organization. Modality may

be classified according to the level of linguistic organization on which it operates.
Specifically, it’s useful to distinguish subsentential, sentential, and discourse
modality.

1. Subsentential modality operates below the level of proposition expressed by a

complete sentence. It includes modal adjectives, modal nouns, propositional
attitude verbs (verbs which take an argument which expresses a proposition),
verbal mood, and infinitives, among other types.

(9) (a) A possible solution to this problem is to call the recalcitrant reviewer.
(b) The probability of success is low.
(c) I think/hope/regret that she arrived on time.
(d) I hope to be happy.

Of course, many of these elements also affect the meaning of the complete
sentence, but they do so via the meaning of some smaller constituent. For
example, the subject phrases in (9a–b) have noun phrase-type meanings
which have been built up using the modal concepts expressed by possible and
probability, and the predicate in (9c) denotes a property, like other predicates
do, but this property involves consideration of not-necessarily-real situations
which are important in the speaker’s mental life.
2. Sentential modality operates at the level of the complete proposition. In other
words, if a sentence contains a constituent which denotes a proposition, and
then a modal element combines with this to create another propositional
constituent, we have a case of sentential modality. In grammatical terms,
this means that it is typically realized above the level of the main subject–
predicate structure in the clause, that is, above the S, IP, vP, or other roughly

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

background on modality 9

equivalent syntactic unit. Sentential modality affects the primary meaning of

the sentence; thus, in the case of a root declarative clause, it affects the truth
conditions of the sentence; in the case of a root interrogative, it affects what
information is literally requested, and so forth.
Sentential modality has commanded the vast majority of scholarly atten-
tion on modality within linguistics. In English, there has been a great deal
of work on the modal auxiliaries like can in (10a). We can also classify as
expressing sentential modality such elements as modal verbs (by this I mean
verbs which are not auxiliaries, but which are distinct from regular lexical
verbs as well, as in (10b)), modal adverbs, and some tense and aspect forms.

(10) (a) Noah can swim.

(b) Devo partire domani. (Italian; Squartini 2004)
[Link]-1sg leave tomorrow
‘I must leave tomorrow.’
(c) Probably, Ben was there.
(d) She will be late.
(e) The team is building a bridge.

In some cases, the decision whether to treat a given modal element as

representing subsentential or sentential modality requires a careful syntactic
and semantic analysis. For example, my labeling (10b) as sentential is based
on the assumption that dovere (‘must’) is syntactically distinct from regular
lexical verbs, and the idea that the progressive in (10e) exemplifies sentential
modality assumes that it is a sentence-level operator. If either of these points
turns out not to be correct, then they are not examples of sentential modality.
3. Discourse modality is the direct contribution of modality to meaning in
discourse. I say “direct contribution” because, of course, any type of modality
ultimately contributes to the broader discourse; what distinguishes discourse
modality is that it does not do so via an effect on the regular semantic meaning
expressed by a sentence. Though (10b) may cause the discourse to contain the
information that the speaker favors future situations in which she leaves over
ones in which she does not, this effect is mediated by its ordinary sentence-
level semantics; hence, it is not discourse modality. In contrast, discourse
modality is separate from any modal contribution towards regular sentence
meaning. Some plausible examples of discourse modality are evidentiality,
sentence mood, and the performative meanings of modal auxiliaries and
modal verbs.

(11) (a) Para-sha-n-mi/-si/-chá. (Cusco Quechua, Faller 2006b)

rain-prog-3-bpg/rep/conj
‘It is raining.’ (Direct evidence/Reported/Conjectural.)
(b) Leave right away!
(c) It might rain.

In (11a), we see the three evidential markers in Cusco Quechua, the first
indicating that the speaker has direct evidence (what Faller calls ‘best possible

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

10 introduction

grounds’), the second that he is conveying a report, and the third that he is
making a conjecture. In (11b), we have an imperative. These two exemplify
discourse modality on the assumption that the evidentials and imperative
form do not cause the sentence to denote a modalized proposition—for
example that (11a) does not mean ‘I have direct evidence that it is raining’ (if
it did mean this, it would show sentential modality), but rather conveys this
meaning without affecting the primary proposition ‘it is raining.’ Example
(11c) can be considered discourse modality if we accept the claim (Portner
2008) that it not only means that rain is compatible with our information
(its sentential modality), but also makes the question of whether it will rain a
topic of conversation (additional discourse modality).

The examples of each subtype above are given only in order to help make clear what
should fall under each subcategory. It wouldn’t be surprising if further research
showed some of them to be miscategorized or even not modal at all. For example,
while many think that the progressive is modal, this is somewhat controversial
(Portner 2011a).

Classifications of modality by meaning. Modality involves saying something

about, or on the basis of, situations which need not be real, and it is useful to
classify modal elements according to the way in which its meaning is based on
those situations. Roughly speaking, (12) says that the relevant situations (those in
which the girl gets the prize) are good (perhaps more specifically, they are good in
relation to the criteria set up for the contest):

(12) That girl should get the prize.

In the literature on modality, we find various systems for classifying modal ele-
ments, especially sentential modals, along such parameters; in Portner (2009),
I outline a top-level classification into epistemic, priority, and dynamic modality,
with various subtypes:

1. Epistemic modality has to do with what can be concluded based on

someone’s knowledge.
True epistemic modality is not commonly divided into subtypes, but in
logic and philosophy, there is consideration of such related concepts as
alethic modality (concerning concepts of logical possibility and neces-
sity) and metaphysical modality (concerning metaphysical possibility and
necessity). It is not clear that natural language modal elements ever express
these concepts, except in the context of technical logical/philosophical dis-
course. Historical modality is a kind of metaphysical modality where the
relevant worlds are those with a future that is compatible with everything that
has already occurred, and should perhaps be classified as a kind of dynamic
modality.
2. Priority modality has to do with reasons for assigning priority, or prefer-
ence, to one type of situation over another.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

background on modality 11

There are several subtypes, such as: deontic modality concerns priority
based on rules or right and wrong, buletic modality concerns priority based
on desire, and teleological modality concerns priority based on goals. We
do not assume that these subcategories are mutually exclusive.
3. Dynamic modality has to do with the possible courses of events in the
world, based on the factual circumstances.
The most prominent subtype is volitional modality. Volitional modals
concern the actions available to a volitional individual, with sub-subtypes
including ability modality (focus on the individual’s abilities), opportun-
ity modality (focus on the circumstances surrounding the individual), and
dispositional modality (focus on the individual’s dispositions). There are
also forms of dynamic modality which are not tied to a volitional individual,
and I will call these intrinsic modality. A somewhat special variety of
dynamic modal are the quantificational modals, which seem to involve
quantification over individuals.

Examples of all of these subtypes from Portner (2009, ch.4) are given in (13)–(15):

(13) Epistemic
(a) A typhoon may hit the island.
(b) Mary must have a good reason for being late.
(14) Priority
(a) Deontic: The rich must give money to the poor.
(b) Buletic: You should try this chocolate.
(c) Teleological: You could add some more salt to the soup.
(15) Dynamic
(a) Volitional:
(i) John can swim. (ability)
(ii) You can see the ocean from here. (opportunity)
(iii) Mary will laugh if you tell her that. (dispositional)
(b) Intrinsic:
(i) The cup is breakable.
(ii) Every empire eventually falls. (historical)
(c) Quantificational:
(i) A spider can be dangerous. (existential)
(ii) A spider will be dangerous. (universal)

Following Kratzer (1981, 1991), priority and dynamic modality are often grouped
together as circumstantial modality (i.e. modality which makes reference
to factual circumstances, rather than only an individual’s knowledge or beliefs),
and following this terminology, dynamic modality can be called “pure” cir-
cumstantial modality (circumstantial modality where priorities do not play
a role).
One feature of traditional classifications which might be less than ideal is the
great difference it implies between two uses of words like likely, certain, and chance.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

12 introduction

According to the standard classification above, likely is ambiguous between an

epistemic meaning in (16a) and a dynamic meaning in (16b).

(16) a. Given the evidence, Bob is likely to be hiding in the basement.

b. The way that storm is developing, it is likely to spawn a tornado.

It is useful to have a cover term for types of epistemic and dynamic modality which
express either subjective or objective chance; we can use the term predictive
modality for this class.
I will refer to the differences in meaning along the dimensions outlined in (13)–
(16) as differences of judgment type. (In the literature and especially in the spoken
jargon of semantics, they are often described as “flavors” of modality.) See Portner
(2009) for discussion of other classification schemes. Note that many of the above
examples involve English modal auxiliaries. Other varieties of modality are found
in other languages, and in other constructions within English, but these have not
made their ways into the generally shared terminology of semanticists.
Yet another parameter along which modal meanings vary is that of strength.
Scholars who study modality make at least a two-way distinction between strong
and weak modals, although not every language may actually have modals of both
strengths (Deal 2011). Both should and may can be deontic, but (17a) is stronger
than (i.e. it entails) (17b):

(17) (a) He should vote for her. (strong)

(b) He may vote for her. (weak)

Similar oppositions exist within each of the other subtypes of modality, as can be
seen in (13)–(15). In those examples, must, should, and will would be classified as
strong modals, while the others would be classified as weak. Strong modals are
sometimes called necessity modals, and weak ones possibility modals, on the
grounds that it is necessary that is strong and it is possible that is weak.
When we look beyond modal auxiliaries, it becomes clear that strength is not
a two-way distinction. In the following, (18a) is stronger than (18b), and so forth
down the line:

(18) (a) It will certainly rain.

(b) It will almost certainly rain.
(c) It will probably rain.
(d) There is a reasonable chance that it will rain.
(e) It is just possible that it will rain.

It seems that modal strength is gradable, and it is natural to think of this gradability
as being similar to the gradability of concrete properties, such as height and weight.
(Gradable modality is currently a topic of much study in semantics; see for example

 Rubinstein et al. () find that speakers have significant difficulties making the distinction

between epistemic and dynamic modality in certain texts.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

background on modality 13

Portner 2009; Yalcin 2010; Katz et al. 2012; Klecha 2014; Lassiter 2016.) As we see in
these few examples, variation in strength can arise from a combination of lexical
choice (certainly is stronger than probably) and compositional semantics (almost
certainly is weaker than certainly). Among strong modals, elements including
should and ought are sometimes called “weak necessity modals,” because they feel
weaker than other (“strong”) necessity modals like must. (The terminology is some-
what confusing here, because a weak necessity modal is still a strong, or necessity,
modal by our terminology. A weak necessity modal is the weaker subtype of strong
modal.) It’s not yet clear whether weak necessity modals are logically weaker than
their strong counterparts, or whether they differ in meaning in some other way
(e.g. von Fintel and Iatridou 2008; von Fintel and Gillies 2010; Rubinstein 2012).

1.3.2 Modality in possible worlds semantics

In order to fully engage with the ideas and debates in the analysis of mood which
will occupy the bulk of this book, it is important for readers to have a good
understanding of semantic theories of modality. The goal of Portner (2009) was
to introduce the analyses of modality within four semantic frameworks: modal
logic, Kratzer’s ordering semantics, dynamic semantics, and cognitive semantics.
I cannot recapitulate that discussion here, but instead suggest that readers without
sufficient background begin with that book or other resources on the semantics
of modality. In this section, I will review some of the key ideas of the important
tradition of modal semantics based on possible worlds. This tradition began within
modal logic and has evolved into the standard account within formal (linguistic)
semantics. It is of the most relevance here because the bulk of theoretically-
informed work on mood takes place within this framework. My goal with this
review of possible worlds analysis of modality is to emphasize aspects of the
approach which will prove especially important in the rest of the book, to establish
my own terminological and notational preferences, and to jog the memories of
readers who may know about the semantics of modality, but not have thought
about it in a while.
The standard theory of modality within possible worlds semantics focuses on
sentential modal constructions, and aims to explain their judgment type and
strength.

Judgment type. Within possible worlds semantics, judgment type is determined

by the particular set of possible worlds on which the modal’s meaning is based.
For example, with the epistemic example (13a), the judgment being made is that a

 Subsentential and discourse modality are considered important topics to which the theory should

be extended (and much of this book concerns such attempts), but not exemplars of the standard
theory as it stands. An exception to this statement is subsentential modality that gets treated as if it
were sentential modality, for example the use of modal adjectives with sentential complements: It is
necessary/likely/possible that it is raining. These treatments generally ignore features of the subsentential
modal constructions which are not shared by sentential modals.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

14 introduction

typhoon hitting the island is compatible with the speaker’s knowledge. Suppose we
identify the set K of worlds compatible with what the speaker knows; in that case,
we can express the judgment of (13a) by saying that some of the worlds in K contain
situations in which a typhoon hits the island. In more formal terms, modal logic
bases the semantics of a particular modal element on an accessibility relation
between worlds R. In (13a), we might use Rep :

(19) For any worlds w∗ , and w: Rep (w∗ , w) iff everything the speaker knows in w∗
also holds in w.

Taking w∗ to be the actual world, the set of accessible worlds K is {w : Rep (w∗ , w)}.
We can define the truth conditions of (13a) as follows:

(20) A typhoon may hit the island is true in a world w∗ iff there is some world w
such that both Rep (w∗ , w) and a typhoon hits the island in w.

Though the definitions in (19)–(20) show technically how an accessibility relation

works (and so might be useful for some purposes), it is far too simple. For instance,
it assumes that the only person whose knowledge we care about is the speaker’s
and the only knowledge of his that we care about is that which he has at the present
moment. If there are examples where the knowledge of an individual other than
the speaker, or at a time other than the present, is relevant, we’d need a different
accessibility relation.
Let us examine some data to see who can be the “knower” and what can be the
“knowing time” in cases of sentential epistemic modality.

(21) [Two children are discussing whether the creature they caught is a newt or a
salamander.]
(a) This might be a salamander.
(b) It might have been a salamander.
(c) Ryan said that it might be a salamander.

In (21a), the knower is either the child speaking or the two children jointly, and
the knowing time is the speech time. In (21b), the knower is again the speaker
or two children jointly, while the knowing time could be either the speech time
or some time in the past, for example, when the children still had the creature in
their hands. (To see the latter possibility, consider the continuation . . . but it turns
out it wasn’t.) In (21c), Ryan is the relevant knower, and the knowing time is the
(past) time at which he spoke. Clearly, the details of the accessibility relation can
vary from case to case, and yet this variation is very much limited by grammatical
factors. It would be very difficult for one of the children to use (21a) to make a
modal statement based on what Ryan knew.
It is an important goal for modal semantics to come up with an adequate theory
of the kind of variation illustrated in (21). One way to do this is to incorporate a
context situation s into the definition of the accessibility relation. The knower and
knowing time are extracted from the context situation, while the context situation

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

background on modality 15

itself (as the name suggests) is determined by the linguistic or extralinguistic

context, in ways to be determined:

(22) For any context situation s and worlds w∗ , w: Rep (s)(w∗ , w) iff everything
that the thinking participant(s) of s know in w∗ at the time of s also holds
in w.

There is some redundancy in (22) because the set of accessible worlds depends on
both a context situation s and a world w∗ . But if we assume that each situation is
only part of a single world, and that it only determines accessibility from the world
of which it is a part, we can base the accessibility relation on the context situation
alone, as follows:

(23) For any context situation s and world w: Rep (s, w) iff everything which the
thinking participant (or participants) of s know in s also holds in w.

In root sentences, it is always possible (and usually preferred, if not outright

required) for s to include the speaker as among the thinking participants, and
for the time of s to include the time at which the sentence itself was used. Thus,
the typical interpretation of a root sentence with might concerns the speaker’s
knowledge at the time of utterance. This is what we see in (21a). However, the time
can diverge from the speech time, particularly in the presence of perfect aspect
(have been in (21b)), and it’s even possible for the thinking participants not to
include the speaker, for example in a narrative. In embedded clauses, the situation
s is typically determined by grammatical factors; in (21c), we see the matrix verb
said controlling the interpretation of the modal in the embedded clause by making
sure that Ryan is the thinking participant of s.
Other judgment types can be associated with accessibility definitions similar to
(23), such as the following for a deontic modal used to make a statement about what
the law requires (for example, That guy should put a coin in the parking meter):

(24) For any context situation s and world w: Rlegal (s, w) iff all of the laws in force
in s are fully complied with in w.

We describe a world accessible by the relation associated with an epistemic modal

as an epistemically accessible world. Similarly, we may talk about deontic-
ally accessible worlds, buletically accessible worlds, and so forth. We can
also use less esoteric language like “knowledge worlds” (for epistemically accessible
ones) and “desire worlds” (for buletically accessible ones).

Strength. Differences in strength are analyzed within modal logic and semantic
systems closely based on modal logic as a difference between universal and
existential quantification. The strength of a given modal is also known as its

 Hacquard () argues that the meanings expressed by modal auxiliaries are relative to events in

a way analogous to (). She also assumes that the judgment type can be determined from s, so that a
single general-purpose R can work for all flavors of modality.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

16 introduction

modal force. Within the meaning of the weak modal might in (13a), we find the
phrase “there is some world,” an existential quantifier over accessible worlds. A
strong modal like must, in contrast, is used to make a statement to the effect that all
relevant worlds have a certain property. In other words, it is a universal quantifier
over accessible worlds. The universal meaning of epistemic must in (13b) can be
stated as follows:

(25) Mary must have a good reason for being late, used in a context situation s in
w∗ , is true in w∗ iff, for every w such that Rep (s, w), Mary has a good reason
for being late in w.

This is like (20) except that it involves universal quantification and it has been
improved through the inclusion of the context situation motivated above.
The quantificational analysis of modal strength in standard possible worlds
semantics only can distinguish two strengths, strong (universal quantification) and
weak (existential quantification), and it is clear that this is not enough to explain
the full range of variation in strength. This point is most clear with subsentential
modality, as illustrated by (18) above, but it may also be seen in the relation
between strong and weak necessity modals. Many authors (including von Fintel
and Iatridou 2008; Finlay 2010; Kolodny and MacFarlane 2010; Rubinstein 2012;
Lassiter 2016) discuss the fact that the strong necessity must and have to appear to
be stronger than weak necessity should and ought to (example from von Fintel and
Iatridou):

(26) Everybody ought to wash their hands; employees must.

The simplest approach to explaining intuitive differences in strength beyond the

basic strong–weak contrast focuses on the choice of accessibility relation. The idea
here follows from a logical observation. Suppose we have two accessibility relations
R and R such that the set of worlds accessible by the former is always a subset of
the set of worlds accessible by the latter:

(27) For any context situation s: {w : R(s, w)} ⊆ {w : R (s, w)}.

Then, if we have weak (existential) modal sentences differing in that the first uses
R and the second R , the first will entail the second. The reverse holds for strong
(universal) modal sentences. To see why, consider Figure 1.1, based on Portner
(2009). At a given situation s, R makes accessible the set of worlds indicated by the
arrow labeled R, and R makes accessible a superset of that set. A weak modal is
true, using a given accessibility relation, if some accessible world is one in which
the portion under the scope of the modal (call it S) is true. The portion of the
diagram on the left shows the situation relevant to a weak modal: if S is true at
some world in the smaller set accessible by R, it will obviously be true at some
world accessible by R (at least, the very same world). The right-hand side of the
diagram shows that the opposite holds with a strong modal: if S is true at all worlds
accessible by R , it will obviously also be true at all worlds accessible by R. To
summarize, we have the following:

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

background on modality 17

(28) For any sentence S and any accessibility relations R and R related as in (27):
(a) ♦R (S) entails ♦R (S)
(b) R (S) entails R (S)

(The symbol  is drawn from modal logic to indicate any modal whose meaning is
expressed using universal quantification over worlds, while ♦ is the corresponding
symbol for a modal whose meaning is expressed using existential quantification.)
We might use the observations in (28) to analyze (26) by saying that the accessibility
relation associated with ought makes accessible a subset of the worlds made
accessible by that associated with must. In other words, ought is like R in (28),
while must is like R .
The main issue with trying to explain differences in modal strength in terms of
subset relations among accessibility relations is that it becomes unclear precisely
what content is associated with each relation. In the case of (26), we are to assume
that ought’s R determines a subset of must’s R , but precisely which worlds do R(s)
and R (s) make accessible? Intuitively, we want to say that R(s) makes accessible all
worlds in which both the very important and the less important rules applicable in
s are followed, while R (s) makes accessible all worlds in which the very important
rules in s are followed, but possibly not the less important rules. It would be helpful
if we could explicitly define the accessible worlds for ought and must in terms
of the rules of varying degrees of importance. Kratzer’s premise-based ordering
semantics for modality discussed next allows us to think of differences in modal
strength in precisely this way.

weak modal strong modal

S is true S is true

w∗ R w∗ R

R R

Figure . Strength via accessibility relation

Kratzer’s ordering semantics. In Portner (2009), I presented two versions of

modal logic based on the ideas about modal force and judgment type outlined
above. These theories, a basic modal logic similar to what is taught in introductory
modal logic texts and a “modal logic for linguists” which modifies it to deal with
some of the most obvious differences between natural languages and logical ones,
are useful because they are widely understood and serve as a precise baseline
against which other theories can be measured. While forms of modal logic like
those are useful, they are not taken as realistic theories of modal semantics in
natural language. They simply do not allow one to express in a theoretically
perspicuous way many important ideas about modal semantics and pragmatics.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

18 introduction

For this reason, linguists have extended and modified the framework of modal logic
to produce more linguistically useful theories of modality.
The standard theory of modality within formal semantics was developed
by Kratzer (1977, 1981, 1991, 2012). This influential approach extends the basic
possible worlds analysis of modality in several ways. Perhaps the most important
development is that the modals do not simply classify a possible world as either
wholly accessible or inaccessible, but rather rank or order them according to
some relevant criteria. Exactly which worlds are accessible in a given case can be
determined in a flexible way based on this ordering. We can label Kratzer’s theory
as a prime example of the ordering semantics approach to modality.
Portner (2009) provides a detailed introduction to Kratzer’s theory, but it will be
useful to have a brief review here. One key concept in her framework is that of a
conversational background. A conversational background is a function from
situations to sets of propositions, and can be given by context, linguistic material,
or a combination of the two. A conversational background can serve either of two
basic functions. It can specify a set of relevant worlds or it can define an ordering of
worlds. In its former function, a conversational background is known as a modal
base, while in the latter function, it is known as an ordering source.

(29) Conversational backgrounds

(a) A conversational background is a function from situations to sets of
propositions. (Its domain is S and its range P(W).)
(b) Functions of conversational backgrounds
(i) Modal base. For any situation s, a conversational background m
(functioning
 as a modal base) defines a set of relevant worlds
m(s).
(ii) Ordering source. For any situation s, a conversational background
o (functioning as an ordering source) defines an ordering ≤s , as
follows:
For any worlds w and v, w ≤s v iff, for all p ∈ o(s), if v ∈ p, then
w ∈ p.

The modal base defines as relevant (at a given situation s) those worlds in which all
propositions in the modal base are true. The ordering source ranks worlds so that
one world is ordered ≤ another (with respect to a given situation s) if and only if
all of the propositions in the ordering source which are true in the latter are also
true in the former. So for example:

(30) (a) Modal base: {‘It is hot’, ‘It is sunny’}.

(b) Ordering source: {‘We have cold drinks’, ‘We have a parasol’}.

The relevant worlds defined by the modal base are ones in which it’s both hot and
sunny. The ordering defined by the ordering source ranks those worlds in which

 In Kratzer’s work, a conversational background is a function from worlds to sets of propositions.

I will treat the domain as situations rather than worlds, in order to incorporate the idea of a context
situation.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

background on modality 19

We have cold drinks and a parasol. † Relevant:

w
It’s both hot Ordering w≤v:
and sunny. v

We have We have a parasol;

cold drinks; no cold drinks.
no parasol.

We have neither cold drinks nor parasol.

Figure . Modal base and ordering source

we have both cold drinks and a parasol the highest, worlds in which we have either
cold drinks or a parasol less high, and worlds in which we have neither cold drinks
nor a parasol the lowest. (Worlds in which we have cold drinks but no parasol, and
those in which we have a parasol but no cold drinks, are not ordered with respect
to each other.) These relations are illustrated in Figure 1.2.
Within ordering semantics, the meanings of modal operators are defined in
terms of structures like the one illustrated in Figure 1.2. In simple cases (like the
one in the figure), the modal base and ordering source combine to define a set
of accessible worlds which can be thought of in pretty much the same way as
the accessibility relation in modal logic. The accessible worlds are the best-ranked
relevant worlds; this set is indicated by † in Figure 1.2 (i.e. the grey-shaded worlds
in the upper sector). The meanings of strong and weak modal operators can then
be defined in terms of this set in the usual way. However, the ordering structures
can be more complicated; for example, there may be multiple, incompatible sets
of “best-ranked” worlds (worlds compared to which there are no higher-ranked
worlds), or infinite series of ever-better worlds, with no stopping point at a highest-
ranked world. In those cases, more sophisticated definitions will be called for; see
Kratzer’s papers or Portner (2009) for details.
Kratzer’s framework has been used to provide a better analysis of modal strength
than the one which comes from modal logic. Recall the problem of must and ought
illustrated in (26). In order to explain the fact that must is stronger, we want to
say that the set of accessible worlds for must is always a superset of the set of
accessible worlds for ought. Within Kratzer’s system, the set of accessible worlds for
a modal is determined by the conversational backgrounds which function as modal
base and ordering source. Suppose that it is part of the lexical meaning of deontic
must that it uses a particular conversational background which includes only the
most important priorities and rules in the context; in contrast, deontic should uses
both this conversational background and another one which includes some less-
important priorities and rules. From this intuitively reasonable difference between
must and should, we can derive that the sentence with must is stronger than the
corresponding sentence with should.
Let me show this by way of example. Consider the following conversational
backgrounds:

i i

i i