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 is.subj happy
‘Pierre wants Marie to be happy.’
(5) J’ai rêvé qu’il était président.
I dreamed that he was.indic 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)
must.pres-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
 i i
OUP CORRECTED PROOF – FINAL, 10/10/2017, SPi

i i

20 introduction

(31) a. For any situation s, cbcrucial (s) = the set of the most important priorities
and rules in s (for example: you wash hands, you cook dinner).
b. For any situation s, cbgood (s) = the set of relevant but lesser priorities in
s (for example: you clean up the house before dinner).

Now consider (32):

(32) a. You must cook dinner.

b. You must clean up the house beforehand.
c. You should clean up the house beforehand.

The ordering source for (32a–b) is cbcrucial , while that for (32c) is a combination
of cbgood and cbcrucial . Given this, (32a) and (32c) will be true, but (32b) will
be false because there are best-ranked worlds accessible according to cbcrucial in
which you do not clean up the house. This example illustrates how the notion of
conversational background allows us to state in an explicit and intuitive way how
various contextual factors combine to determine which worlds are accessible in
modal semantics.
Ordering semantics is useful for dealing with a number of other problems in the
semantics of modality, including further variation in modal strength (as in (18)),
comparative modality, and the interaction between modals and conditionals, but it
would take us too far from our main concerns to pursue them here. Our discussion
so far has been enough to highlight two points at which the ordering semantics
approach is helpful in the theory of mood. First, Figure 1.2 is an example of the kind
of partially ordered set of worlds (or posw) which will show up in analyses of both
verbal mood and sentence mood, and which may serve as the basis for developing
a general theory of core mood. And second, the discussion of must and ought
gives a sense of how we can formalize intuitively meaningful factors (like “the most
important priorities and rules”) to give an account of the lexical semantics of modal
expressions. In order to understand verbal mood, we will develop techniques for
explaining a variety of features of lexical meaning within a formal theory of modal
semantics.
Before moving on, it is important to note that ordering semantics as sketched
above is not the last word on the semantics of modality. At the current time, we
see both a great deal of work aiming to modify and extend the basic framework
and some important arguments that the entire framework should be replaced.
When it comes to the semantics of mood, however, ordering semantics remains
the touchstone (though not necessarily the precise form of ordering semantics
proposed by Kratzer). Given the tight link between verbal mood and modality
in human language, it is certain that the correct theory of modality will play an
important role in the correct theory of mood, and so it should be seen as a test of

 There are several different possibilities for how to combine them. We might combine the ordering

sources as λs[cbcrucial (s) ∪ cbgood (s)] or merge them in another way. Or we might assign ought both
cbcrucial and cbgood ordering sources, adjusting the definition of ≤s to allow for multiple ordering
sources. See von Fintel and Iatridou (), Rubinstein (), and Katz et al. () for discussion.

i i

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

i i

the flow of information in discourse 21

old and new theories of modal semantics whether they can produce useful insights
when used for thinking about mood.

1.4 The flow of information in discourse

Linguistic research on sentence mood builds on a very diverse range of ideas
about discourse meaning from pragmatics, philosophy, and logic, and there is
no foundational framework with the same near-standard status that the possible
worlds semantics for modality has in research on verbal mood. In order to
investigate sentence mood, we need some ideas about the nature of “information
flow” in discourse. That is, we need some understanding of how information is
introduced, solicited, and deployed for planning, reasoning, and social interaction
in conversations or other genres like narratives and records. (Exactly what falls
under this term “flow” must be left vague at this point. Theories differ, and it is up
to a specific theory of discourse meaning to state what it encompasses and what it
does not, and to justify its choices.) In this section, I will give a brief overview of two
approaches to information flow which are important in contemporary research on
sentence mood: the dynamic approach and classical speech act theory.

1.4.1 The dynamic approach

The central virtue of the dynamic theories is that, through all their variations,
they give an explanation of the nature of information in discourse which is both
precise and amenable to a simple account of how that information changes. I will
introduce the ideas of dynamic semantics from a slightly unorthodox direction,
beginning with the work of Robert Stalnaker (1970, 1974, 1978, 1984), and then
discuss how Stalnaker’s ideas develop into dynamic semantics in the form of the
File Change Semantics of Irene Heim (1982, 1983, 1988, 1992). I will then give
overviews of two other important, and somewhat different, theories of dynamic
semantics: Kamp’s Discourse Representation Theory (Kamp 1981; Kamp and Reyle
1993) and the “Amsterdam school” of Dynamic Logic (Groenendijk and Stokhof
1990, 1991; Groenendijk et al. 1996). In this section, we will not go into more recent
work which explicitly applies the dynamic approach to understanding verbal mood
or sentence mood; we will delve into such material in Chapters 2 and 3.

Stalnaker on pragmatic presupposition and assertion. Stalnaker (1978) pro-

poses that “the central concept needed to characterize speech contexts” is speaker
presupposition:

A proposition is presupposed if the speaker is disposed to act as if he assumes or believes

that the proposition is true, and as if he assumes or believes that his audience assumes or
believes that it is true as well. (Stalnaker 1978, p.321)

Building on this concept of presupposition and the associated model of the

discourse context, we will be able to characterize the dynamic perspective on
information flow in discourse.

i i

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

i i

22 introduction

Stalnaker takes propositions to be sets of possible worlds (as we have done

above), and so it is possible to conjoin all of the speaker’s presuppositions into a
single proposition which represents all the ways the world could be which match
all of those presuppositions. This proposition is the context set of the speaker.
More generally, any participant in a linguistic exchange has a context set (relative
to that exchange):

(33) The
 context set of A in speaking to B =
{p : A presupposes p in speaking to audience B}

In a given conversation, each participant has her own context set. These context
sets might be different, since one or more participants could be mistaken about
what her audience assumes or believes. If the participants in the conversation
have the same presuppositions in speaking to one another, we have a non-
defective context, and theories of dynamic semantics typically make the ideal-
izing assumption that our understanding of language and communication can
make progress by considering only non-defective contexts. While one might fear
that this idealization takes us too far from the nature of real conversation, in
that most contexts are probably defective, it is perhaps reasonable to assume that
most contexts are “close enough to non-defective” (as Stalnaker says) that their
deficiencies do not matter.
In a non-defective context, we can speak of the common ground of the
context and the context set of the context:

(34) a. The common ground of a non-defective context c:

cgc = the set of presuppositions of all of the participants in c in speaking
to one another. 
b. The context set of a non-defective context c: csc = cgc

The context set of a non-defective context is identical to the context set of any
participant in the conversation (in speaking to any other participant).
In thinking about Stalnaker’s notion of context, it is worth noticing that the
discussion has moved quickly from a definition of presupposition based on a
speaker’s disposition to act as if he assumes or believes some things (in speaking
to an audience) to an analysis of context which relies on the concept of “the
conversation.” Specifically, the definitions of defective and non-defective contexts
only make sense relative to a delineation of a set of participants all of whom count as
speakers towards and audiences of one another. Thus, Stalnaker’s approach seems
to make the assumption that the nexus of speakers and auditors which constitutes a
conversation is clear and unproblematic. Although Stalnaker’s approach to defining
the common ground of a context is sufficient for the purposes for which he and
other dynamic semanticists use it, a more precise definition of context will require

 He or she might even have different context sets for each other addressee in the conversation, for

example if he or she is talking to two people, trying to deceive one with the support of another.

i i

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

i i

the flow of information in discourse 23

a better-developed notion of conversation and with it a clearer understanding of

what it means to be a speaker’s “audience” (cf. Bell 1984).
Given the ability to model the conversational context with the common ground
and context set, Stalnaker discusses the concepts of assertion and linguistic pre-
supposition. He writes that

. . . the essential effect of an assertion is to change presuppositions of the participants in the

conversation by adding the content of what is asserted to what is presupposed. This effect is
avoided only if the assertion is rejected. (Stalnaker 1974, p.323)

These ideas about assertion can be expressed in terms of a rule affecting the
common ground or context set:

(35) a. The essential effect of an assertion of S with respect to a non-defective

context c with common ground cgc is to change cgc to cgc ∪ {[[ S ]] c }.
• As a formula: cgc + S = cgc ∪ {[[ S ]] c }
This effect comes about unless the assertion is rejected.
b. The essential effect of an assertion of S with respect to a non-defective
context c with context set csc is to change csc to csc ∩ [[ S ]] c .
• As a formula: csc + S = csc ∩ [[ S ]] c
This effect comes about unless the assertion is rejected.

The formulaic versions above rely on interpreting the ‘+’ symbol appropriately, as
shorthand for the extended description given by Stalnaker.

Heim’s File Change Semantics. Irene Heim (1982, 1983, 1988) takes us from
Stalnaker’s concepts of assertion and presupposition to the core ideas of dynamic
semantics. In Heim’s work, we can see a number of important developments
which lead from the philosophical and somewhat abstract perspective of Stalnaker
to a framework more suited to research into linguistically interesting questions
about assertion and presupposition. The most obvious differences result from
Heim’s concern for a much wider range of presupposition phenomena. First of
all, whereas Stalnaker expressed skepticism that it is ever crucial to talk about the
presuppositions of linguistic forms (as opposed to the pragmatic presuppositions
of the speakers who use those forms), Heim develops a notion of the felicity
conditions of a presupposition trigger within dynamic semantics. For example,
Heim (1992) defines the felicity condition of too in such a way that the second
sentence in (36a) (uttered in that context and with the coindexing indicated) is
subject to the condition (36b):

(36) a. Mary1 is happy. SusanF is happy too1 .

b. Felicity condition of too, given context c indicated by (36a): csc + SusanF
is happy too1 is defined iff Mary is happy in every world in csc .

Intuitively, the second sentence of (36a) presupposes that the referent associated
with index 1, Mary, is both distinct from Susan and happy. Heim does not formulate

i i

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

i i

24 introduction

this presupposition directly in terms of the speaker’s disposition to act as if he

assumes or believes something, but rather expresses it as a condition on when
csc + S is defined. This way of formalizing the felicity condition triggered by too
ultimately leads back to Stalnaker’s pragmatic speaker’s presupposition, since on
the assumption that csc is non-defective, if the condition in (36b) fails, this is
because the speaker and addressee do not presuppose (in the sense of pragmatic
presupposition) that Mary is happy. However, crucially it is not identical to
Stalnaker’s notion of presupposition, because it is seen as a linguistic property
of the sentence which constrains speakers’ utterances based on their (speaker)
presuppositions.
In the context of (36a), the felicity condition of too will be satisfied, if the
assertion of the first sentence is not challenged, because the first sentence entails
that Mary is happy and the assertion adds this information to the context set. Heim
uses similar ideas about presupposition to attack the famous Projection Problem
for presupposition (see for example Karttunen 1973; Karttunen and Peters 1975;
Gazdar 1979; Soames 1982; Heim 1988, 1992).
Other crucial differences between Heim’s work and Stalnaker’s flow from Heim’s
concern for presupposition phenomena related to anaphora and definiteness.
Example (37b) is felicitous if preceded by (37a), intuitively because a soldier in (37a)
provides a referent for he in (37b):

(37) a. I saw a soldier in the park.

b. He waved to me.

To be more precise, within Heim’s theory the felicity of a pronoun depends on

both the preceding context and indexing of noun phrases in Logical Form. On
the reading indicated by the indexing (38a), the second sentence is felicitous.
On the reading indicated by the indexing in (38b), in contrast, it is not felicitous
unless the referent for he is determined by previous material or extralinguistic
context.

(38) a. I saw [a soldier]1 in the park. He1 waved to me.

b. I saw [a soldier]1 in the park. He2 waved to me.

Definite and indefinite noun phrases show complementary patterns: A definite is

felicitous if it is coindexed with an antecedent and it is entailed that the referent
has the property denoted by the definite’s common noun phrase. (It can also be
felicitous due to extralinguistic context.) An indefinite is felicitous unless it is
coindexed with an antecedent.

(39) a. I saw [a soldier]1 in the park. The soldier1 waved to me.

b. I saw [a soldier]1 in the park. ∗ A soldier1 waved to me.

 The description here leaves out many complexities which are well-studied in the literature on

anaphora and definiteness, but this oversimplified version is sufficient to allow us to understand Heim’s
contributions to dynamic semantics.

i i

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

i i

the flow of information in discourse 25

In (37)–(39) we see, in each case, that the initial sentence I saw a soldier in the park
introduces a soldier into the context, and that soldier can serve as a referent for
a pronoun or definite; it cannot, however, serve as the referent for a subsequent
indefinite, and indeed each indefinite must introduce its own, new individual into
the context. The main effort of Heim’s work on definiteness is to take this intuitive
description and develop it into an explanatory formal theory.
Under the inspiration of Stalnaker’s ideas about presupposition and assertion,
one can see (37)–(39) as showing that the common ground and context set contain
information not just about what is presupposed to be the case, but also about which
individuals the participants presuppose they are able to talk about. For example,
assertion of the first sentence in (38a) results in a context set which represents a
speaker presupposition that the speaker saw a soldier in the park and a speaker
presupposition that subsequent assertions can affect the speaker presuppositions
about this same presupposed soldier. Thus, after the second sentence in (38a) is
asserted, the context set encodes a speaker presupposition that the speaker saw a
soldier in the park who waved at her. The beauty of analyzing anaphora in terms of
the context set is that this overall presupposition of (38a) can be recorded without
there needing to be any actual soldier in the park—in the end, the context set
encodes a speaker presupposition to the effect that the speaker is disposed to act
as if she assumes or believes that she saw a soldier in the park who waved at
her. The soldier being talked about is a mere discourse referent, in Karttunen’s
(1976) terminology. We can understand discourse referents to be the referential
component of speaker presupposition.
The technical side of Heim’s work focuses on giving a precise, compositional
version of the above ideas. She replaces Stalnaker’s notion of the context set of the
context csc , a set of possible worlds, with the satisfaction set of the context satc ,
a set of pairs w, g consisting of a world w and assignment of values to indices g.
Note that g is a partial function from indices to values (not every index need be
in its domain), but all of the assignments in a satisfaction set must have the same
domain. The satisfaction set allows her to represent contextual information both
about what is the case (via w) and about discourse referents (via g). Specifically, if
index i is assigned a value by the assignments in the satisfaction set, i corresponds
to a discourse referent in the context. For example, satc in (40) represents a context
in which Max is a soldier, Al is a baker, we’re not sure whether Max walked, the
first discourse referent is one of the two men Max or Al, and there are no other
discourse referents.

(40) satc = { w1 , g1 , w1 , g2 w2 , g1 , w2 , g2 }, where

i. w1 , g1 :
w1 : Max is a soldier, Al is a baker, Max walked.
g1 (1) = Max. For all i = 1, g1 (i) is not defined.
ii. w1 , g2 :
w1 : (as above)
g2 (1) = Al. For all i = 1, g2 (i) is not defined.

 Instead of talking about the context, Heim talks about the satisfaction set of the “file,” a metaphor-

ical picture of context as a file cabinet filled with shared information.

i i

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

i i

26 introduction

iii. w2 , g1 :
w2 : Max is a soldier, Al is a baker, Max didn’t walk.
g1 : (as above)
iv. w2 , g2 :
w2 : (as above)
g2 : (as above)

Heim has to adjust Stalnaker’s analyses of pragmatic presupposition and assertion

to make sure that pronouns and definite and indefinite noun phrases affect dis-
course referents in the way observed in (37)–(39). Let’s work with a simpler example
than those, (41a):

(41) a. [A soldier]1 walked. He1 waved.

b. i. soldier(x1 ) ∧ walked(x1 )
ii. waved(x1 )

Building on ideas of Lewis (1975), Heim proposes that pronouns and definite and
indefinite noun phrases are realized as variables at logical form, and that, while
these variables become bound in some cases, in simple ones like this they do
not. The logical form of the sequence (41a) is a pair of open sentences (41bi–ii).
Although an open sentence does not have proper truth conditions, we can talk
about the conditions under which an assignment of individuals to the free variables
leads to its being true in a world. For example, a pair consisting of a possible
world w and assignment of individuals to indices g satisfies (41bi) if and only if the
individual g(1) assigned to index 1 is a soldier who walked in w. The satisfaction
set of a sentence S, Sat(S), is the set of world-assignment pairs which satisfy S.
Satisfaction sets allow a treatment of assertion parallel to Stalnaker’s:

(42) The essential effect of an assertion of S with respect to a non-defective

context c with satisfaction set satc is to change satc to satc ∩ Sat(S).
• As a formula: satc + S = satc ∩ Sat(S)
This effect comes about unless the assertion is rejected.

The assertion of (41bi) in a context c results in a satisfaction set in which every

w, g pair has g(1) a soldier who walked in w. For example, if it is asserted in the
context with satc as in (40), the result is { w1 , g1 }.
The satisfaction set also allows an attractive theory of the presuppositions of
definites and indefinites. In the first sentence of (41a), the index 1 is associated with
an indefinite. This means that index 1 is “new”; in other words, it is presupposed
not to be in the domain of the satisfaction set. When we get to the second sentence,
however, 1 is associated with a definite; at this point, 1 is presupposed to be in the
domain of the satisfaction set:

(43) Presuppositions of indefinites and definites:

a. If variable xi is associated with an indefinite NPi in S, then satc + S is only
defined if i is not in the domain of satc .

i i

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

i i

the flow of information in discourse 27

b. If variable xi is associated with a definite NPi in S, then satc + S is only

defined if i is in the domain of satc .

What’s important here for us is that (43) encodes lexically triggered presuppositions
which pertain to discourse referents, not propositional information.
Although it was not explicit in her earliest work, Heim comes to see the formula
in (42) not just as a description of what happens when someone asserts a sentence,
but rather as giving the very meaning of the sentence.

the meaning of a sentence is its context change potential (CCP). . . . A CCP is a function
from contexts to contexts. (Heim 1992, p.185)

In this quote, Heim goes beyond Stalnaker’s perspective on assertion, according to

which a sentence has semantic content (a proposition) which gets used to change
the context when it is asserted. Instead she adopts a more radical view according
to which the context change potential is itself the semantic content. This view is
the central hypothesis of dynamic semantics, and so we can see Heim’s work as
realizing a dynamic semantics analysis of speaker presupposition, assertion, and
felicity conditions.

Versions of the dynamic approach. For the purposes of discussing theories of

mood in Chapters 2 and 3, it will be useful to have terms for a range of broadly
“dynamic” perspectives on meaning. We will use the term dynamic pragmatics
for the Stalnakerian view that sentences have a static (i.e. non-dynamic) semantic
content such as a proposition, which is then used to define a context change
potential (for example, assertion). We will use dynamic semantics for the view
which says that the content of a sentence is precisely its context change potential.
The difference between dynamic pragmatics and dynamic semantics concerns the
relative priority between static content and context change potential. For dynamic
pragmatics, static content is basic and is dynamicized through operations native to
pragmatics; for dynamic semantics, context change is basic, and if static content is
needed for some analytical purpose, it has to be recovered from the context change
potential. We should note that hybrid views are possible: perhaps some sentences
have a static content while others do not, or perhaps sentences have two aspects of
meaning, one of which conforms to the dynamic pragmatics view and the other
of which requires dynamic semantics. As I will use it, the broad term dynamic

 To give the complete meaning of the sentence in this framework would be to define c+S generally,

not just the effect of S on the satisfaction set of c, satc + S.

 One might wonder whether the difference is more than terminological, since the Sat(S) plays a

role in () parallel to the content of S in Stalnaker’s (b), but be that as it may, Heim is clearly endorsing
a move to understanding sentence meaning in essentially context-change terms.
 Lauer () uses “dynamic pragmatics” for a rather different approach to discourse meaning.

Schlenker () employs the phrase to describe Stalnaker’s theory of presupposition and assertion,
and Stalnaker (to appear) uses the term in the way I do here. Stalnaker () further explores and
advocates for the dynamic pragmatics approach.
 The discussion of the performativity of modals by Portner () suggests this last possibility.

i i

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

i i

28 introduction

Table . Versions of the dynamic approach

The dynamic approach

Dynamic pragmatics Dynamic semantics Combined approaches

Static semantic content, and Semantic content Semantic content

CCP via pragmatics = CCP is varied

approach encompasses dynamic pragmatics, dynamic semantics, and all of these

other possibilities. These versions of the dynamic approach are summarized in
Table 1.2.

Discourse Representation Theory. Beginning at the same time as Heim’s work

on File Change Semantics but continuing for a longer time, Hans Kamp and col-
leagues developed an important representative of the dynamic approach, Discourse
Representation Theory (DRT; Kamp 1981; Kamp and Reyle 1993, among others).
DRT differs in an important way from File Change Semantics and the Stalnakerian
ideas on which the latter is based: According to DRT, the dynamic aspects of
meaning are computed with the essential mediation of a new level of representation
known as the discourse representation structure. A discourse representation
structure, or DRS, is a level of Logical Form which can incorporate the contribu-
tions of a sequence of sentences in discourse. For example, the operation of DRS-
construction applied to the sequence (41a) from above results in, first, (44b), and
then (44c):

(44) a. Initial DRS:

x
b. + A soldier walked. ⇒ soldier(x)
walked(x)

xy
soldier(x)
c. + He waved ⇒ walked(x)
waved(y)
x=y

The box structures in (44) are the conventional way of representing DRSs; officially,
they are just set-theoretic pairs consisting of a set of variables (discourse referents,
on top) and formulas in a logical language (conditions, at bottom). A DRS is

i i

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

i i

the flow of information in discourse 29

interpreted through techniques familiar from logic, and the simple examples in
(44) can be interpreted extensionally in a first-order model. Specifically, the DRS
in (44b) is true in a model M iff there is a soldier who walked in M, and the one in
(44c) is true in M if there is a soldier who walked and waved in M.
From the above sketch, it should be apparent that the semantic mechanisms
of DRT are not dynamic since each DRS is interpreted with a standard static
semantics. What is special about DRT is that the dynamism in the theory is realized
through the DRS-construction process. For example, the dynamic meaning of He
waved in (44c) is captured by the introduction into the DRS of discourse referent
y and conditions waved(y) and x = y. In this way it differs crucially from Heim’s
framework in which the context change potential of a sentence is a function from
contexts (represented in a simplified way as satisfaction sets) to contexts.
I am not aware of any research on the semantics and pragmatics of mood in
which the representationalism of DRT would differentiate it in a crucial way from
other dynamic approaches. In the very basic example used here, note the similarity
between the logical forms used by File Change Semantics, (41), and the DRSs
constructed by DRT, (44). The differences are merely (i) that the DRSs contain
a layer enumerating the discourse referents (x, y) and (ii) that they introduce a
separate discourse referent y for the pronoun which is equated with the previously
introduced one, x. The first difference is due to the fact that, in order to assign
traditional static semantic values to DRSs, all discourse referents must be implicitly
bound at some point, and the discourse referent layer identifies the scope at which
this occurs. (If box (44b) is nested inside a larger box, the value of x cannot be
accessed from the containing box.) The second difference corresponds to the fact
that Heim assumes that anaphoric relations are already resolved in syntax (and
represented by coindexing, as in (41a)), while Kamp assumes that they are resolved
in a “pragmatic” phase of DRS-construction which follows the introduction of
syntactically overt material.

Dynamic logic. Beginning with an important series of papers by Groenendijk

and Stokhof, dynamic logic developed the dynamic approach to meaning through a
series of formal innovations (e.g. Groenendijk and Stokhof 1990, 1991; Groenendijk
et al. 1996). The earliest work in dynamic logic explored the best ways to formally
analyze anaphora in a dynamic framework; the approach opposes the representa-
tional analysis of DRT and in this way is similar to File Change Semantics. However,
in contrast to the way in which Heim’s commitment to dynamic semantics became
clear only gradually, the founders of dynamic logic were interested in dynamic
semantics from the start. Based on these initial results focused on anaphora,
dynamic logic went on to explore many other important topics, including modality
and sentence mood. In this section, we will briefly review some important ideas
about anaphora and epistemic modality developed within dynamic logic. For a
more detailed introduction, see Portner (2009). We will return to dynamic logic
in Chapter 3 as we study its contribution to theories of sentence mood.
Dynamic logic is, as the name indicates, a logic, meaning that it is a syntax and
semantics for a syntactically simple artificial language. It is important to note that
the fundamental ideas of dynamic logic have been employed in several different

i i

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

i i

30 introduction

ways; they have been used to give a semantics to the standard syntax for predicate
logic according to which quantifiers have non-standard scope (Groenendijk and
Stokhof 1991), to interpret a logical language which translates sentences of English
in a Montague Grammar-like fragment (Groenendijk and Stokhof 1990), and to
give an explicitly dynamic semantics to modal predicate logic (e.g. Groenendijk
et al. 1996). Because of the attention it pays to modal semantics, we will focus on
the last of these here.
We can begin thinking about dynamic logic by relating some of its ideas to
what we’ve already discussed about Heim’s work. In File Change Semantics, the
context change potential of a sentence is defined in terms of the satisfaction set
of the context, satc , and the satisfaction set of the sentence, Sat(S). In dynamic
logic, in place of the satisfaction set of the context, we talk about an information
state. An information state is essentially a model of the context which focuses
on the same type of information present in a satisfaction set. Similarly, in place
of the context change potential, we say that the semantic value of a sentence is its
update potential. The update potential of φ is notated as [φ]. Update potentials
are formally functions from information states to information states, and we write
s[φ] to indicate the result of using φ to update s. Note that the meaning of a sentence
is precisely its update potential, and so when dynamic logic is applied to analyze
the meaning of natural language expressions, the result is clearly an example of
dynamic semantics as we have defined it above.
We begin with the concept of an information state. In basic dynamic logic,
information states are sets of possibilities, where a possibility is a triple w, r, g
whose three parts function as follows:

(45) A possibility is a triple w, r, g , where:

1. Possible world: w is a possible world.
2. Referent system: r is a partial function from variables to pegs.
3. Assignment to pegs: g is a function from the pegs in the range of r to
individuals (in the domain D of the model).

Together, r and g serve to assign values to variables, and so serve the function
of the assignment function components of the elements of the File Change Se-
mantics satisfaction set, but instead of assigning values to the variables directly, a
possibility first associates variables with a computation-internal representation of
the discourse referent, a peg, and then assigns an individual as the referent of the
peg. The role of pegs is not too important for our purposes in this book, and we
can think of them as just keeping track of which variables correspond to discourse
referents that are currently accessible.
Given the definition of a possibility, an information state will look like this:

 I have reordered the components to make the presentation connect in a more simple way to that

which has come earlier in this section.

i i

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

i i

the flow of information in discourse 31

(46) s = { w1 , r, g1 , w2 , r, g1 w1 , r, g2 , w2 , r, g2 }, where
i. w1 , r, g1 :
w1 : Max is a soldier, Al is a baker, Max walked.
r(x1 ) = d1 . For all i = 1, r(i) is not defined.
g1 (d1 ) = Max. For all i = 1, g1 (di ) is not defined.
ii. w2 , r, g1 :
w2 : Max is a soldier, Al is a baker, Max didn’t walk.
r: (as above)
g1 : (as above)
iii. w1 , r, g2 :
w1 : (as above)
r: (as above)
g2 (d1 ) = Al. For all i = 1, g2 (di ) is not defined.
iv. w2 , r, g2 :
w2 : (as above)
r: (as above)
g2 : (as above)

This information state is the dynamic logic counterpart of the satisfaction set in
(40). It represents a context in which one discourse referent (i.e. one peg), d1 ,
is active and associated with the variable x1 . This discourse referent could refer
to Max (a soldier) or Al (a baker), and it’s not known whether Max walked
or not.
Whereas File Change Semantics treats indefinite noun phrases as open sen-
tences, one of the most apparent innovations of dynamic logic is that it produces
a dynamic meaning for the existential quantifier ∃. The job of ∃ is to introduce a
new discourse referent (or peg) associated with its variable, and then generate the
information state which corresponds to all of the ways that an individual can be
associated with that peg while meeting the conditions set out in the formula in its
scope. Formally, the rule looks like this:

(47) s[∃xφ] = ∪d∈D (s[x/d][φ])

The sentence A soldier walked would be represented as (48):

(48) ∃x1 [soldier(x1 ) ∧ walked(x1 )]

And the result of updating an information state s with (48) is an information

state (call it s ) in which every possibility w, r, g has r(x1 ) = d1 and g(d1 ) is a
soldier who walked in w. The dynamic feature of ∃’s meaning here is that it leaves
a trace in the output state s ; specifically, r records that there is a discourse referent
d1 active and linked to the variable x1 . So, if the formula waved(x1 ) with a free
occurrence of x1 is used after (48), r will associate its x1 with d1 . So, when we
compute s [waved(x1 )], the information state will gain the information that this
same discourse referent waved (by losing those possibilities where g(d1 ) did not
wave in w).

i i

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

i i

32 introduction

In addition to this analysis of indefinites and discourse anaphora, similar to the

earlier frameworks of File Change Semantics and DRT, dynamic logic develops
some innovative ideas about epistemic modality. As with the analysis of ∃, the
semantic rules for the modal operators  and ♦ have something in common with
the standard treatments in modal logic, but they relate to the discourse context in
a different, inherently dynamic way.
Epistemic modality is the judgment type which concerns someone’s knowledge,
and within standard modal semantics, this judgment type is analyzed using an epi-
stemic accessibility relation. The key to understanding the dynamic logic analysis
of  and ♦ is to note that the information state s plays a role very similar to the
set of accessible worlds in the standard analysis, but that, because s only provides
a set of worlds, and not an accessibility relation, it is impossible to give a modal
formula like ♦φ a standard propositional meaning; that is, it is impossible to say
that it denotes the set of worlds from which some φ world is accessible. Instead, its
meaning must be defined in terms of the set of the “knowledge worlds,” that is the
worlds in s, directly.
Intuitively, one can think of ♦φ as performing a check on whether any φ worlds
are accessible—in other worlds, as checking whether there are φ worlds in s. If there
are, s is left unchanged and the conversation can proceed as if ♦φ had never been
uttered; but if there are not, the conversation has reached a dead end. Formally
speaking:

(49) a. s[♦φ] = s, if s[φ] = ∅;

b. s[♦φ] = ∅, otherwise.

The “absurd” state ∅ can also be reached by accepting a contradiction (that is, by
updating s with a sentence which is inconsistent with the information in s), and
so the outcome in (49b) is a clear signal that something has gone wrong with the
discourse. As an example, suppose we have the information state (46). In that case,
updating for (50a) leaves the information state unaffected, while updating for (50b)
leads to the absurd state:

(50) a. Max might not have walked.

b. Max might not be a soldier.

Dynamic semantics and assertion. In Stalnaker’s model of presupposition and

assertion, sentences have a classical, propositional denotation and their effect on
the context set, if they are felicitous, is defined by an intersective assertion operation
(35b). Heim maintains the idea of a general intersective update, although her
writings show some ambivalence in whether it is a separate operation which applies
to the non-dynamic sentence meanings (satisfaction sets) or a property of dynamic
sentence meanings (context change potentials). But either way, within Heim’s
theory, the final CCP can always be broken down into an intersective operation
applied to a satisfaction set. Dynamic logic is different in that there are several
operators, including ∃ and ♦, which cannot be defined in terms of the intersection
of s with any proposition-like set defined within the theory. In the case of ♦, what

i i

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

i i

the flow of information in discourse 33

is crucial is that ♦φ cannot be assigned a meaning independent of the information

state s which it is used to update. That is, there is no single [[ ♦φ ]] such that
s[♦φ] = s ∩ [[ ♦φ ]] . In this respect, these operators are “essentially dynamic”
(see Portner 2009 for discussion).
We should not overstate how essential the dynamism of the inherently dynamic
operators is, however. It is possible to reproduce the effects of ♦φ on an information
state s in terms of an intersective assertion operator if the static meaning of ♦φ is
allowed to depend on s. More specifically, ♦ can be given a standard modal logic
semantics if it may use an accessibility relation which always makes accessible all
of the possibilities in s. In that case, ♦φ will always denote a superset of s or the
empty set, and by intersecting with s we will always get s or ∅, as on the dynamic
logic analysis.
To see how this semi-dynamic meaning for ♦ would work, suppose that among
the accessibility relations available to modal semantics, we have Rs :

(51) For any information state s, Rs = { i, i : i ∈ s and i ∈ s}

Notice that what this accessibility relation relates are dynamic logic possibilities,
not possible worlds.

(52) For any information state s:

a. [[ ♦φ ]] s = {i : for some i , i, i ∈ Rs and {i }[φ] = {i }}.
b. s[♦φ] = s ∩ [[ ♦φ ]] s

Seen this way, the dynamic ♦ is an epistemic possibility modal whose judgment
type is based on the “knowledge” in the common ground.

The expanding range of the dynamic approach. Early work within the dynamic
approach contributed important ideas concerning assertion, presupposition, dis-
course referents, and modality. Since that time, the approach has been improved
in various ways, both by incorporating findings from other areas of semantics
and by expanding the range of the dynamic approach itself. For the purpose
of understanding mood, two developments have been most crucial: first, an
attempt to integrate the mechanisms of dynamic semantics into subsentential,
compositional semantics; and second, a variety of proposals which extend our
notion of context change beyond assertion. Both of these developments will
be important in later chapters, but it may be useful to have a very brief pre-
view here:

1. Several scholars have used the assertive update operation ‘+’ as part of the
compositional mechanism by which certain subsentential modal expressions
combine with their arguments. In Chapter 2, we will see how this strategy
develops from several directions, including Heim’s (1992) analysis of presup-
position projection with attitude verbs and Farkas’s (2003) attempt to account
for verbal mood selection in a File Change Semantics framework. (Other
work on verbal mood hints at the connection to the dynamic approach, but

i i

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

i i

34 introduction

does not make it explicit; see, for example, Portner 1997; Giannakidou 1999;
and Quer 2001.)
2. In Chapter 3, we will see that the dynamic approach has been used to account
for other ways of updating the context beyond assertions. Building on a long
tradition of research on the semantics and pragmatics of interrogatives, schol-
ars have developed various strategies for incorporating questions into the
dynamic approach to meaning. Some of the most representative work along
these lines is Hamblin (1971), Gazdar (1981), Ginzburg (1995a,b), Roberts
(1996/2012, 2004), and Ciardelli et al. (2013). More recently, imperatives have
also been discussed within the dynamic approach, for example by Portner
(2004, 2007, 2016), Roberts (2004), Mastop (2005), Charlow (2011, 2013), and
Starr (2010, 2013). Portner (2004, 2007) and Starr (2010, 2013) try to explain
the universal properties of clause type systems from the perspective of the
dynamic approach.
A related theme has been the idea that our model of discourse meaning
should make a place for propositions which do not yet have the presupposed
status represented by inclusion in the common ground. Hamblin (1971) and
Gunlogson (2001) develop models of the context containing representations
of the individual commitments of each participant in the conversation. Using
such a framework, we can model the state of a conversation which is biased in
favor of a proposition and the status of a proposition which has been proposed
for acceptance into the common ground but not yet accepted. We see similar
ideas in Portner’s (2008) work on epistemic modals, Farkas and Bruce’s (2010)
on polar particles, and Murray’s (2014) on evidentials.

There is, of course, much additional important work on both of these topics.
Here I only mean to point out the fact that much of it grows out of the dynamic
perspective in a more or less direct way. We will study these and other theories of
verbal mood and sentence mood in Chapters 2 and 3.

1.4.2 Speech act theory

Most readers of this book are no doubt familiar with the basic tenets of our
second framework for understanding the flow of information in discourse, classical
speech act theory. Austin (1962) described the differences between constative
utterances, those for which the main point is whether they are true or false, and
performative utterances, those which perform some action in the world. While
some of the actions performed by utterances bring about dramatic changes in
the social world (“I pronounce you married”) or in linguistic convention (“We’re
naming her Bea”), others pertain only to someone’s temporary psychological state
(“Let me warn you not to trust her”). The relevance of speech act theory to sentence
mood is obvious even at this initial level of understanding; clearly imperatives and
interrogatives normally do things the main point of which is not whether they
are true or false. And while there are seemingly clear cases of declaratives used
constatively to make a factual statement, like (53), it is very difficult to explicate
an understanding of sentence types and speech act theory according to which

i i

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

i i

the flow of information in discourse 35

imperatives and interrogatives are performative, while declaratives are constat-

ive.

(53) The bird is out of the cage.

Austin (1962) writes:

Would it be correct to say that when we state something . . . we are doing something as well
as and distinct from just saying something?
. . . Surely to state is every bit as much to perform an illocutionary act as, say, to warn or
to pronounce. (Austin 1962, p.573)

Even in its most canonical constative use, (53) does something, and in the passage
Austin calls what it does stating.
Austin recognized that his ideas were most likely to prove fruitful if used to
understand better all of the speech acts which various sentences may perform,
including those performed by constatives. To this end, he distinguishes three levels
of speech acts:

1. The locutionary act lumps together all of the acts intrinsic to the language
system, including phonetics, phonology, morphology, syntax, and truth-
conditional semantics, which the speaker performs when he utters a sentence.
2. The illocutionary act is the central act of intentional communication
which the speaker performs in making the utterance.
3. The perlocutionary act, or acts, are those which the speaker performs
through, or as an effect of, the locutionary and illocutionary acts.

In the case of (53), the locutionary act includes subacts like moving the tongue to
an interdental position, using the English word ‘the’, referring to a particular bird,
and (I will assume) expressing the proposition which (53) literally denotes in the
context in which the utterance is made. The illocutionary act might be to state this
proposition, or it might be something else (or in addition), such as to warn the
addressee that it’s not okay to open the window. The perlocutionary act might be
to convince the addressee not to open the window.

Illocutionary force and the logical form of illocutionary acts. Illocutionary acts,
as described by speech act theorists beginning with Austin, serve the many diverse
functions of language, including communication, planning, and managing social
relationships. The acts which serve these functions are correspondingly diverse.
We have acts such as asking one’s interlocutor a question, giving a subordinate an
order, providing a stranger with needed information, making a motion in a formal
meeting, promising to help a friend with his problems, and threatening an enemy.

 Though it is not part of the original framework of speech act theory, the theory probably

should allow a single utterance to perform multiple illocutionary acts, as noted by Levinson ()
among others.

i i

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

i i

36 introduction

The type of act of which a particular illocutionary act is an example is known

as its illocutionary force. Thus, we can speak of an utterance as having the
force of asking, the force of ordering, etc. Every illocutionary act has a force, and
many scholars who take a functional approach to language have seen the concept
of illocutionary force as providing a framework for explaining some important
properties of language and dialogue.
Searle (1965, 1969) offers a reassessment of Austin’s locutionary act which makes
explicit where the crucial components of semantic interpretation fit into speech act
theory. In the place of the locutionary act, he gives us both utterance acts (uttering
words, phrases, and sentences) and propositional acts. Propositional acts are
the speech acts of semantics; among them are the acts of referring, predicating, and
expressing a proposition. In uttering a sentence like (53), one normally performs
propositional acts like referring to a particular bird, predicating of it the property
of being outside of the cage, and expressing the proposition that the bird (the
particular one referred to) is not in the cage.
While the propositional acts are not part of the illocutionary act, they play
a special role in the illocutionary act. Searle proposes that illocutionary acts in
general have the logical form F(c), where F is its illocutionary force and c is its
content. The propositional acts determine the content of the illocutionary act. In
most cases discussed by philosophers, the content is the proposition expressed as a
propositional act, or in some cases it can be the object referred to (for example,
Down with Caesar!; Searle 1969). For example, when (53) is used to make an
assertion, the content is the proposition that the bird is outside of the cage and
the illocutionary force is assertion. In the case of wh-questions, Searle (1989)
proposes that the content is a propositional function.
As for the force of an illocutionary act, in many cases it is obviously determined
by some grammatical or lexical feature. Such linguistic material is known as
the force indicator. For example, the declarative sentence form (in English,
indicative verbal mood plus non-interrogative syntax) is an indicator of assertive
force, while interrogative syntax would serve as an indicator of the force of asking.
According to the literal meaning hypothesis (Gazdar 1981), every sentence
has a force indicator, and any utterance of it which is an attempt to perform
an illocutionary act includes an attempt to perform the illocutionary act F(c)
where F is the force indicated by the force indicator and c is the proposition
expressed by the propositional act. As we will see, this literal meaning hypothesis
(alternatively called the literal force hypothesis, when focusing on the status
of F) is critiqued in later literature.
Searle (1965, 1969) also clarifies the status of the rules which govern the perform-
ance of illocutionary acts. By and large, speech acts are defined by linguistic and
social conventions, and those conventions place conditions on when an individual
can perform the act. Illocutionary acts, in particular, are associated with felicity
conditions pertaining to the propositional content, the speaker’s psychological
state, and the actual state of affairs. (We will not go into the details of the kinds of

 As Searle notes, the idea that the use of a sentence has a logical form divided into a force-like part

and a content-like part has deep roots, at least to Frege’s judgment stroke. See Pagin () for a recent
overview. I find Dudman (, ) to provide very helpful insights.

i i

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

i i

the flow of information in discourse 37

conditions and rules which apply in this version of the theory, because they will be
incorporated into more recent versions of the Searlean approach discussed below.)
The ideas just outlined comprise what we may call the classical version of speech
act theory. In this framework, each of the utterances listed in (54) has the logical
form indicated (see Searle 1969, ch. 2):

(54) Where p = the proposition that the bird is out of the cage:
a. The bird is out of the cage. (asserted)
(p)
b. I warn you that the bird is out of the cage. (warned)
W(p)
c. The bird is out of the cage. (warned)
W(p)
d. Is the bird out of the cage? (asked)
?(p)
e. What is out of the cage? (asked)
?(λx[x is out of the cage])
f. Let the bird be out of the cage! (ordered)
!(p)

In the case of (54a), the indicative verbal mood and lack of interrogative structure
indicates the sentence’s assertive force, . In (54b), the performative verb and its
two nominal arguments I warn you is the force indicator, while the complement
clause that the bird is out of the cage gives the content. (Borrowing terminology
from the old performative–constative distinction, examples with a performative
verb as the force indicator are often called explicit performatives, or sometimes
just “performatives.”) In the case of (54c), the force of warning is not indicated in
any obvious way (and it is not clear what the indicative verbal mood is doing),
but rather inferred from context, while the propositional content continues to be
identified with that denoted by the subject–predicate structure. Because its force
is not the same as the one for which it contains an indicator, (54c) is known as an
indirect speech act. (If the literal force hypothesis is correct, it fundamentally
has the force , and the logical form with W has some type of secondary status.)
Normally, in this classical version of speech act theory, content is understood to
be the meaning computed by standard compositional semantics. Force is officially
understood to be indicated in diverse and sometimes complex ways, but in point
of practice, the theory largely focuses on cases where force is indicated either by
sentence mood or by a performative verb and its associated arguments.
Much research in linguistics has built upon the classical version of speech act
theory. For example, we often find claims in syntactic theorizing that a particular
piece of structure (e.g. the head of a functional projection, a specified element in C,
or a cluster of properties) is the “force indicator.” Because this work in linguistics
is devoted to analyzing linguistically interesting properties of sentences, it normally

 See Chapter  for discussion. Representative references include Rivero (); Rivero and Terzi

(); Rizzi (); Han (); for different views, see Michaelis and Lambrecht (); Ginzburg
and Sag (); Zanuttini and Portner ().

i i

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

i i

38 introduction

focuses either on clause types like interrogatives or imperatives, or on syntactically

uniform subclasses of these clause types. This literature thereby skirts some of the
most difficult challenges for speech act theory concerning explicit performatives
and indirect speech acts.

Language games. Austin was not the only philosopher who saw how important it
is to recognize the diversity of functions of language. Wittgenstein (1953) famously
presented the metaphor of language use as a variety of games and pursued the
idea that the meaning of a sentence is constituted by the ways in which it can be
used in the language game (or games). This perspective has a clear resemblance
to the speech act model. Stenius (1967) clarifies and develops Wittgenstein’s
ideas in a way which has been influential in more recent work in semantics and
pragmatics. Working within a Wittgensteinian conception of language use, he
divides the sentence’s logical form into two parts, the modal element and the
sentence radical. The modal element is not a modal in the sense we have used
the term in this chapter, but rather a sentence mood, and it indicates the use to
which the descriptive content (expressed by the sentence radical) is put in the
language game. (Stenius used the term ‘mood’ to refer to the type of use a sentence
has, but this is confusing in the present context, since we are using ‘mood’ to
represent categories of grammatical form. The term ‘force’ is more appropriate.) An
interrogative sentence thus has the logical form Q(S), with Q the force-indicator
(his “modal”) and S the sentence radical; it has meaning ?(p), with ? the force
(“mood”) and p the content. This way of talking about the logical form of sentences
obviously has much in common with speech act theory.
The language game picture lends itself to a certain way of thinking about the
role sentences play in interaction. Declaratives, interrogatives, and imperatives play
different roles in a game. For example, a game which includes uttering declaratives,
imperatives, and yes/no interrogatives as moves might have the following rules
(Stenius 1967, pp. 268, 273):

(55) a. Produce a sentence in the indicative mood only if its sentence radical is
true.
b. React to a sentence in the imperative mood by making the sentence-
radical true.
c. React to a sentence in the interrogative mood by saying ‘yes’ or ‘no,’
according to whether its sentence radical is true or false.

Stenius’s presentation of the language game metaphor has influenced much work
on discourse meaning. Although the rules in (55) are not dynamic (they do not
update the context), they lend themselves to further development in that direction,
as we see in work by Lewis (1979b) and Krifka (2001), for example. We will return to
Stenius’s ideas within the context of a broader discussion of sentence mood in 3.1.3.

 Searle’s remarks on the influence of Wittgenstein on Austin are interesting (Searle ).

i i

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

i i

the flow of information in discourse 39

Challenges for classical speech act theory. Indirect speech acts and explicit
performatives are both cases where there appears to be a mismatch between the il-
locutionary force of a sentence and the force which is marked by overt grammatical
means. It is clear that there is no explicit marker of the force of warning in (54c); the
explicit marker (here, indicative mood and non-interrogative grammar) normally
marks assertion, and speech act theorists have generally assumed that it does mark
assertion here as well. That is, they normally assume that the literal force hypothesis
is correct for these cases, and therefore that the force of warning comes about
through additional pragmatic processes like Gricean implicature (Gordon and
Lakoff 1971, 1975; Searle 1975a; Bach and Harnish 1979; Asher and Lascarides 2001).
This perspective is critiqued by Gazdar in an influential article (Gazdar 1981); he
argues that not every utterance can be assigned a speech act with the force indicated
by its force-indicator, and that not every utterance can be assigned a speech act
with the content assigned to it by the semantics. For example, considering (56)
he argues that there are contexts in which its force cannot be that of questioning
and its content cannot be [[ can you pass the salt ]] c .

(56) Can you pass the salt?

Rather, its force in such contexts is requesting (and only requesting) and its content
is [[ you will pass the salt ]] c . Gazdar, in other words, seeks to overturn the literal
meaning hypothesis.
Explicit performatives also show a mismatch between the force indicated by
their grammatical form (assertion) and the force they are assumed to actually have
(in the case of (54b), warning). On this way of looking at things, it is natural to
think of explicit performatives as indirect in a sense, deriving their observed force
from a (direct) force of assertion by means of additional reasoning (Lewis 1972;
Bach and Harnish 1979; Ginet 1979).
We can identify three main responses to these issues.

1. We find scholars who conclude that classical speech act theory is not particu-
larly useful for purposes of understanding language and communication. The
problem of indirect speech acts is seen by them as an example of the broader
fact that the function or functions of an utterance are dependent in complex
ways on the context in which it is used and on the interaction in which
it plays a role; Levinson (1981) gives an early statement of this perspective
and it has generally led to a decline in interest in speech act theory within
sociolinguistics.
2. We have scholars who aim to develop a more sophisticated theory of illocu-
tionary forces, with the goal of allowing a more transparent relation between
grammatical form and illocutionary force. Among other goals, Searle and
Vanderveken aim to explain the declarative form of explicit performatives
through this approach (Searle 1975b, 1989; Searle and Vanderveken 1985;

 See also Sadock (); Searle (a); Morgan (); Bach and Harnish (), among many

others.

i i

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

i i

40 introduction

Vanderveken 1990, 1991), and Beyssade and Marandin (2006) attack the
problem of indirect speech acts. We will sketch the basics of Searle and
Vanderveken’s theory in the remainder of this section and will return to the
more general issues in Chapter 3.
3. We can identify a trend of moving away from some of the ideas of classical
speech act theory and towards the dynamic approach. The dynamic notion
of update potential can be combined with speech act theory in various
ways. Most conservatively, it might be incorporated into classical speech act
theory to provide a formal model of the essential communicative effect of
an illocutionary act; more radically, update potential might replace the entire
notion of illocutionary force. While this trend does not resolve the difference
of opinion between scholars who assign (56) a literal meaning in which it
has the force of asking and those who do not, it shifts the issue by letting it
play out within an idealized formal model of discourse. It may well be that
there is a level of analysis at which it is correct to say that uttering (56) asks
a question, even though language users normally know right away that the
speaker is making a request. We will outline the central ideas of this work in
Section 1.4.3 and will return to it in Chapter 3.

The compositional structure of illocutionary force. Beginning in the mid-

1970s, Searle and Vanderveken undertook to improve upon the classical notion of
illocutionary force (Searle 1975b, 1989; Searle and Vanderveken 1985; Vanderveken
1990, 1991). Their goal was to develop a more systematic taxonomy of illocutionary
acts and a system which explains both the range of possible illocutionary acts and
the logical properties of and relations among those acts. This effort is important
for our purposes here because it led to, in the first place, a new approach to
explicit performatives, and then eventually to a theory of illocutionary forces
according to which the final force of a sentence is to some extent compositionally
derived from the sentence’s grammatical properties. This compositional approach
to illocutionary force provides the most well-developed basis within speech act
theory for the analysis of sentence mood. In this section, we will focus on the
analysis of illocutionary force presented by Searle and Vanderveken (1985) and
the way Searle applied it to the problem of explicit performatives (Searle 1989).
In Chapter 3, we will attend to the treatment of sentence mood developed by
Vanderveken (1990, 1991, 2002).
According to the system of Searle and Vanderveken, an illocutionary force is
determined by six properties:

1. The illocutionary point of a speech act is the core conventional pragmatic

purpose of the act. When the speech act has a propositional content p, its
point can be described in terms of that content.
(a) The five illocutionary points are:
i. An act with the assertive point says that the world is a certain way;
specifically, it represents an actual state of affairs as being p.

 My wording here draws extensively on Searle and Vanderveken’s () and Vanderveken’s (,

ch.).

i i

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

i i

the flow of information in discourse 41

Examples of illocutionary forces with the assertive point: asserting,

reporting, claiming.
ii. An act with the directive point attempts to get the hearer to do
something; specifically, it attempts to get the addressee to make it the
case that p.
Examples of illocutionary forces with the directive point: ordering,
asking, suggesting, advising.
iii. An act with the commissive point attempts to commit the speaker
to doing something; specifically, it attempts to commit the speaker to
making it the case that p.
Examples of illocutionary forces with the commissive point: prom-
ising, vowing.
iv. An act with the declarative point performs a world-changing action;
specifically, the speech act in and of itself makes it the case that p.
Explicit performatives have the declarative point (resigning, nomin-
ating).
v. An act with the expressive point expresses the speaker’s attitude;
specifically it expresses a feeling or attitude towards p.
Examples of illocutionary forces with the expressive point: thanking,
apologizing.
(b) Each illocutionary point has a direction of fit:
i. The assertive point has the words-to-world direction of fit; the
speaker has responsibility for choosing p which matches an actual
state of affairs.
ii. The directive and commissive points have the world-to-words direc-
tion of fit; someone has responsibility for making actual state of affairs
match p.
(With the directive point, the addressee has responsibility, while with
the commissive point, the speaker has responsibility.)
iii. The declarative point has the double direction of fit (both words-to-
world and world-to-words); the speech act itself guarantees that p
matches an actual state of affairs.
iv. The expressive point has the null direction of fit; whether p matches
an actual state of affairs is not at issue.
2. The mode of achievement of an illocutionary act is the way in which the
illocutionary point of that act is achieved. For example, an act of taking
an oath achieves the speaker’s commitment to do something by calling
upon a higher authority to penalize the speaker if the commitment is not
fulfilled.
3. The degree of strength of an illocutionary act is the theory’s way of saying
that some speech acts are stronger than others. For example, an oath commits
the speaker in a more serious way than a promise does.
4. The propositional content conditions place restrictions on the proposi-
tional content p to which the force may be applied. For example, an oath must
concern the future.
5. The preparatory conditions are the presuppositions of the speech act. For
example, an oath presupposes that the speaker can make it the case that p and

i i

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

i i

42 introduction

that the higher authority which is called upon would endorse the speaker’s
making it the case that p.
6. The sincerity conditions identify the psychological states which the
speaker expresses in performing the speech act. For example, in taking an
oath, the speaker expresses the intention to make it the case that p.

The most primitive illocutionary forces may be defined on the basis of the five
illocutionary points. For example, the basic force of asserting has the assertive
point and only those other properties which follow from having the assertive point,
namely it has words-to-world direction of fit, its strength is medium (the default
strength), it presupposes that the speaker has reason to believe p (a preparatory
condition), and it expresses that the speaker believes p (a sincerity condition). It
has no special mode of achievement and no propositional content conditions.
Other illocutionary forces can be grouped with the basic one with which they
share the illocutionary point, and differ from it in changing the degree of strength,
adding a mode of achievement, or adding propositional content, preparatory, or
sincerity conditions. For example, the act of reminding is based on assertion and
differs from assertion only in having the preparatory condition that the speaker
believes or presupposes that the addressee previously believed p (Vanderveken
1990, p. 174). Searle also admits complex illocutionary acts, built out of simple ones
with logical connectives like (illocutionary) conjunction.
Given this framework, we will look next at how Searle addresses the problem of
explicit performatives. In our earlier discussion, we noted the problem posed by
examples like (54b), repeated here as (57).

(57) I warn you that the bird is out of the cage.

It is not clear what the force indicator of this sentence is. On the one hand, we
can think of it as I warn you that, and from that perspective the sentence has the
standard logical form F(p). However, the sequence I warn you that does not always
function as a force marker, and the sentence also contains the normal assertive
force indicator (indicative mood). For this reason, many scholars have argued that
its basic force is some form of assertion and that its force of warning is indirect
(Searle 1989 cites Lewis 1972; Bach 1975; Bach and Harnish 1979; and Ginet 1979).
The idea of this indirect approach is that the speaker asserts that, by uttering
the very sentence (57) itself, she warns the addressee that the bird is out of the
cage; in virtue of this she represents the world as being one in which she warns
the addressee (by uttering that sentence), in which she is committed to believing
that she has made such a warning, and in which she has good grounds for this
belief. All together, according to the proposal, because of all of the entailments,
presuppositions and inferences based on the speaker’s assertion, the assertive act
guarantees that the utterance also performs an act of warning.
Searle, however, is skeptical of this proposal. While an assertion of (57) would
(in the right context) commit the speaker to giving a warning, this is not enough
to guarantee that she actually does give a warning. In order for a speech act of
warning to occur, the speaker must actually intend to give a warning, and even

i i

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

i i

the flow of information in discourse 43

if she is committed to having this intention, she might not (for example, if she
intends to mislead the addressee by pretending to care). As a result, according to
Searle the assertive utterance of performative-type sentences can at best receive
the illocutionary force of the performative verb via inference, that is as an indirect
speech act. Yet he thinks the use of (57) to give a warning is not indirect; thus, he
thinks that a different account is called for.
Searle’s (1989) solution is to analyze explicit performatives not in terms of
the force of assertion (or any other force with assertive point) or in terms of the
force explicitly named (in the case of (57), not warning), but rather as having a
special force designed precisely to account for explicit performatives. This force
is declaration. The special property of acts of declaration is that they count as
fulfilling the key conditions on the kind of act named by the performative verb in
the sentence itself. For example, when she uses (57) in an act of declaration, the
speaker meets the key condition on performing an act of warning the addressee
that the bird is out of the cage—crucially, she manifests the intention to warn the
addressee that the bird is out of the cage. Assuming that all of the other conditions
on giving this warning are met (such as that the bird is indeed out of the cage,
that the speaker knows this, and that the speaker and addressee agree that it is
something to be careful of), the warning will in fact be given. One can think of the
force of declaration as a special meta-force which identifies a possible speech act
A (which would make true the sentence uttered) and then fulfills as many of the
conditions which must be met for A to be performed as can be fulfilled by a speech
act. As a result, the act of declaration amounts to a performance of A, provided the
remaining conditions on A are met.
Even given this interesting analysis, it is still not clear what the force marker in
an explicit performative like (57) is. In form, it still looks like its force should be
assertion (or another force with assertive point), so how is the force of declaration
assigned in this case? Vanderveken (1991, pp. 140–1) offers some guidance on this
issue. After proposing “the law of an assertive commitment in the achievement of
the declarative point,” which states that “every declaration contains an assertion
of its propositional content,” he states that the indicative marks the assertive point
unless some additional element like hereby further specifies the illocutionary act
in such a way that the whole construction is a marker of declarative point. The
intuition seems to be that the assertive point is somehow entailed by the declarative
point and that this relation is realized in the close relation between the ways in
which the two points are marked, but the exact mapping between grammatical
forms and types of speech acts remains somewhat vague.

1.4.3 Update potential and illocutionary force

Speech act theory was the main approach to information flow in discourse from
the mid-1960s through at least the late 1970s, and as a result of its dominance
during this formative phase of pragmatics, it dramatically affected the ways in

 See Condoravdi and Lauer () for a recent analysis of performatives in terms of assertion which

aims to address Searle’s critique.

i i

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

i i

44 introduction

which all linguists interested in discourse meaning think and talk about the topic.
As the dynamic approach began to emerge with the work of philosophers like
Hamblin (1971) and Stalnaker (1978), researchers in pragmatics faced the task of
understanding the relation between the two approaches. In this section, we will
consider a number of perspectives on their relation; as we will see, some proposals
can be seen as incorporating the tools of the dynamic approach into classical speech
act theory, while with others, the relation is much vaguer. As we discuss the range of
proposals, I will make an attempt to classify them in a way which will prove helpful
when we return to a detailed investigation of sentence mood in Chapter 3.

The dynamic force hypothesis. In an important but often-overlooked early

discussion, Gazdar presents a theory which combines speech act theory with the
central idea of the dynamic approach. Specifically, he proposes that illocution-
ary acts should be analyzed as update potentials as understood within dynamic
pragmatics. Given the traditional division of the logical form of illocutionary acts
into an illocutionary force and a propositional content, this perspective leads to a
position I refer to as the dynamic force hypothesis:

(58) The dynamic force hypothesis: Illocutionary force is to be analyzed in terms

of update potential as understood within the dynamic approach.

The dynamic force hypothesis aims to improve upon classical speech act theory by
providing a better explanation of the nature of illocutionary acts.
The key ideas of Gazdar’s model can be summarized as follows:

1. The content of a speech act. The semantic value of any sentence can be the
propositional content of a speech act. These contents have various semantic
types, for example:
(a) The meaning of a declarative sentence is a proposition.
(b) The meaning of an interrogative sentence is a set of propositions.
2. Illocutionary force. An illocutionary force is a function from contents to
update potentials.
3. Update potential. An update potential is a function from contexts to contexts.
4. Contexts. Though Gazdar doesn’t commit to understanding contexts this
way, he suggests Hamblin’s (1971) commitment slates as a good basis for
providing a definition of contexts. (See Section 3.3.1 for a detailed discussion
of Hamblin’s ideas.)
5. Speech act assignment. A speech act assignment is a pair f , c consisting of
a force f and a content c.
6. Speech acts. A speech act is f (c), for any speech act assignment f , c .

 Quite a few important works in dynamic semantics make little or no reference to speech acts (e.g.

Aloni and van Rooy ; Ciardelli et al. ); while Groenendijk et al. () mention speech act
theory as an important predecessor, they leave open whether they think it is being improved upon or
superseded by their work.

i i

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

i i

the flow of information in discourse 45

To give an example, (54a: The bird is out of the cage) might (on its literal, assertive
interpretation) receive a speech act assignment , p , where  could be defined as
follows:

(59) For any proposition p, (p) = the function g which assigns to any context
c (satisfying the felicity conditions for the assertion of p by the speaker) the
context c just like c except that the speaker is committed to p in c .

Gazdar uses this framework to express several important criticisms of classical

speech act theory. Specifically, he critiques the literal meaning hypothesis in two
ways. Recall that the literal meaning hypothesis says that every utterance performs
the speech act defined by the sentence’s force indicator and semantic content. In
Gazdar’s terms, this is to say that every utterance e of sentence d receives the speech
act assignment f , c , where f is indicated by the force indicator of d and c is the
denotation of d in the context. He argues that it is not always possible to make a
speech act assignment with content [[ d ]] , and that it is also not always possible
to make an assignment with force f . As noted above, he uses (56), among other
examples, to press these points, claiming that there is not good evidence that it can
(in the stereotypical type of context) have assignment ?, Q , where ? is the force
of asking and Q is the set of propositions {‘the addressee can pass the salt,’ ‘the
addressee cannot pass the salt’}.

Recent perspectives. It remains unclear in Gazdar’s discussion whether he sees

Points 1–6 above as representing the beginnings of a new approach to speech acts,
or as a method of bringing out some serious problems in speech act theory. But
whatever Gazdar’s intention may have been, the ideas themselves have turned
out to be prescient. Indeed, at around the same time, Levinson (1983) identified
the dynamic force hypothesis as a promising direction for developing a new and
improved theory of speech acts. In his introductory textbook (Levinson 1983,
sect. 5.6), Levinson presents the outlines of the dynamic force hypothesis in terms
very similar to Gazdar’s (citing, in addition to Gazdar, Hamblin 1971; Ballmer
1978; and Stalnaker 1978). Labeling it the “context-change theory of speech acts,”
Levinson assumes with Gazdar that a theory of speech acts must account for the
full range of illocutionary forces identified by classical speech act theory, but he
also makes clear that the explanatory value of such a theory depends on its utilizing
a limited range of basic types of update. In this way, Levinson identifies a crucial
tension which exists within Gazdar’s presentation: On the one hand, classical
speech act theory treats all illocutionary forces, both basic ones like assertion and
more obscure ones like vowing and exhortation, as full-fledged forces; on the other,
sentence mood systems in natural language are not so complex, and it is natural
to try to explain them in terms of a restrictive theory of context change based on
a simple model of context.

 It’s not clear how “the speaker” is represented in Gazdar’s speech act model.

i i

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

i i

46 introduction

More recently, the literature has given more priority to the second horn of this
dilemma. In other words, the rich and complex structure of illocutionary acts has
played a diminishing role in work on mood and force, while models which focus
on a smaller number of basic forces have gained influence. We see this tendency
in the work of such modern speech act scholars as Krifka (2001, 2014), Charlow
(2011), and Kaufmann (2012). As a result of this trend, the difference between
speech act theory (with the dynamic force hypothesis) and the dynamic approach
has grown smaller.
It also worth noting that many scholars do not invest much effort in explaining
the relation between their ideas and speech act theory. For example, Roberts
(1996/2012) states:

Note that moves, on the interpretation I will give them, are not speech acts, but rather the
semantic objects that are expressed in speech acts: A speech act is the act of proffering a
move. (Roberts 2012, p. 12)

It is not clear in Roberts’ discussion whether the speech acts she is referring to are
the illocutionary acts of classical speech act theory, or just a similar construct. We
see similar tangential references to speech act theory in a great deal of recent work
which makes use of ideas from the dynamic approach (for example, Lewis 1979b;
Ginzburg 1995a,b, 1996; Asher and Lascarides 1998, 2001; Portner 2004; Rett 2014;
and Murray 2014).
This discussion of the relation between speech act theory and the dynamic
approach has touched briefly on a number of quite different theories, and it is
important not to get lost in the details. The crucial point is that we can see
ideas from dynamic semantics and pragmatics playing important roles in a variety
of contemporary theories, some of which are naturally understood as versions
of speech act theory, and others of which are not. The term “dynamic force
hypothesis” is most useful when applied to theories which aim to contribute to an
improved version of classical “Searlean” speech act theory, and I will use it this way.
I consider the works of scholars like Krifka, Kaufmann, and Charlow to fall within
this category. Other scholars like Portner (2004) and Starr (2013) have clearly set
their ideas apart from speech act theory, and these I will treat as falling under the
dynamic approach. Still others, like Lewis (1979b) and Roberts (2012), indicate that
their ideas can be integrated with speech act theory into a broader conception of
information flow in discourse, but do not develop a detailed proposal for the role
which will be played by the core ideas of classical speech act theory. These authors’
works too will be discussed as falling within the dynamic approach. Of course,
in dividing theories into two groups, we simplify them and risk making certain
theories seem more similar or dissimilar than they really are. Readers should keep

 It is worthwhile to point out Condoravdi and Lauer’s (, ) work on performatives and

the illocutionary force of imperatives. While their main proposals are explicitly presented as falling
within speech act theory, they offer an alternative formulation of their ideas about imperatives which
is compatible with the dynamic approach (Condoravdi and Lauer , sect. ), and the later ideas of
Lauer () are not as tightly tied to classical speech act theory.

i i

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

i i

looking ahead 47

in mind that, while the classification used here is appropriate for the purposes
of this book—specifically, for the goal of determining what type of theoretical
framework is best suited to analyzing sentence mood—the same classification may
be unhelpful in other contexts.

1.5 Looking ahead

In Sections 1.3–1.4, I have presented some of the most essential background for
recent work on verbal mood and sentence mood. Specifically: (i) the possible
worlds semantics for modality which developed out of modal logic provides the
framework in which the most important analyses of subsentential modality have
been developed, and these analyses in turn provide the basis for theories of verbal
mood; and (ii) the major theories of information flow in discourse, namely the dy-
namic approach and speech act theory, serve as the basis for attempts to understand
sentence mood. Up until now, there has been a division in semantic research on
mood into these two fields, one focusing on verbal mood and the other on sentence
mood, with little interaction between them. This is so despite the fact that verbal
mood clearly plays a role in the determination of sentential force. The reason for the
division seems to be that the phenomena associated with verbal mood and sentence
mood overlap in an odd way. The central questions in the analysis of sentence
mood (how to analyze the declarative, interrogative, and imperative clause types
and associate them with conversational functions) are partially independent of
verbal mood, in that both declaratives and interrogatives typically have indicative
verbal mood in root clauses and can be indicative, subjunctive, or infinitival in
embedded clauses. In contrast, imperatives can have a distinct verbal mood, but
the connection between imperative mood and other verbal moods and mood-like
forms (indicative, subjunctive, and infinitive) is not well understood.
In the next two chapters, we will examine verbal mood (Chapter 2) and sentence
mood (Chapter 3). Only after we have gained a detailed understanding of these two
topics will we be able to return to thinking about the nature of mood in general.
In Chapter 4, we will investigate whether verbal mood and sentence mood are
fundamentally similar enough to be analyzed in a unified theory of core mood,
and we will consider their relation to other phenomena like reality status and
evidentiality which are usually seen as residing at the margins of the category
of mood.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

2
Verbal mood

Mood is, as we said in Chapter 1, “an aspect of linguistic form which indicates
how a proposition is used in the expression of a modal meaning.” In light of this
definition, verbal mood can be understood as the subcategory of mood with the
primary function of indicating how the proposition is used in the computation of
subsentential modal meaning. Such a conception of verbal mood is close to being
just a formal statement of the traditional practice of studying mood in particular
languages. Both in traditional grammars of languages with a distinction between
indicative and subjunctive clauses and in syntactic and semantic studies of those
languages within formal linguistics, one finds lists of elements and constructions
which require or allow subjunctive or indicative clauses. The subjunctive list will
include lexical items like ‘want’ and ‘happy (that)’ and the indicative list will include
ones like ‘know’ and ‘say.’ And, it is important to notice, not only do the elements
in the lists for the most part express modal meaning, but they are described,
classified, and explained in terms which suggest that their modal meaning is what
is crucial for understanding verbal mood. For example, we are told that predicates
which express the concepts of desire, will, necessity, and emotion select subjunctive
complements (in those languages which have the subjunctive). These types of
modality are all examples of the same judgment type, priority modality.
In order to understand verbal mood, we will need prior understanding of
subsentential modality. Section 2.1 will lay out this background. Then, in the
remainder of the chapter (Sections 2.2–2.3), we will examine the distribution of
verbal mood forms and the semantic theories which have attempted to explain
their distribution. As we will see, researchers in this area have made slow but
steady progress in improving the semantic theory of verbal mood, and we have
now reached the point where it seems possible to enumerate many of the key ideas
which must be part of an explanatory theory of this domain. We will see that the
following are important:

1. Ideas from ordering semantics seem to be important in explaining the distri-

bution of verbal mood forms.
2. Differences in the meanings of clauses marked by different verbal moods
seem to be based on a variety of semantic properties connected to indexicality,
including in particular varieties of de se interpretation.
3. Scholars have developed a number of detailed, insightful theories of indicative
and subjunctive complement clauses, but we do not have such thorough
analyses of verbal mood in other grammatical positions or the meanings and

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, 26/9/2017, SPi

i i

subsentential modality 49

functions of other forms which may fall into the category of verbal mood,
such as infinitives and dependent modals.

Apart from these points about the semantics of verbal mood in and of itself,
studying verbal mood yields insights into the nature of modality generally. Verbal
mood gives us evidence about how to classify modal expressions in a linguistically
relevant way, and highlights grammatically important properties of modals. As a
result, the analysis of verbal mood has implications for our understanding of both
subsentential and discourse modality.

2.1 Subsentential modality

As mentioned in Section 1.3.2, the standard theory of modality in possible worlds
semantics was developed for sentential modal constructions. As a result, we rarely
find discussions of the difference between the examples in (1). The fact that must is a
modal auxiliary (a sentential modal) while necessary is an adjective (a subsentential
modal) is glossed over by the fact that both can be symbolized in modal logic as
α (where α = ‘we pay this bill today’).

(1) (a) We must pay this bill today.

(b) It is necessary that we pay this bill today.

However, modal adjectives show differences from sentential modals in having

typical properties of adjectives. First, since adjectives are an open lexical class, there
are many more of them and as a result, they distinguish the parameters of judgment
type and strength in a more fine-grained way. Second, they are gradable expressions
and the fact that they can be combined with degree modifiers leads to even more
variation in strength. These points are illustrated in (2):

(2) (a) It is possible/likely/probable/necessary/desirable/believable that Ben

will win.
(b) It is somewhat/quite/100 likely that Sam will lose.

At a general level, the facts are similar with modal nouns:

(3) (a) There is a possibility/chance/likelihood that Ben will win.

He has a strong desire/belief that Ben will win.
(b) There is a tiny/slight/significant/excellent possibility that Noah will win.

 The role of the modifiers of modal auxiliaries in (i)–(ii) is not completely clear:
(i) Ben positively must be winning.
(ii) Ben to some extent must be winning.
Positively in (i) is an extreme modifier (Rett b; Morzycki ; Portner and Rubinstein ) while
to some extent in (ii) lessens the strength of must in some sense.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

50 verbal mood

As mentioned in Section 1.3.2, Kratzer applied her version of ordering semantics to

some of the variation in strength among modal adjectives and nouns. We saw how
she defined the concept of ‘slight possibility,’ which can be used to give an account of
the English there is a slight possibility that. She likewise provides definitions, using
ordering semantics, of necessity, weak necessity, possibility, good possibility, and
better possibility and associates these with adjectival and nominal constructions in
English and German; see Kratzer (1981, 1991) and Portner (2009) for her treatments
of these concepts. However, the fact that the concepts are not given a compositional
treatment—slight possibility is not derived from possibility through the effects of
the modifier slight—means that Kratzer’s version of ordering semantics cannot be
the full correct analysis as it stands.
Lexical verbs with modal meaning have received much more extensive study
within semantics than modal adjectives and nouns. While modal verbs which
correspond to auxiliaries in English, such as Spanish poder (‘can’), are typically
treated as sentential modals, the set of verbs which describe aspects of the mental
lives of individuals has been the focus of much research within semantics and
philosophy. Example (4) lists some of these words, known as propositional
attitude verbs because they can be thought of as denoting an attitude (mental
relationship) which a thinking individual, the attitude holder corresponding to the
subject argument Ben, holds towards a proposition, corresponding to the sentential
complement that John can fly:

(4) Ben believes/hopes/knows/doubts/fears/realizes/says that John can fly.

Note that modal adjectives also sometimes express propositional attitudes. For
example, in Ben is certain that birds can fly, the adjective certain describes a
relation between Ben and the proposition that birds can fly very similar to the
ones indicated by some of the verbs in (4). Therefore, whatever we have to say
about the semantics of the verbs in (4) should be extendable to a significant
extent to adjectives like certain. But other modal adjectives don’t have analogues
in the realm of modal verbs—there is no verb which expresses the same basic
meaning as probable (it’s related to prove, but probable does not mean ‘prov-
able’). The realm of subsentential modality is a collection of overlapping types,
some based in semantics and some based in syntax, and we can expect that
the properties we observe are due to a mixture of interacting semantic and
morphosyntactic factors.

Propositional attitude predicates as strong modals. Since the work of Hintikka

(1961), propositional attitude verbs have been treated as modal operators within the
framework of possible worlds semantics. Let us see how this works taking the verb
believe as a case study. In order to treat believe as a modal element within modal
logic (Section 1.3.2), we need to know two things: its accessibility relation and its
modal force. The accessibility relation of a propositional attitude verb is determined
by a combination of the lexical meaning of the verb itself and the grammatical
context in which it occurs. Specifically, the set of worlds relevant to the meaning
of believe can be identified from the beliefs of the referent of its subject (the
attitude holder) at the time specified by the sentence’s tense. For example, in (5),

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

subsentential modality 51

the accessible worlds are those which match Ben’s beliefs at the speech time in the
actual world:

(5) Ben believes that John flew away.

In terms of the tools of modal semantics outlined in Section 1.3.2, the meaning of
believe can be stated in terms of a doxastic accessibility relation.

(6) For any situation s: Rdox (s) = {w : everything that the thinking participant
of s believes in s at the time of s holds in w}.

Let us assume that the verb sits in a syntactic structure something like the one in
(7a). If so, the semantics needed for believe will be that in (7b):

(7) (a)
Present S

s NP VP

Ben VP CP

believe that John flew away

(b) Believe denotes the function which takes a proposition p, individual a, and
situation s as arguments, and yields true iff a is the thinking participant
in s and every world w ∈ Rdox (s) is one in which p is true.
(c) [[ believe ]] = [λpλaλs . a is the thinking participant of s and for every
world w ∈ Rdox (s), w ∈ p]

The meaning in (7b) can be given as (7c) using the standard λ-notation of formal
semantics.
Notice that (7b–c) treats believe as a strong modal; it involves universal quan-
tification over accessible worlds. Treating it as a weak (existential) modal would
give obviously wrong results. Suppose that Ben has no opinion about whether John
flew away; in that case, there will be some worlds compatible with what he believes
in which John flew away. On the assumption that believe has existential force, (5)
would incorrectly be predicted to be true. In contrast, defining the meaning of
believe as a strong modal gives the right result in this case.

 It is sometimes clearer to state the accessibility relation as an actual relation, and not a function

from situations to a set of accessible worlds. This would look very similar to ():
(i) For any situation s and world w: Rdox (s, w) iff everything that the thinking participant of s believes
in s at the time of s holds in w.
In a format closer to modal logic, the accessibility relation would be a relation between worlds:
(ii) For any individual a, situation s, and worlds w and w∗ :
Rdox(a,s) (w∗ , w) iff everything that a believes in the situation s in world w∗ is the case in w.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

52 verbal mood

Applications of strong modal semantics for propositional attitudes. Some

other propositional attitude verbs can be treated as strong modals simply by
changing the accessibility relation, but as we turn to the full diversity of these
predicates, various complexities arise. In some cases, the analysis may be correct
as far as it goes, but be incomplete. Consider the situation with know. Suppose it is
treated in parallel with believe above:

(8) [[ know ]] = [λpλaλs . a is the thinking participant of s and for every world
w ∈ Rep (s), w ∈ p]

This meaning fails to give an account for the verb’s factivity. We could try to
incorporate factivity either by saying more about the accessibility relation in (8),
requiring it to be reflexive, or by adding the presupposition that p is true as a
separate component of the lexical meaning.
Another type of issue can be illustrated with the propositional attitude verb
learn. If we define the accessibility relation for learn as in (9) below, we fail to model
the fact that simple sentences with learn are eventive rather than stative.

(9) For every situation s: Rlearn (s) is the set of worlds w such that everything that
the thinking participant of s learns in s is also the case in w.

The reason that learn describes an event, of course, is that to learn something is
to undergo a change from not knowing it to knowing it. This description indicates
that learn does not express a simple propositional attitude, but rather has a meaning
which can be defined in terms of a more basic propositional attitude, knowledge,
as something like the following:

(10) [[ learn ]] = [λpλaλs . a is the thinking participant of s and s has an initial

part s1 and a final part s2 such that:
i. it’s not the case that for every world w ∈ Rep (s1 ), w ∈ p; and
ii. for every world w ∈ Rep (s2 ), w ∈ p]

These discussions of know and learn indicate that the possible worlds semantics for
modality often provides only one component of the full semantic analysis of a given
predicate. Unlike sentential modals, which are often (perhaps wrongly) assumed to
be nearly pure exemplars of modal meaning, propositional attitude verbs and other
subsentential modals have lexical meanings of which the pure modal component
captured by the possible worlds analysis of Section 1.3.2 is only part of the overall
meaning.
Desire verbs like want and hope form one especially complex and important
subclass of subsentential modals. Suppose that we analyze want as a strong modal
in the same way as believe above using a buletic accessibility relation:

 See Portner (, ch.) for explanation concerning the use of properties of accessibility relations

in modal logic.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

subsentential modality 53

p: Nicolas takes free ride

on the Concorde
q: Nicolas rides the Concorde

w∗
Rbul(s)

Figure . Entailment pattern under strong modal analysis

(11) For any situation s: Rbul (s) is the set of worlds w such that everything the
thinking participant of s wants in s is the case in w.

This analysis has problems. Consider the following sentences based on an example
of Asher (1987):

(12) (a) Nicolas wants to ride the Concorde.

(b) Nicolas wants to take a free ride on the Concorde.

Riding the Concorde (a supersonic airplane now no longer in service) is very

expensive, and Nicolas does not think it is worth the money. Therefore (12a) is false.
However, it would be fun, so (12b) is true. On the assumption that the subject and
situation arguments of want are the same in the two examples, this combination
of attitudes is not possible according to the analysis which treats want as a modal
with universal force. Figure 2.1 illustrates how the semantics works: evaluated in a
situation s in the real world w∗ , the set of worlds accessible by Rbul is a subset of the
worlds p in which Nicolas takes a free ride on the Concorde. This is what we need
to make (12b) true in w∗ . Since p is a subset of q, the set of accessible worlds is also
a subset of q, and hence (12a) must be true as well. (In general, for any modal ‘’
with universal force, if p entails q, p will entail q.) Yet obviously, (12b) does not
in fact entail (12a), and so a different analysis is needed.
Stalnaker (1984) and Heim (1992) develop ideas about the semantics of desire
predicates which use the ingredients of modal semantics in an even more indirect
way than we saw with learn. The intuition behind this analysis is that desire
predicates indicate a preference for one hypothetical scenario over another; for
example, to evaluate (12a), Nicolas must consider both what it would be like if he
rode the Concorde, and what it would be like if he did not. Then, the sentence is

 There are similar logical problems with the modal analysis of believe in () as well, including the

famous “problem of logical omniscience.” The question remains open whether they show that the entire
approach should be modified in a radical way or even abandoned, or whether the unintuitive predictions
of the theory outlined in the text can be defended. An introduction to the foundational issues can
be found in many handbook and encyclopedia articles on propositional attitudes and possible worlds
semantics, for example Swanson (b).

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

54 verbal mood

true if he prefers the first kind of scenario over the second. This idea is made formal
by Heim (1992, (31)) as follows (with adjustments to fit with our notation):

(13) [[ want ]] = [λpλaλs . a is the thinking participant of s and for every w ∈

Rdox (s), every world in p maximally similar to w is more desirable to a in s
than every world maximally similar to w in ¬p]

It will be easier to see how Heim’s analysis works by introducing several definitions;
the following are based on her presentation:

(14) (a) For any worlds w1 and w2 ,w1<s w2 iff w1 is more desirable to the thinking
participant of s in s than w2 .
(b) For any sets of worlds p1 and p2 , p1<s p2 iff for every w1 ∈ p1 and w2 ∈ p2 ,
w1 <s w2 .
(c) For any world w and proposition p, Simw (p) = {w : w ∈ p and w
resembles w no less than any other world in p}.
(d) [[ want ]] = [λpλaλs . a is the thinking participant in s and for every w ∈
Rdox (s): Simw (p) <s Simw (¬p)]

The set of worlds Simw (p) represents the scenario that Nicolas rides the Concorde
from the point of view of world w; it is the set of worlds most similar to w in which
he does so. (Heim’s use of similarity to represent what happens if p is true is based
on Stalnaker’s 1975 analysis of conditionals.) Likewise, Simw (¬p) represents the
scenario that he does not ride the Concorde. The ordering <s indicates Nicolas’s
preferences. It is defined first as a relation between worlds in (14b), but then
applied to propositions in (14c). The definition in (14d) uses an ordinary (doxastic)
accessibility relation Rdox , putting it together with the preceding definitions to
say that, relative to every doxastically accessible world, the scenario that Nicolas
rides the Concorde based on that world is preferable (according to Nicolas) to the
scenario that he doesn’t ride the Concorde.
These definitions are complicated and important in the discussion later in this
chapter, so let’s consider in detail how they might work to determine the truth value
of (12a) in a particular world w∗ . We need to consider worlds doxastically accessible
from a speech-time situation s in w∗ which has Nicolas as its thinking participant.
Pick one of them, w, and suppose that this is a world in which Nicolas does not get
to ride the Concorde. Then, since any world is more similar to itself than any other
world could be, Simw (¬p) is just {w}; in other words, from the point of view of w,
the scenario in which he does not ride the Concorde is just {w}. But we also must
consider Simw (p), the scenario in which he rides the Concorde. Simw (p) does not
contain w, but rather the world or worlds most similar to it in which he rides the
Concorde. Then, the core condition of (14d) will be satisfied if all of the worlds in
Simw (p) are preferable to this doxastically accessible world w. In other words, it is

 Heim uses a more traditional system in which the base for the doxastic accessibility relation is a

world, rather than a situation. This means that her version of (d) has the equivalent of Rdox (a, w∗ )
in the place of our Rdox (s).

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

subsentential modality 55

satisfied if the scenario based on w in which he rides the Concorde is preferable

to the world w itself in which he does not. If this condition is satisfied for every
doxastically accessible world, (12a) is true.
The meaning in (14) is not the final word on a comparative semantics for desire
predicates. Heim discusses several further issues. One is the question of whether
the compared sets of worlds, Simw (p) and Simw (¬p), should be restricted to belief
worlds. Such details will become relevant when we discuss Villalta’s (2008) work in
Section 2.2.2 below (see also von Fintel and Iatridou 2008; Rubinstein 2012).
Because the contrast between belief predicates and desire predicates is central
to patterns of mood selection in many languages, the Stalnaker/Heim analysis
of want will prove important in what follows. However, the analysis was not
developed with the goal of explaining verbal mood. Rather, Stalnaker was inter-
ested in the foundations of possible worlds semantics and links to such topics as
conditionals, while Heim focused on the behavior of anaphora and presupposition
in the complements of attitude predicates. Eventually, the specific form of Heim’s
analysis will become relevant, but what’s important to observe for now is how the
analysis uses the ingredients of the theory of modality in a rather complex way to
express the meaning of a subsentential modal element. Moreover, even though the
Stalnaker/Heim theory did not originate as a theory of mood, it brings out several
ingredients of modal semantics which will prove important in the development of
such theories. Let us list them briefly:

1. Most obviously, the analysis makes use of a doxastic accessibility relation,

Rdox . Unlike the accessibility relation of a sentential modal, this one is
lexically determined, and its thinking agent and temporal component are
grammatically determined.
2. The use of an ordering relation is reminiscent of the function of the ordering
source in ordering semantics. While <s is not defined from an ordering
source in the style of Kratzer’s theory, this may not be an important difference.
On the one hand, ordering semantics can be done without the ordering
source (e.g. Brown 1988; Hansson 1990, 2001; Veltman 1996; Halpern 2003;
Portner 2009); on the other, one could rewrite the definitions in (14) so as to
use an ordering source. There is an important difference between <s and a
Kratzerian ordering source, though: the former is lexically determined, while
the latter is given by context.
3. The analysis connects the semantics of want to the theory of conditionals. The
function Simw (p) is included to pick out what things are like if p is or were to
be true.

The functioning of accessibility relations, the use of ordering, and connections to

conditionals are all important themes in the literature on mood, and we will return
to them in subsequent sections.

De se and de re interpretations. According to the ideas about subsentential

modality we have discussed so far, modal expressions take as a semantic argument a
clause with propositional meaning. The meanings of believe, know, learn, and want
all begin with “λp,” and p is understood to be the standard content of a sentence in

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

56 verbal mood

the possible worlds semantics framework, namely a proposition represented as a set

of possible worlds. And in each case, the semantics of the modal expression is stated
in terms of a set of accessible worlds being a subset of p; for example, in the simple
case of believe, all doxastically accessible worlds are described as in p. Yet there is
strong evidence that this view of subsentential modality is not always sufficient.
Two types of problematical cases are known as de se and de re interpretation.
Chierchia illustrates de se and de re interpretations with the following minimal
pair in Italian (Chierchia 1989, context quoted from p. 3, examples from p. 24).

(15) Imagine Pavarotti looking at a mirror without realizing it and seeing a man
whose pants are on fire. The man Pavarotti is seeing in the mirror is in fact
Pavarotti himself, but he does not realize that.
a. Pavarotti crede che i suoi pantaloni siano in fiamme.
Pavarotti believes that the his pants be.subj.3pl on fire
Ma non si è accorto che i pantaloni sono i proprio.
but not refl is realized that the pants are the his
‘Pavarotti believes that his pants are on fire. But he hasn’t realized that the
pants are his own.’
b. #Pavarotti crede che i propri pantaloni siano in
Pavarotti believes that the his pants be.subj.3pl on
fiamme. Ma non si è accorto che i pantaloni sono i
fire but not refl is realized that the pants are the
proprio.
his
‘Pavarotti believes that his pants are on fire. But he hasn’t realized that the
pants are his own.’

The examples differ in how the anaphor ‘his’ is expressed, and they have different
truth conditions. The version with suoi (15a) can describe a situation in which
Pavarotti is not aware that the person whose pants he sees on fire is himself, while
the one with propri (15b) cannot describe such a situation. Under the assumption
that the meaning of a full clause is always a set of worlds, one would think that the
complement clauses in (15a) and (15b) denote the very same one, namely the set of
worlds in which Pavarotti’s pants are on fire. What is the difference between suoi
and propri here which leads to this difference in meaning?
The belief sentence of (15a) has a de re interpretation. It reports that Pavarotti
has a belief about a particular pair of pants, identified by us as his pants but
identified by him as the pants of the guy he is looking at. The belief sentence of
(15b) has a de se interpretation. It reports that Pavarotti has a belief about a
pair of pants he identifies as his own. It is even more accurate to describe the de se
interpretation as one which reports a belief of Pavarotti about himself. He believes
himself to be a person whose pants are on fire.

 There are many important works which investigate the aspects of meaning related to de se in more

detail. Pearson () and Lasersohn () are especially useful and provide excellent entries into the
related literature.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

subsentential modality 57

The semantic analysis of de re and de se interpretations is a complex topic. To

begin with, the terminology leads us to think that we should classify sentences as
de re, de se, or neither (we’d call this case de dicto), and then assign appropriate
logical forms on this basis. On this way of thinking, there would be a de re logical
form, a de se logical form, and a de dicto logical form. However, de re is really
a property of particular expressions within the modal sentence, and not an overall
property of the sentence. Sentence (16) can be understood de se with respect to
Pavarotti’s pants, but de re with respect to his hat:

(16) Pavarotti crede che i propri pantaloni siano dello stesso

Pavarotti believes that the his pants be.subj.3pl of the same
colore del suo cappello.
color of the his hat
‘Pavarotti believes that his pants are the same color as his hat.’

This sentence reports a situation in which Pavarotti is aware that the pants
he’s thinking about are his own, but might incorrectly think that the hat is
someone else’s.
In addition, as Lewis (1979a) points out, all belief can be thought of as, in a basic
respect, de se. To take the example which we previously described as de re, (15a), this
sentence can be understood as describing a situation in which Pavarotti believes
himself to be someone who is looking at a guy (who happens to be Pavarotti
himself) whose pants are on fire. This again shows that we should think of de re
and de se as features of the meanings of propositional attitude expressions, not as
readings of potentially ambiguous sentences or expressions.
The semantics of de re and de se have played a relatively minor role in theories
of verbal mood, but we will see evidence that de se is actually quite important.
As we go forward it will be necessary to have a basic knowledge about each.
The key feature of de re interpretation is that a phrase identifies a referent and
then contributes that referent (but not the way in which it was identified) to the
propositional content of the sentence in which it occurs. The mechanism by which
it does this is often understood in terms of scope. For example, we can understand
(15a) as resolving the reference of i suoi pantaloni outside the scope of ‘believe.’ It
means, roughly, ‘concerning Pavarotti’s pants, Pavarotti believes that they are on
fire.’ Theories can differ in what they understand to be involved in “resolving the
reference” of a definite description like ‘his pants.’ If ‘his pants’ is a quantifier, it can
simply take wide scope in the ordinary way. If it is an anaphor, the idea would be
that it finds its referent in a higher level of discourse representation. Either way, the
point is that the pants which Pavarotti believes to be on fire are described as ‘his
pants’ not from Pavarotti’s point of view, but rather from the speaker’s.
Given the above, the de re meaning of (17a) could be represented as (17b), in
contrast to the de dicto representation (17c):

(17) a. Pavarotti believes that one violinist is beautiful.

b. ∃x[violinist(x) ∧ Bel(P, x is beautiful)]
c. Bel(P, ∃x[violinist(x) ∧ x is beautiful])

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

58 verbal mood

The de re reading can be brought out by considering a context in which Pavarotti

feels that the orchestra he is working with has a uniformly unattractive violin
section, but is entranced by the sight of one woman waiting on the side. He is not
aware that she is a violinist.
There are problems with this simple scope-based view of de re. The most famous
one is due to Quine (1956):

There is a certain man in a brown hat whom Ralph has glimpsed several times under
questionable circumstances on which we need not enter here; suffice it to say that Ralph
suspects he is a spy. Also there is a gray-haired man, vaguely known to Ralph as rather a
pillar of the community, whom Ralph is not aware of having seen except once at the beach.
Now Ralph does not know it, but the men are one and the same. Can we say of this man
(Bernard J. Ortcutt, to give him a name) that Ralph believes him to be a spy?
(Quine 1956, p.179)

Both of the following sentences appear to be true:

(18) a. Ralph believes that Ortcutt is a spy.

b. Ralph believes that Ortcutt is not a spy.

Scope will probably not help with the analysis here. On the assumption that names
are scopeless, the only representation of (18a) would be (19a). And even on the
questionable assumption that Ortcutt can be assigned scope, both the wide scope
(19b) and narrow scope logical forms would be equivalent to (19a).

(19) a. Bel(R, O is a spy)

b. ∃x[x = O ∧ Bel(R, x is a spy)]

Therefore, it is very difficult to see how to account for the compatibility of (18a–b)
using a system in which the only tool for assigning meanings distinct from (19a)
is scope.
We need to capture the intuition that the beliefs reported in (18a–b) are different
because they involve Ortcutt seen in two different ways, as the man on the beach
and as the pillar of the community. That is, (18a) is true because of Ralph’s belief
that the man on the beach is a spy, and (18b) is true because Ralph believes that the
pillar of the community is not a spy. Both of these can be expressed with the name
Ortcutt simply because of the fact that Ortcutt can be identified both as the man
Ralph saw on the beach and the man Ralph knows as a pillar of the community. In
order to capture the relevance of these two descriptions of Ortcutt to (18a–b), we
introduce a relation of acquaintance into the logical form. An acquaintance
relation C(x, y) holds between x and y iff the first argument x is related to the second
y in a specific way which allows x to have a propositional attitude involving y. In
Quine’s examples, there are two different acquaintance relations in play, roughly
‘saw at the beach’ and ‘pillar of the community.’

(20) a. ∃C[C(R, O) ∧ Bel(R, the x such that C(R, x) is a spy)]

b. ∃C[C(R, O) ∧ ¬Bel(R, the x such that C(R, x) is a spy)]

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

subsentential modality 59

Example (18a) is true under logical form (20a), because C can take the value ‘saw at
the beach’; it is true because Ralph saw Ortcutt at the beach, and believes that the
man he saw at the beach is a spy. Likewise (18b) is true under logical form (20b)
because C can take the value ‘knows as a pillar of the community.’
Next we turn to the de se feature of meaning. Chierchia’s examples (15a–b)
show that propositions are not fine-grained enough to differentiate all of the
beliefs which can be reported using the verb ‘believe.’ Within linguistics, the
most influential solution to this problem, due to Lewis (1979a) building on Perry
(1977), replaces propositions as the objects of attitudes with properties. That is,
the proposition that Pavarotti’s pants are on fire does not serve as the argument
of ‘believe,’ as we have assumed up until now, but rather one of the following
properties does:

(21) a. The property which an individual has in a world iff Pavarotti’s pants are
on fire in that world =
λxλw . Pavarotti’s pants are on fire in w
b. The property which an individual has in a world iff that individual’s pants
are on fire in that world =
λxλw . x’s pants are on fire in w

We are going to use the λ-bound variable x in the second property to account for
the fact that (15b) means that Pavarotti believes that his own pants are on fire. But
in order to give a truly respectable semantics, we need to come up with a meaning
for ‘believe’ different from the ones given above. Instead of being treated as a strong
modal quantifying over the subject’s doxastically accessible worlds, it has to accept
a property type argument and differentiate (15a–b) in the right way.
The difference between the beliefs reported by the two belief sentences in (15a)
and (15b) concerns Pavarotti’s knowledge, or lack thereof, of his role in the scene he
sees. In the scenario, he believes that a large, handsome man’s pants are on fire, but
does not identify himself as that man. The scenario thus indicates that belief is not
just a matter of what the world is like, but also who one is within the world. We
can represent this with a change to our understanding of the doxastic accessibility
relation. Instead of being fundamentally a relation between situations and worlds,
it should be a relation between a situation and an individual and a world.

 The semantics of de re is a complex topic and the analysis used here is at least oversimplified, if not

altogether incorrect. It will suffice, though, as background for investigating verbal mood. See, among
many others, the following for more sophisticated discussion: Lewis (a), Cresswell and von Stechow
(), Percus and Sauerland (), Maier (), Keshet (), Charlow and Sharvit (), McKay
and Nelson ().
 Cherchia’s example does not quite prove this conclusion, since Pavarotti believes that there is a

window several feet in front of the man whose pants are on fire, but in reality what he thinks is a window
is a mirror.
 The formulation of de se semantics here is superficially different from Lewis’s because I wish to

remain consistent with the way modal semantics is presented elsewhere in this work. Note that Lewis
proposes that individuals only exist in a single world, and so for him the world component is not needed.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

60 verbal mood

(22) For any situation s, individual b, and world w: b, w ∈ Rdox (s) iff everything
that the thinking participant a in s believes in s is compatible with a being b
in w.

This new accessibility relation can be used to define a meaning for ‘believe’ which
takes properties as arguments:

(23) Believe denotes the function which takes a property P, individual a, and
situation s as arguments, and returns true iff a is the thinking participant
in s and for every b, w ∈ Rdox (s), P(b, w) = 1

To use Lewis’s way of describing de se belief, according to this semantics ‘A believes

P’ means that A ascribes to himself the property P. This analysis works for (15b): it
says that everything Pavarotti believes in the situation requires that he be a person
who has burning pants. This is not the case in the scenario, and this is why the
sentence is false.
The analysis of (15a) does not work out quite so readily. If we allow ‘believe’ to
take property (21a) as argument, the sentence would be true if everything Pavarotti
believes in the situation requires that his, Pavarotti’s, pants be on fire. This gets right
the fact that he is not aware that the person whose pants are on fire is himself,
but remember that we actually think that i suoi pantaloni has the de re feature
of meaning. The property in (21a) does not capture the fact that he identifies his
pants indirectly, by seeing himself in a mirror. The right property should involve
an acquaintance relation between Pavarotti and his pants:

(24) The property which an individual has in a world iff there is one man who
that individual sees in that world through what appears to be a window, and
that man’s pants are on fire in that world =
λxλw . the pants of the man who x sees in w are on fire in w

The de re sentence (15a) as a whole is true because Pavarotti sees himself (through
what appears to be a window frame), and ascribes to himself the property of seeing
(through what appears to be a window frame) the man whose pants are on fire.
Before concluding this subsection, I would like to point out one important fact
which will become relevant later. Chierchia (1989) shows that some elements are
grammatically required to be de se. Specifically, the subject of a control infinitive
is always understood de se (i.e. as the unsaturated individual argument of the
property complement of a de se attitude verb). Consider (25a):

(25) a. Pavarotti wants to drink espresso.

b. The property which an individual has in a world iff that individual drinks
espresso in that world = λxλw . x drinks espresso in w

 Morgan () identified the difference in interpretation between the null subject of a control

infinitive and the corresponding pronominal subject of a finite clause.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

subsentential modality 61

Roughly speaking, (25) says that every b who has (in w) what Pavarotti wants in
s drinks espresso (in w). The unexpressed subject of to drink espresso corresponds
to the argument x in (25b), bound by the λx. (A more thorough analysis would
combine a de se doxastic modal base with a comparative semantics using a de se
buletic ordering relation, along the lines of (14) but modified to work with prop-
erties rather than propositions.) Crucially, it cannot mean that Pavarotti’s desires
would be satisfied by letting the large man he is looking at drink espresso; to express
that meaning, we would need a sentence which supports a de re reading of the
subject, such as Pavarotti wants for that man to drink espresso (and amusingly he
doesn’t know that that man is Pavarotti himself). Given that an infinitival subject
argument (i.e. PRO) can be obligatorily de se, we might consider the possibility
that other arguments can be obligatorily de se as well. This idea has been taken up
in some of the literature on verbal mood (see Sections 2.2.2.4 and 2.3.1.1).
This discussion has been rather subtle at times, and it will be helpful to highlight
two important points before moving on to other topics. First, there is good reason
to believe that attitude verbs take a property-type, rather than a proposition-
type argument. They should be called “property-oriented attitudes” rather than
“propositional attitudes.” The simple way to put this is to say that attitudes are
fundamentally de se, with other readings being special cases. Second, when a
constituent is understood de re, it is analyzed using an acquaintance relation, and
moreover the acquaintance relation itself is based on de se semantics because it is
between the “self ” of the attitude and a particular “res.” But despite the fact that a de
se semantics is almost certainly appropriate for all attitude verbs, the literature on
verbal mood only rarely uses property-oriented meanings; as a result, we will for
the most part work with logical forms and meanings which reflect the traditional
propositional attiude analysis. We will consider de se and de re interpretation only
when the particular ideas being discussed require it.

Recent issues in the theory of subsentential modality. As mentioned in

Section 1.5, there is a developing set of ideas about modality which differs substan-
tially from the more standard approach discussed so far in this chapter. These ideas
involve several major claims, in particular: (i) that sentences involving sentential
and subsentential modality have a semantics which should be represented in terms
of mathematical probabilities; (ii) that root sentences do not have truth condi-
tions, but rather some type of meaning we might call non-factual or “expressive”;
and (iii) that the semantic function of clauses embedded under propositional
attitude verbs is to impose constraints on mental states which are themselves
appropriately modeled in terms of probabilities. I will label this general set of ideas
the probability-based approach.

 In place of probabilities, we might rather talk of credences. Utilities and expected values might be

involved as well for priority-type modal meanings.

 Not all authors who will be cited as contributing to the probability-based approach accept all of

the major claims I associate with it. For example, Lassiter () assigns truth conditions to modal
statements of the form It is probable that S, and so does not follow point (ii). For different views on (ii),
see Yalcin (, , ), Swanson (, ), and Rothschild ().

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

62 verbal mood

In order to avoid confusion, I would like to be explicit about why I often

use phrases like “mathematical probabilities" in discussing the probability-based
theory. When we talk about the semantics of a predicate like probable, it is no doubt
appropriate to say that its semantics concerns probabilities just as it is appropriate
to say that the predicate intelligent concerns how much intelligence people have.
But a crucial part of understanding the semantics of these items is to figure out
exactly what types of things probabilities and intelligence are. In both of these
cases, there are famous existing concepts which might seem to readily answer the
question: Mathematicians have developed a precise formal model of probability
no doubt known to all readers of this book in at least a basic form: the theory
of probability we invoke when we say things like the probability of a fair coin
coming up heads is .5 or 50 per cent. Similarly, psychologists have developed a
notion of degrees of intelligence known as IQ. Of course, we would not want to
assume that the semantics of intelligent involves IQ. Indeed, many believe that the
IQ measure is fundamentally misguided, and I doubt anyone would want to claim
that X is intelligent just means something like ‘X’s IQ is above threshold i.’ Likewise,
we should not simply assume that the probabilities relevant to probable’s meaning
have anything to do with the mathematical theory according to which they can
appropriately be represented as numbers like .5. We need to keep the semantically
relevant notion separate from the mathematical one so that we can clearly test the
hypothesis that the two should be identified. It is for this reason that, when I wish
to consider the idea that the probabilities relevant to the semantics of probable do
have to do with the mathematical theory, it will sometimes be clearer if I talk about
“mathematical probabilities.”
The probability-based approach draws its motivation from several major points.
The first two are directly relevant to the issue of whether mathematical probabilities
should be used to express the semantics of words which express subsentential
modality:

1. Subsentential modals are often gradable (see Villalta 2008; Portner 2009;
Yalcin 2010; Klecha 2014; Lassiter 2016; Portner and Rubinstein 2016, among
others).

(26) It is more likely that this meat is chicken than squirrel.

(27) It is highly probable that there will be rain tomorrow.

 I don’t wish to suggest that mathematical probabilities have anything like the questionable status

of IQs. The point of the analogy is simply to make apparent that it’s a substantive hypothesis that they
play a role in explaining the semantics of such a word as probable.
 At a deeper level, the issue does not have to do with probabilities represented mathematically, in

the sense of using numbers in the familiar way, as opposed to “non-mathematically.” Kratzer’s ordering
semantics is presented in a set-theoretical system which is “mathematical” in its own way. The real issue
concerns whether something close to probability theory as standardly presented is needed, or whether
a Kratzerian premise semantics giving ordering relations among worlds is more appropriate. Of course,
the ideal theory might be based on a formal model somewhere in between, or something altogether
different. See Holliday and Icard () for useful discussion.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

subsentential modality 63

(28) Your actions are barely permissible.

The comparative construction and modifiers in these examples show that the
modal items likely, probable, and permissible are gradable in much the same
way that non-modal adjectives like tall and happy are. As mentioned above,
while Kratzer’s semantics for modality gives meanings to some sentences
involving gradable modality, it does not do so in a compositional way. It
seems natural to analyze gradable modality in terms of the probability-
based theory by construing scales associated with modal expressions in terms
of probabilities and utilities. Thus, for example, (26) might be analyzed as
conveying that the mathematical probability associated with the meat being
chicken is higher than that associated with the meat being squirrel. But we
should not simply assume that gradability provides evidence for a system
based on mathematical probabilities; it may well also be possible to construct
an appropriate semantics for gradable modality by extending the ordering
semantics system (see Katz et al. 2012; Kratzer 2012; Portner and Rubinstein
2016).
2. Several scholars have argued that Kratzer’s analysis cannot correctly model
the logic of subsentential modals which intuitively have to do with probability,
in particular likely, probable, and certain. The most important of these chal-
lenges come from the work of Yalcin (2010), Lassiter (2011), and Moss (2015),
and while they raise a number of puzzles, we will focus on only one puzzle
for purposes of illustration here. The standard version of Kratzer’s analysis
(Kratzer 1981, 1991) predicts that if a proposition is as likely as its negation, it
is at least as likely as any other proposition whatsoever (Yalcin 2010, p. 922):

(29) φ is as likely as ¬φ.

φ is at least as likely as ψ.

It’s easy to see why (29) holds if we make the limit assumption. In that case,
Kratzer’s (1981) semantics says that, when comparing φ and ψ in likelihood,
we compare the highest ranked worlds making them true. Hence, the premise
of (29) says that the highest ranked φ worlds have the same ranking as the
highest ranking ¬φ worlds. The conclusion says that the highest ranking φ
worlds are at least as highly ranked as the highest ranking ψ worlds—and
this must be the case, because the highest ranking ψ worlds must be either φ
worlds or ¬φ worlds. If a given highest ranked ψ world is a φ world, it can’t
be higher ranked than the highest ranked φ worlds, and if it is a ¬φ world, it
can’t be higher ranked than the highest ranked ¬φ worlds. The probability-
based theory would analyze the argument as follows: the premise states that
the probability associated with both φ and ¬φ is .5, and this clearly does not
rule out that there are other propositions with probability above .5.

 Yalcin’s and Lassiter’s papers demonstrate the problem without the limit assumption.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

64 verbal mood

There is debate about whether this problem afflicts the whole ordering semantics
framework or just the particular version of it cited by Yalcin. Kratzer (2012)
proposes a different definition of the world-ordering which avoids the inference
in (29), though Lassiter (2011) argues that the new definition fails to fully solve the
problem. Klecha (2012) and Katz et al. (2012) provide alternative analyses for some
of the data which had been claimed to support the probability-based approach.
There is no consensus on whether mathematical probabilities, ordering semantics,
or some other tools provide the best basis for the semantics of probability operators;
nor is it clear how representative such operators are of the full range of gradable
modal expressions. What is important here is simply that points like 1–2 above
have motivated the development of a theory of subsentential modality which uses
mathematical probabilities.
Within the literature which develops the probability-based approach, scholars
have noted other important issues in modal semantics. One of them is especially
relevant to this chapter:

3. The semantics for embedded probability expressions does not seem intuit-
ively correct. According to the standard semantics for believe (as a strong
modal) and probably (as expressing Kratzerian human necessity), example
(30) would be true if and only if Mary believes that the highest ranked worlds
(according to the ordering source for probably) are ones in which the thing
being talked about is a chicken.

(30) Mary believes that this is probably a chicken.

Assuming that the ordering source would represent Mary’s beliefs, such an
analysis amounts to saying that she believes that she holds it likely that it
is a chicken. However, the sentence does not seem to report Mary’s belief
about what she holds likely, but rather a judgment that it’s probable, ac-
cording to her beliefs, that it’s a chicken (Yalcin 2007, 2011; Swanson 2011a;
Rothschild 2012). One intuition about this case is that the verb and the
adverb combine into a single modal operator, “x holds . . . likely in terms of x’s
beliefs.”
Several scholars have argued that embedded probability expressions like
probably in (30) point to the need to abandon the assumption that sentences
like (31) express propositions (e.g. Yalcin 2011; Swanson 2011a, 2015).

(31) This is probably a chicken.

According to ordering semantics, (31) expresses a proposition which is true

if and only if the highest ranked (according to the ordering source) worlds in
which the thing is a chicken are more highly ranked than the highest ranked
worlds in which it is anything else. A sentence with these truth conditions
should be consistent with (32):

(32) This is not a chicken.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

subsentential modality 65

Given that they should be consistent, it is surprising that (33a–b) are

incoherent:

(33) (a) ?This is probably a chicken and (but) it’s not a chicken.
(b) ?Suppose that this is probably a chicken and that it’s not a chicken.

The embedded version (33b) poses an especially serious problem. As long as

the proposition expressed by the complement is consistent, John should be
able to suppose that it is true. According to the standard version of ordering
semantics, doing so would amount to making the supposition that his own
evidence suggests that it is a chicken even though it is not; this is the kind
of thing a humble person is willing to suppose. Given that (33b) cannot be
directed even towards a humble person, it appears that the complement of
(33b) does not have the meaning implied by the standard ordering semantics
analysis. From this point, it is a short step for authors like Swanson (2011a,
2015) and Yalcin (2011, 2012) to the conclusion that sentences like (30)–(33)
do not denote propositions which describe the world, but rather are “non-
factual,” “non-descriptive,” and “expressive.” See the papers cited for details.

Though the usefulness of mathematical probabilities is often broached within the

same literature which argues against the assumption that sentences believe and
probable denote assertable, truth-conditional propositions, the claims are largely
separable. In what follows, we will focus on the hypothesis that the semantics of
certain subsentential modal expressions involves mathematical probabilities. We
will return to other issues raised in connection with the probability-based approach
later on; specifically, in Section 2.2.2 we will encounter Yalcin’s analysis of sentences
like (33) as it relates to a particular theory of verbal mood, and in Section 3.3.1 we
will briefly discuss Swanson’s proposal about the discourse function of sentences
like (31).
With these preliminaries out of the way, we can now turn to the question of
how one might incorporate probabilities into the semantics. The simplest way is to
allow the relevant predicates, which take propositions as their semantic arguments,
to describe those propositions in terms of a probability measure derived from
the lexical semantics of the predicate, arguments of the predicate, and contextual
information. For example, along the lines of proposals made by Lassiter (2011) and
Yalcin (2010), probable might mean the following:

(34) [[ probable ]] = λpλw[Pw (p) > .5]

(35) [[ It is probable that it is raining ]] = {w : Pw ([[ it is raining ]] ) > .5}

 These authors do not reach exactly the same conclusions about the semantics of attitude verbs and

probability expressions. I should note that much of Yalcin’s work is presented in an exploratory frame,
so it may be going too far to state that he endorses any particular system.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

66 verbal mood

Following Lassiter, we can assign similar meanings to a priority expression like

good by using a measurement of expected value. A proposition is good if its
expected value is above some threshold t:

(36) [[ good ]] = λpλw[Vw (p) > t]

(37) [[ It is good that it is raining ]] = {w : Vw ([[ it is raining ]] ) > t}

Under the approach in (34)–(37), probability and value information is only relevant
to semantics at the points where a probability operator is being interpreted. It does
not challenge the standard assumption that the meanings of sentences should be
thought of in terms of truth conditions specified as a set of possible worlds.
A more radical version of the probability approach proposes that sentence
meanings themselves involve probabilities. There are several ways to carry this
idea out, and for explanatory purposes I sketch an approach based on Yalcin (2012,
sect. 5) and Swanson (2015). Within standard possible worlds semantics, sentence-
level and modal meanings are all stated in terms of propositions, thought of as sets
of possible worlds. In this version of the probability-based semantics, in place of
possible worlds, we have probability spaces, and in place of propositions we have
credal information states. A credal information state is a set of probability
spaces:

(38) a. A credal information state I is a set of probability spaces.

b. A probability space is a pair c, P, where:
i. c is a set of worlds, and
ii. P is a probability measure over a set of subsets of c.

So how can a credal information state represent the meaning of a sentence?

Consider the simple example (39):

(39) Willie is happy.

If someone says this to you and you say “thanks for telling me,” what have you
learned? It seems that, in terms of probability spaces, you’ve learned that worlds in
which Willie is not happy can be put out of mind, and you don’t need to assign any
probability weight to them. We express this in terms of credal information states
by assigning (40) as the meaning of (39):

(40) [[ Willie is happy ]] = {c, P : c ⊆ {w : Willie is happy in w}}

 Note that P need not assign a probability to every subset of c; whether it does or not depends on

the exact role we want P to play in relating information states to real cognitive states.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

subsentential modality 67

[[ Willie is happy ]] is a set of probability spaces whose domains are limited to

worlds in which Willie is happy. Someone who accepts the information in this
credal information state does not assign any probability to worlds in which Willie
is not happy.
In this approach, an epistemic sentential modal places constraints on the prob-
ability measure P. A simple idea about might would be that it leads the sentence to
denote a set of probability spaces according to which the prejacent has a probability
above zero:

(41) a. Willie might be happy.

b. [[ Willie might be happy ]] =
{c, P : P(c ∩ {w : Willie is happy in w}) > 0}

Although (41b) is not as sophisticated as Swanson’s (2015) proposal, it shares the

key feature that the meaning of both modal and non-modal sentences are sets of
probability spaces.
Now we turn to the semantics of the subsentential modal expression believe in
the credal information state theory. Within the standard possible worlds approach
to attitude verbs, the meaning of believe is given in terms of the set of doxastically
accessible worlds. Since we have provisionally replaced the notion of proposition
(a set of worlds) with a credal information state (a set of probability spaces), it
seems that the set of doxastically accessible worlds should translate into a set of
probability spaces as well. Here’s how that goes:

(42) [[ believe ]] = λpλx[{c, P : c ⊆

{w : for every probability space i which is compatible with x’s beliefs in w,
i ∈ p}}]
(43) [[ John believes that Willie might be happy ]] = {c, P : c ⊆
{w : for every probability space i which is compatible with John’s beliefs in
w, i ∈ [[ Willie might be happy ]] }}

Statement (43) is the set of probability spaces which do not include any worlds in
which it is compatible with John’s beliefs to assign zero probability to all of the
worlds in which Willie is happy.
We have seen two versions of the probability-based approach: a conservative
one which uses probabilities as part of the lexical meanings of subsentential modal
expressions, and a radical version which replaces possible worlds propositions with
credal information states as the central model of semantic content. The question
for us here is whether either of these have any significance for our understanding
of mood. Unfortunately, the answer is not clear. None of the scholars who have
worked on this approach have applied themselves to analyzing verbal mood, and

 One might think that simply assigning all worlds in which Willie is not happy no probability

would be sufficient to record the information that Willie is happy, but Yalcin goes farther than this and
excludes them from c entirely.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

68 verbal mood

so it will be up to us to consider whether the probability-based approach suggests

any new ways of thinking about the topic. It may be useful to make three points:

1. We see a significant difference between the entries for probable and good
given in (34) and (36). The former involves probabilities, while the latter uses
expected values. Expected value is defined as a combination of probability and
utility—roughly, how good an outcome is, weighted for how likely it is. Thus,
we can say that probable uses probability alone, while good uses probability
and utility. If one wants to explain verbal mood within some version of the
probability-based approach, this difference might well be important.
2. We see another significant difference between the entries for probable and
good, on the one hand, and that for believe in (42), on the other. Setting
aside the fact that the latter is more complex because it was used to explain
the radical version of the approach, note that the meanings of probable and
good both involve comparison: they use > to relate the content of their
complements to some standard (.5 or t). In contrast, (42) uses universal
quantification. This difference corresponds to the contrast within Kratzer’s
theory of modality between predicates which employ an ordering source, and
those which do not. As we will see in Section 2.2, one of the most important
approaches to verbal mood is based on the difference between comparative
and non-comparative predicates—the idea is that comparison of propositions
or the use of an ordering source triggers the subjunctive form. While there is
as of yet no reason to think that basing our theory of mood on the probability
approach, rather than some version of ordering semantics, would result in
a better theory of verbal mood, at least we can foresee the possibility of
translating the theory based on comparative vs. non-comparative predicates
into a system that uses probabilities and utilities.
3. Much of the work in developing a theory of verbal mood comes in trying to
analyze particular aspects of the lexical semantics of predicates that embed
indicative or subjunctive clauses. The probability-based approach represents
the meanings of such predicates using different tools, and so, if one were
to develop a probability-based theory of verbal mood, the way to argue in
favor of it would be to claim that the lexical semantic properties relevant
to verbal mood are better captured using those tools. But there is also the
chance that this type of consideration could support the opposite conclusion.
The probability-based approach is quite general and powerful, and this might
make it too easy to represent all subsentential modality in a simple and
uniform way. With such a theory, it might be difficult to explain why the
semantic differences which are reflected in verbal mood are important in
human language.

In sum, nothing we know about verbal mood rules out the probability-based
approach as a theory of subsentential modality, but nothing we know supports
the approach either, and moreover we can foresee where some of the pitfalls for
it would lie. Linguists should seek to evaluate this approach in terms of whether it
can contribute to an explanatory theory of verbal mood.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 69

2.2 Indicative and subjunctive

With this background in place on the semantics of subsentential modal expres-
sions, we can now begin the study of verbal mood. The most basic and well-
known categories of verbal mood are the indicative and subjunctive, and the
most important theories of verbal mood within linguistics aim to explain the
distribution of indicative and subjunctive clauses. In this section, we will study the
indicative/subjunctive contrast with the ultimate goal of developing an integrated
theory of subsentential modality and the semantics of the different verbal moods.
To begin with, in Section 2.2.1, I will give some high-level orientation on the
main ideas in the semantics literature concerning the distribution of indicatives
and subjunctives. Then we will focus in Section 2.2.2 on the main grammatical
contexts in which the subjunctive/indicative contrast has been studied: comple-
ment clauses. That subsection both outlines the empirical patterns which inform
semantic theories of verbal mood and reviews a number of the most important
such theories. As a conclusion to the discussion of indicatives and subjunctives,
Section 2.2.3 reviews recent research on non-argument clauses, including adverbial
clauses, relative clauses, if clauses, and root clauses. My aim in this section and more
broadly in Chapter 2 is to contribute to a better understanding of verbal mood by
identifying the most useful and promising ideas in the existing literature and by
suggesting how to refine those ideas and integrate them with one another. I think
that the results of this chapter point very clearly in the direction of a better theory
of verbal mood, though I will not be proposing a new theory of indicatives and
subjunctives here.
A major difficulty in following the line of research just outlined concerns
crosslinguistic variation. Almost all semantically-oriented work on verbal mood
focuses on a small number of European languages, and while the mood systems
of those languages have much in common, there are also significant differences.
Anyone who wants to understand the literature in this area has to give a lot of
weight to the major, crosslinguistically valid patterns, but also must keep track
of the areas of difference between languages. And it is not at all easy to know
which differences among languages are as important as the general patterns, and
which are idiosyncrasies which should be understood in terms other than semantic
theory. In terms of presentation, this chapter faces a problem of wanting to
identify important cases of variation while not getting bogged down in a detailed
description of every language which has been carefully studied by linguists within
a more or less well-developed semantic perspective. So, as I aim to maintain the
correct balance, readers should keep in mind that many details are not discussed,

 If we look beyond these European languages whose verbal mood systems have been the subject

of formal theories, the variation is much greater, and it is by no means clear that we are dealing with a
unified category. See Wiltschko () for an approach which aims to extend the theory of subjunctive
to languages which are quite different from the standard European ones. In Section ., we will discuss
the notions of reality status and mode and consider their relation to mood.
 For example, they may have historical, morphological, or psycholinguistically-based explanations

which are important to our understanding of language.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

70 verbal mood

and that undiscussed details could turn out to be important. It is up to future

research to figure out whether this is so.

2.2.1 Ideas about the indicative/subjunctive contrast

Semantically oriented research on verbal mood has developed two main intuitions
about the basis of mood selection:

1. Comparison: The subjunctive mood is associated with grammatical contexts

which express a comparison between alternatives in which the clause is true
and some other relevant alternatives. The comparison-based approach has
been connected explicitly to ordering semantics from the time of Giorgi and
Pianesi’s (1997) important book on tense, aspect, and mood.
2. Truth: The indicative is associated with contexts which imply that the clause
is true throughout some designated set of possible worlds. This approach
finds its roots in the work of Farkas (1985, 1992b), and we see an early, direct
implementation of the idea in Portner’s (1997) analysis of Italian.

Both intuitions have their basis in traditional ideas about the contribution of
mood: Comparison connects directly to the idea that mood choice in complement
clauses is determined by a distinction between evaluative/emotive predicates (the
subjunctive) and intellectual/cognitive ones (indicative). Truth relates to the idea
that indicative clauses are realis and subjunctives irrealis.
It is important to keep in mind at this point that we are entertaining two intu-
itions about the indicative/subjunctive contrast, but not necessarily two competing
theories. Indeed, in Giannakidou’s important work on mood in Greek, we see
references to both ideas as she develops her analysis (Giannakidou 1994, 1995,
1997). There are two points I’d like to make about the relation between “comparison”
and “truth.” First, they could be two ways of making the same semantic distinction.
To see the plausibility of this, note that under standard versions of ordering
semantics, a predicate with a comparative meaning like ‘want’ fails to entail that
its argument proposition is true throughout the set of worlds compatible with the
modal base; instead, it only entails that the complement is true throughout the
best-ranked subset of the worlds compatible with the modal base. If we assume
that the worlds compatible with the modal base constitute the “relevant” worlds for
the truth-based approach, a comparative predicate is predicted to take subjunctive
according to both the idea that comparison triggers subjunctive and the idea that
truth triggers indicative. In contrast, a non-comparative predicate like ‘believe’
does entail that its complement is true throughout the set of worlds compatible

 A list of basic references on the semantics and pragmatics of verbal mood from outside of the

formal semantics tradition would include the work of Bolinger (, ), Terrell and Hooper (),
Givón (), Lunn (), Bybee (), Bhat (), Palmer (), Haverkate (), Jary (),
de Haan (), and Gregory and Lunn ().

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 71

with the modal base, and so should take indicative according to both criteria. It
is this perspective—where both of the ideas mentioned above are seen as correct
when expressed within a version of ordering semantics—which we can consider to
be the current state of the art in mood semantics.
The second point about the relation between the two ideas under discussion is
that, rather than being unified into a single idea as part of a deeper analysis, the two
ideas may have independent roles to play in the overall theory of mood selection.
We will see Farkas (2003) pursue this strategy as she attempts to analyze variation
in mood selection crosslinguistically in terms of partially competing requirements.
The idea is that each of the leading ideas above corresponds to a real requirement on
verbal mood, with one favoring indicative, the other subjunctive. When they im-
pose contradictory requirements, a language-specific principle determines which
holds sway. Of course, such a strategy cannot work if the ideas are really two sides
of the same coin.
In the review of particular theories of verbal mood in the following subsection,
I have done my best to appropriately classify particular scholars’ work in terms of
the two main intuitions outlined above, but because of the fact that they can be seen
as two versions of the same intuition, or as both being a part of the overall story,
it is in some cases difficult to do so. In such circumstances, I have tried to make
the classification based on the most explicit hypotheses he or she gives concerning
mood selection. We will return to the various ways in which the two intuitions
might be combined in Section 2.2.2.5.

2.2.2 Semantic theories of verbal mood in complement clauses

In this section, we will discuss theories of verbal mood based on the distribution of
indicative and subjunctive complement clauses. As mentioned above, such theories
typically begin with a classification of predicates in terms of whether they can
combine with an indicative complement, a subjunctive complement, or both. They
then develop some principle which aims to explain the choice of indicative or
subjunctive as the complements of particular predicates in terms of the lexical
semantics of the predicates, possibly in combination with additional pragmatic
factors. For some of the theories to be discussed, the entire analysis rests on the
lexical semantics of the selecting predicates; in such cases, the idea is that verbal
mood simply marks a feature of the clause’s semantic context, without contributing
any relevant meaning of its own. In contrast, other theories also include claims
about the internal semantics of the indicative- or subjunctive-marked clauses
themselves (for example, Portner 1997; Schlenker 2005; Villalta 2008), and the
distribution of each verbal mood is supposed to follow from interaction between
the meaning of the predicate and the meaning of the mood-marked clause itself.

 The alignment between non-comparative meaning and the entailment of truth throughout the

modal base worlds might come apart with negative predicates and predicates with a modal force of
possibility. If the assumptions of the standard ordering semantics analysis of propositional attitudes are
correct, it is with such predicates that we can try to see whether one of the two leading ideas is more
important than the other.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

72 verbal mood

... Outline of data: complement clauses The empirical basis for this discus-
sion is the patterns of mood selection by semantically defined classes of predicates.
I will outline these patterns in terms of the following more-or-less traditional
classes:

1. Predicates of knowledge and belief: words with meanings similar to know

and believe.
These are the classical propositional attitude predicates; they involve an indi-
vidual’s cognitive state which is naturally described in terms of a proposition.
2. Predicates of assertion: words with meanings similar to say that.
These predicates report assertive speech acts of the kind which would be
performed with an indicative declarative clause.
3. Interrogative-selecting predicates: words with meanings similar to wonder
and ask.
These predicates have barely been discussed in the literature on verbal mood.
They should be placed into subclasses:
(a) Predicates of inquisitiveness, such as wonder: the interrogative-selecting
counterparts of predicates of belief/knowledge (wonder means some-
thing like ‘want to know’).
(b) Predicates of inquiry, such as ask: the interrogative-selecting counter-
parts of verbs of assertion (ask means something like ‘wants to be told’).
4. Predicates of fiction and mental creation: words with meanings similar to
imagine and dream.
5. Commissive predicates: words with meanings similar to promise.
6. Perception predicates: words with meanings similar to see, notice, and hear.
7. Preference predicates: words with meanings similar to want, like, and fear.
These report on the individuals’ desires, and more broadly their preferences
and dispreferences.
8. Directive predicates: words with meanings similar to order and require.
These report directive speech acts of the kind which would be performed
with imperative clauses.
9. Causative and implicative predicates: words with meanings similar to make,
cause, and manage.
10. Negative counterparts of predicates of knowledge/belief and predicates of
assertion: words with meanings similar to doubt and deny.
There is a strong intuition that words in this class “contain negation,” in some
sense.
11. Modal predicates: words with meanings similar to possible and necessary.
These are often thought of as the most simple, pure modal words in natural
language, with meanings similar to the  and ♦ of modal logic. Other
predicates which are just as surely modal, for example probable and likely,
are less frequently discussed.
12. Factive predicates: words which presuppose the truth of their complement,
such as regret and remember. This class has two important subclasses:
(a) Neutral factive predicates: factives with meanings lacking a strong
evaluative or emotive dimension, such as remember. Know also falls into

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 73

this class, though it is more typically discussed together with believe as

part of a knowledge/belief class. This class is also known as the semi-
factives (Karttunen 1971), because they seem not to be factive in certain
contexts.
(b) Emotive factive predicates: factives which express an evaluative meaning
similar to what would be seen in a preference predicate, e.g. happy (that),
odd (that), and regret.

The kind of classification given above tends to obscure the issue of how we deal
with lexical items which fall into multiple classes and those with multiple meanings
or multiple uses. Consider the example of require: this verb is listed as directive,
because a sentence like Mary required John to return early could be true because
Mary commanded John to return early through the use of a directive speech act.
However, require is not limited to such uses—we can also say things like Mary’s
health required her to avoid going out in the heat, which does not depend for its
truth on any speech act of the sort which could be performed with an imperative.
Should this latter use be considered to exemplify directive meaning? If so, this
shows that the description of “directive” under point 8 above is quite rough and far
from an adequate definition. Or should it be considered to fall into a distinct class?
Or should we not worry about assigning it a second class at all, on the assumption
that the core uses of require are directive and that this fact determines its relation
to verbal mood? Fundamentally, the question is whether we aim to classify lexical
items as morphological entities, meanings of lexical items, or uses of them. The
answer to this question has both empirical consequences (might we expect require
to combine with different verbal moods on its different uses?) and theoretical
ones (how do we specify the principles which explain the selection of mood, and
are these principles syntactic, semantic, or pragmatic?). There may, of course, be
no single answer; perhaps in some cases the different meanings of a polysemous
predicate are relevant to mood selection, while in others they are not.
With some understanding of these semantic classes of predicates in the back-
ground, we can turn to their mood selection properties. While it is well known that
the classes are not uniform, both in that there are lexical exceptions to the general
patterns of mood selection and in that languages do not all behave alike, many of
the classes are generally understood to have default or typical selection properties.
The literature tends to identify the defaults with the patterns found in the majority
of the most well-known Romance languages. Examples come at the end.

1. Indicative governors
(a) Predicates of knowledge and belief: normally take indicative, though
belief predicates can take subjunctive, as in Italian. The factive ‘know’
also falls in the neutral factive class (see below), and is perhaps better
categorized there. Examples: (44)–(45).
(b) Predicates of assertion: are usually thought of as taking indicative,
because of the fact that they routinely select indicative in Romance
languages. They also are seen as selecting indicative in Greek, but
notice that what the verb actually selects is a clause introduced by the

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

74 verbal mood

complementizer oti; it is not entirely clear whether an oti clause should

be classified as a case of verbal mood, but in any case, it is treated in the
semantics literature as falling under the same theory as verbal mood, so
we include it here. Example: (46).
German does not follow the same pattern as Romance and Greek; pre-
dicates of assertion can take subjunctive or indicative, with no apparent
change in meaning. In addition, German can employ the subjunctive
with other classes of predicates that normally select indicative with the
semantic effect of indicating that someone has said something with the
meaning of the embedded clause. Example: (47).
Although predicates of assertion show crosslinguistic variation in
mood selection, this fact has not been seen as having deep consequences
for the theory of mood. Rather, scholars seem to make an assumption
that the German subjunctives illustrated in (47) exemplify a special use,
the reportative subjunctive, which plays out on top of a more basic
indicative–subjunctive contrast. This is the reason I list verbs of assertion
under “indicative governors” in this summary; I do not, however, wish to
dissuade anyone from an attempt to challenge that assumption.
(c) Predicates of inquiry: have barely been discussed, but initial investigation
points to their selecting the indicative. Example: (48).
(d) Neutral factive predicates: normally take indicative. In Greek, they may
take the indicative complementizer oti used with belief and assertion pre-
dicates (cf. (46b)), or another “factive” complementizer, pu. Giannakidou
(2009) considers them both to mark indicative, but, following Christidis
(1981), describes a semantic effect of complementizer choice, with pu
incorporating a “subjective dimension” of meaning. Example: (49).
(e) Predicates of fiction and mental creation: routinely take indicative. Ex-
ample: (50).
(f) Commissive predicates: routinely take indicative. Example: (51).
(g) Perception predicates: take the indicative (for example, in Spanish). They
take na complements in Greek, which would normally indicate subjunct-
ive, but Giannakidou (2009) argues that the na clause after perception
and aspectual verbs does not mark subjunctive, but rather introduces a
“smaller” clause more similar to an infinitive or gerund. Example: (52).
2. Subjunctive governors
(a) Predicates of inquisitiveness: have barely been discussed, but initial in-
vestigation points to their selecting the subjunctive. Example: (53).
(b) Preference predicates: ‘want’ and other predicates of desire and prefer-
ence typically take subjunctive, though ‘hope’ can take indicative, for
example in French. Examples: (54)–(55).
(c) Directive predicates: routinely take subjunctive. Example: (56).

 Note that some French speakers prefer the subjunctive (Anand and Hacquard ); also note

that Spanish esperar ‘hope’ can take indicative when it means ‘anticipate/expect’, but not when it means
‘hope’, according to Villalta ().

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 75

(d) Causative and implicative predicates: typically take subjunctive, but not
in Greek. Example: (57).
(e) Negative counterparts of predicates of knowledge/belief and predicates
of assertion: routinely take subjunctive. Example: (58).
(f) Modal predicates: The most basic modal predicates, ‘necessary’ and ‘pos-
sible’, take subjunctive, but others, such as probable in French, sometimes
take the indicative. Examples: (59)–(60).
3. A class showing crosslinguistic variation
Emotive factive predicates: may take subjunctive (for example, subjunctive is
preferred in French) or indicative (for example, in Romanian, Greek). Farkas
(1992b, 2003) focuses on the crosslinguistic variation shown by emotive
factives. Example: (61).
While there is abundant crosslinguistic variation in mood choice, with
the emotive factives we see an entire semantic class showing distinct mood
selection properties in otherwise similar languages. For this reason, this case
of variation cannot be downplayed as mere idiosyncratic lexical variation,
and has received a significant amount of attention within semantic theories
of mood.
Examples of mood selection in complement clauses This data is mainly taken
from the semantics literature. Please see the above discussion for an explanation of
what each example is meant to show.

(44) (a) Pierre sait que Marie est heureuse. (French)

Pierre knows that Marie is.indic happy
(b) Pietro sa che Maria è felice. (Italian)
Pietro knows that Maria is.indic happy
(45) (a) Pierre croit que Marie est heureuse. (French)
Pierre believes that Marie is.indic happy
(b) Pietro creda che Maria sia felice. (Italian)
Pietro believes that Maria is.subj happy
(46) (a) Je dis que le temps est beau. (French)
I say that the weather is.indic pretty
(b) O Pavlos ipe oti efije i Roxani. (Greek)
the Paul said that.indic left the Roxani
(47) (a) Er behauptete, dass jemand das Auto angefahren habe,
he claimed that somebody the car on-driven have.subj
…(German)

 Sources are as follows: Portner and Rubinstein (): (a), (a), (a), (), (), (a), (a).

Raffaella Zanuttini, p.c.: (), (). Giannakidou (): (b), (c), (c). Farkas (): (b),
(a,b). Fabricius-Hansen and Saebø (): (). Villalta (): (a), (b), (b). Giannakidou
(): (b). Mulder (): (), (). Blanco (): (a). Quer (): (b). Quer (): (b).
Proust, Le Coté de Guermantes, /: (a). Laca (): (b).

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

76 verbal mood

(b) Ich bat Dr Stroncickij, er möge doch kurz in mein

I begged Dr Stroncickij he may.subj part briefly in my
Büro kommen. (German)
office come
(c) Der Minister war überrascht, dass die EG nicht informiert
the Minister was surprised that the EC not informed
worden sei. (German)
become be.subj
(48) Gli avevo chiesto se ci sono corsi d’inglese. (Italian)
him have-1sg asked if there be.3pl.indic courses of English
‘I asked him whether there are English courses.’
(49) (a) Je me souviens que le temps est beau. (French)
I myself remember that the weather is.indic pretty
(b) Thimame pu/oti ton sinandisa sto Parisi.
remember.1sg that.fact/that.indic him met.1sg in Paris
(Greek)
(50) J’ai rêvé qu’il était président. (French)
I dreamed that he was.indic president
(51) Marie a promis à Bill qu’elle amènerait le dessert à la
Marie has promised to Bill that she bring.cond the dessert to the
fête. (French)
party
(52) (a) Siento que va a haber un problema. (Spanish)
sense.1sg that go.indic to have a problem.
(b) Vlepo tin Ariadne na kolymbai. (Greek)
see.nonpast1sg the Ariadne na swam
(53) Mi chiedo se ci siano corsi d’inglese. (Italian)
me wonder-1sg if there be.3pl.subj courses of English
‘I wonder whether there are English courses.’
(54) (a) Pierre veut que Marie soit heureuse. (French)
Pierre wants that Marie is.subj happy
(b) Victoria quiere que Marcela venga al picnic. (Spanish)
Victoria wants that Marcela come.subj to-the picnic
(c) Thelo na kerdisi o Janis (Greek)
want.1sg subj win the John
(55) (a) J’espère que tu fais toujours l’effort d’écouter tes
I hope that you make.indic always the-effort to listen your
parents. (French)
parents

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 77

(b) Espero enormemente que venga mi hermano. (Spanish)

hope.1sg enormously that come.subj my brother
(56) Il exige que tu partes maintenant. (French)
he demand that you leave.subj now
(57) (a) Juan hizo que María comiera lentejas. (Spanish)
Juan made that Maria eat.subj lentils
(b) Fas que marxi abans d’hora. (Catalan)
make.2sg that leave.subj before of time
(58) Paul doute que tout le monde soit venu. (French)
Paul doubts that everybody be.subj come
(59) (a) Il est possible que cet échantillon soit dissolu dans
it is possible that this sample be.subj dissolved in
l’eau. (French)
the water
(b) És possible que dimiteixi. (Catalan)
Is possible that resign.subj
(60) (a) Saint-Loup a raison et il est probable que le prochain
Saint-Loup has reason and it is probable that the next
Service en Campagne portera la trace de cette évolution, dit
Service en Campagne carry.indic the trace of the evolution said
mon voisin. (French)
my neighbor
(b) Es probable que surjan conflictos. (Spanish)
is probable that arise.subj conflicts
(61) (a) Marie regrette que Paul soit mal. (French)
Marie regrets that Paul be.subj badly
(b) Ion e trist că Maria e bolnavă. (Romanian)
Ion is sad that Maria is.indic sick
(c) O Pavlos lipate pu efije i Roxani. (Greek)
the Paul is sad that.fact left the Roxani

... The link between selecting predicate and complement clause In or-
der to construct a theory of verbal mood in complement clauses, one needs
to understand the relation between a sentence-embedding predicate (such as
‘believe’ or ‘want’) and the clause with which it combines. In simple standard
versions of formal semantics, this relation is just that the predicate denotes a
function which takes the proposition denoted by the clause as its argument. On
this view, the meanings of the predicate and clause are independent, just as the
meanings of the verb and object are in a simple VP like saw Mary. However,
the contrast between indicative and subjunctive clauses shows at least that the

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

78 verbal mood

form of a complement clause can depend on the predicate which combines with
it, and within a semantic theory of verbal mood, this dependency needs to be
encoded somehow in the meanings of the predicate and clause. It would perhaps
be ideal if this dependency could be captured by independently needed differences
in meaning between indicative and subjunctive clauses and between those verbs
which combine with one or the other. For example, a theory in which subjunctive
clauses have a type of meaning which can be naturally combined with ‘want’, but
not with ‘believe,’ would certainly be attractive. However, most semantic theories
of verbal mood propose that the contribution of mood is only to indicate whether
the embedding predicate is appropriate for that particular mood. From this point
of view, mood is a kind of dependent modal element, similar to agreement—we
could say that subjunctive “agrees with” ‘want’ but not ‘believe.’ In order to carry this
idea out in a formal semantic theory, we need a way for the mood marker within
the complement clause to connect to or “see” the relevant features of meaning
of the embedding predicate. This implies that the relation between predicate and
complement is not simply that of a function and argument.
In the literature on verbal mood, we see two ideas about the relation between
a predicate and its complement clause which help to solve the problem of how
indicatives and subjunctives relate to a higher predicate. Because these ideas are
important in themselves, and because different analyses carry them out in slightly
different ways, it will be useful to discuss them in a general way before we dive into
the details of specific analyses. The ideas are as follows:

1. The choice of verbal mood in a complement clause is determined by the

modal parameters of interpretation of the matrix predicate. Within an
approach based on classical modal logic, the relevant parameter is the access-
ibility relation. Within one based on Kratzer’s version of ordering semantics,
several parameters might be relevant: the modal base, ordering source, and
the modal force.
2. The relation between the complement clause and those modal parameters
of interpretation is analogous to the relation between a root clause and the
conversational context in which it is used. For example, suppose we accept
the dynamic semantics/pragmatics idea that the primary function of a root
declarative sentence is to update the common ground. The analogy states that
certain embedded clauses relate to the modal parameters (the accessibility
relation or modal base) associated with their embedding predicate in a way
similar to assertion.

As we will see in the remainder of this section, scholars have given various technical
implementations to these two ideas. Two strategies are fairly perspicuous, and so

 A purely syntactic theory of mood selection would not have this problem. The fact that some

verbs combine with indicative and others with subjunctive would be analogous to the fact that
some prepositions take accusative objects while others take ablative.
 The analyses of Portner (), Schlenker (), and Villalta () have the ambition to provide

explanations of this kind. However, none of them fully manages to analyze verbal mood in terms
of independent semantic values for the predicate and complement clause, and in the end they all
implement one of the methods discussed below.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 79

that it will be clear what is going on when we encounter them below, I would like
to briefly introduce them here. I label the two strategies the method of shifting
parameters and the method of derived contexts.
We will discuss these strategies using a simple sentence with the matrix verb
think, (62):

(62) Ben thinks that the peach is ripe.

The method of shifting parameters proposes that the modal parameters of in-
terpretation for think are available when semantic interpretation applies to the
complement clause that the peach is ripe. The modal parameters for a complement
clause are determined by the matrix verb.

(63) [[ A thinks that φ ]] P0 = [[ thinks ]] P0 ([[ that φ ]] PΘ )([[ A ]] P0 )

The whole sentence is interpreted with respect to some default or discourse-level

parameter P0 . In a basic modal logic, the important parameter is the accessibility
relation; in Kratzer’s system, the relevant parameters include a modal base and
ordering source. Note that both the subject A and the verb think are interpreted
with respect to the same parameters (i.e., we have [[ A ]] P0 and [[ thinks ]] P0 ).
However, in the piece of (63) which references the compositional interpretation
of the complement clause, [[ that φ ]] PΘ , we have a different set of parameters PΘ .
PΘ indicates that, as the complement clause is interpreted, some or all of the
parameters in P0 have been replaced by others which come from the matrix verb
think. Roughly speaking, the contextual parameters P0 are replaced by parameters
which stand for A’s thoughts, such as a doxastic accessibility relation or modal base.
We see this parameter-shifting technique in many analyses of mood, including
Farkas (1992b), Portner (1997), Giannakidou (1997), Quer (2001), Schlenker (2005),
and Anand and Hacquard (2013).
The second strategy for implementing the intuitions listed above, the method
of derived contexts, argues that we should incorporate the update functions of
dynamic semantics/pragmatics into the compositional rules applied to sentences
like (62). Suppose we employ the familiar + notation for contextual update (see
Section 1.4.1). Then, the meaning of a declarative sentence φ is implicitly defined
in the formula c + φ = c . Assuming that the “context” here is merely a set of
worlds (the context set) and that assertive update is merely intersection, we can
say that the assertive meaning of φ is given by c + φ = c ∩ [[ φ ]] . The strategy
for building a theory of verbal mood based on the dynamic approach incorporates
‘+ φ’ (or something similar) into the compositional meaning of sentences like (62).
This might work roughly as follows:

(64) c + A thinks that φ = c ∩ {w : w + φ = c }

 Formally, we can achieve this by treating the modal parameters as parameters of the interpretation

function [[ ]] in a way parallel to how the possible world and model are parameters of interpretation
for formulas of modal logic.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

80 verbal mood

Once again,  is derived from thinks, but in this formula, those parameters are
understood as a “context” which is updated by + φ: w + φ. Since we are thinking
of a context as a context set, a set of worlds, w must be a set of worlds and hence
 itself is a function from worlds to sets of worlds. We already have a name for
this kind of function:  is an accessibility relation, the very prototype of a modal
parameter. In particular, in this case with the verb think, it is a doxastic accessibility
relation: w is the set of worlds in which everything A thinks in w is true.
Formula (64) is not complete because it does not say what c is. Heim (1992)
points out that setting c to be w itself yields a meaning very close to the standard
treatment of propositional attitudes as strong modals. That is, with the meaning
(65), we have an analysis almost equivalent to (7), except that it is built using + φ.

(65) c+A thinks that φ = c ∩ {w : w + φ = w }

Heim seeks to explain the behavior of presuppositions triggered within φ in terms

of the idea that w counts as a derived context in the sense of Stalnaker (1988,
2014). The guiding intuition of this analysis is that w counts as the “context”
for purposes of satisfying the presuppositions of φ. (The theory of presupposition
projection is complex, and we cannot go further into Heim’s theory here. See her
paper for details.)
In the remainder of this chapter, we will repeatedly encounter the idea that
verbal mood is determined by the “context” with respect to which it is interpreted.
Although the intuition will be presented in various ways, in many cases one of the
two strategies identified here—the method of shifting parameters or the method
of derived contexts—will be discernable. It is not yet clear exactly what is at stake
in choosing one or the other of these strategies. In my opinion, they might turn
out to be interchangeable as far as the theory of verbal mood goes, and the choice
between them will probably come down to how well each fits into a broader theory
of context dependency, presupposition, and the relation between verbal mood and
sentence mood.
... Theories based on comparison Portner and Rubinstein (2013) describe
the claim that mood selection is based on the principle that predicates with a
comparative semantics select subjunctive as the “proto-standard analysis of mood.”
They point out that a majority of recent semantically oriented analyses of mood
selection depend on the assumption that mood-selecting predicates have a modal
semantics which is to be given in terms of some version of ordering semantics, and
in particular the idea that we should explain the truth conditions of subjunctive-
selectors by saying that they use an ordering source to rank the proposition
expressed by the complement clause with respect to alternatives. We will label this
approach the comparison-based analysis of verbal mood.
(66) The comparison-based analysis of verbal mood: A predicate P selects the
subjunctive iff its ordering source, gP , is not empty, leading to a comparative
semantics. It selects indicative otherwise. (Portner and Rubinstein 2013,
p. 464)

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 81

We will begin by going through how the comparative analysis deals with the most
well-studied data. Then, we will examine specific proposals in the literature which
can be counted under this analysis, to see why their proposals should be seen as
versions of the comparison-based theory. I think that doing things this way (rather
than explaining each proposal in its own terms, and then deriving the statement
in (66) from them) will make it easier to see how each specific version works, and
will better allow us to identify the most important strengths and weaknesses of this
group of theories.
We can expect a comparison-based theory to make correct predictions on some
of the most well-known patterns:

• Indicative governors with non-comparative meaning

1. Predicates of knowledge and belief: Standardly analyzed as involving uni-
versal quantification over accessible worlds.
2. Predicates of assertion: Standardly analyzed as involving universal quanti-
fication over accessible worlds.
3. Neutral factive predicates: No standard analysis, but there is no obvious
sense in which these predicates compare alternatives.
4. Predicates of fiction and mental creation: No standard analysis, but there is
no obvious sense in which these predicates compare alternatives.
• Subjunctive governors with comparative meaning
1. Preference predicates: Standardly analyzed as involving preference-based
comparison of alternatives.
2. Directive predicates: No standard analysis, but naturally analyzed as in-
volving rule-based comparison of alternatives.

However, there are several classes of predicates which do not, at first glance, fit the
predictions of (66), and others for which it is hard to say whether they should get
a comparative semantics:

• Problematical types
1. Commissive predicates: Naturally analyzed as involving a comparison
between worlds in which the promise is fulfilled and those in which it is
not; the former are ranked higher, according to a deontic ordering, than
the latter. Yet this class normally takes the indicative.
2. Negative counterparts of indicative-selecting predicates: Negation does not
affect the non-comparative semantics associated with the corresponding
positive predicates, yet this class normally takes the subjunctive.
3. Emotive factive predicates: Naturally analyzed as involving comparison
based on the subject argument’s subjective evaluation of alternatives, but
this class shows crosslinguistic variation.
4. ‘Believe’ when it takes the subjunctive, as in Italian: There is no reason that
‘believe’ should be comparative in Italian but non-comparative in other
languages.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

82 verbal mood

• Types about which the approach does not make clear predictions
1. Perception predicates: Standardly assumed to involve a relation to an event
or situation, not a proposition (Barwise 1981).
2. Causative and implicative predicates: The semantics of causatives is highly
controversial, and we would have to accept one analysis in order to identify
predictions.
3. Modal predicates: If they use a non-null ordering source, these are com-
parative and are expected to take the subjunctive. However, weak modal
predicates like ‘possible’ sometimes express simple possibility, and in such
cases would be expected to select the indicative. It is also unexpected
that ‘probable’ would ever take the indicative, since within the order-
ing semantics framework, it is standardly assumed to require a non-null
ordering source.

In the remainder of this subsection, we will see how several authors have un-
derstood and tried to formalize the concept of comparison. Among the works
we will discuss, only Giorgi and Pianesi’s (1997) directly addresses any of the
problems identified here, but even they do not attempt to challenge any standard
assumptions concerning which predicates have a comparative or non-comparative
semantics. Instead, they suggest that other factors (such as factivity) can play
a role in addition to comparison. Thus, I think it’s fair to say that the above
points enumerate the empirical strengths and weaknesses of the comparison-based
analysis.

Giorgi and Pianesi. Giorgi and Pianesi (1997) come very close to giving the
comparison-based theory in the form we have done, as seen in the following:

(67) “…whenever the ordering source is non-null the context of evaluation is

non-realistic, and the subjunctive is consistently selected across languages.”
(Giorgi and Pianesi 1997, p.17)

Though this statement is quite clear, the details of Giorgi and Pianesi’s analysis
make their position a bit less so. The statement above targets all languages, but
they also propose that in some languages a weaker condition governs the choice
of subjunctive. Recall that ‘believe’ in Italian selects the subjunctive. Because they
assume that ‘believe’ has a null ordering source (thereby explaining why it takes
indicative in French), other factors must be in play in Italian. So, to understand
Giorgi and Pianesi’s analysis fully, we first must understand what these other factors
are and how they are integrated into the comparison-based analysis.
Giorgi and Pianesi’s theory is based on classifying predicates in terms of a
Kratzer-style modal semantics. They assume that modal accessibility relations are
defined from conversational backgrounds which function as modal bases and
ordering sources. For example, ‘want’ is treated as a strong modal utilizing a
doxastic modal base and buletic ordering source. They define a series of properties
which conversational backgrounds can have, using these to classify predicates.
Here are the key definitions, derived from Giorgi and Pianesi (1997):

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 83

(68) Given a common ground cg and a conversational background h:

(a) h is realistic
 if h is a subset of the common ground: h(w) ⊆ cg, for all
w ∈ cg.
(b) h is non-realistic if it is not realistic.
(c) h is totally realistic
 if h is identical to the common ground: h(w) = cg,
for all w ∈ cg.
(d) h is weakly  realistic if h overlaps the common ground: h(w) ∩ cg = ∅,
for all w ∈ cg.

(It is actually somewhat unclear in their discussion whether ‘non-realistic’ is

meant to be the complement of ‘realistic,’ or the complement of ‘weakly realistic.’
I understand it as the complement of ‘realistic,’ and so a single background can
be both non-realistic and weakly realistic.) A deontic conversational background
is typically non-realistic, since normally not all rules in a system of rules are
presupposed to be always followed. (Otherwise, why would we need the rules?)
Indeed, a deontic background might not be even weakly realistic, if it is assumed
to be possible that none of the rules would be followed. A doxastic conversational
background is normally non-realistic, because sometimes we talk about people
whose beliefs are incompatible with the information in the common ground.
Giorgi and Pianesi make a number of other specific proposals about the conver-
sational backgrounds for particular lexical items and constructions; for example,
they say that the modal base for an indicative conditional and the ordering source
of the adjective odd are both totally realistic.
Giorgi and Pianesi’s theory also makes use of a concept of the context of eval-
uation (or domain of evaluation) for a complement clause, but it is not clear in
their discussion how the context of evaluation is to be defined. In some places, they
seem to assume that the context of evaluation is simply the set of accessible worlds,
analyzed as in standard ordering semantics. In this sense, the context of evaluation
is the subset of the worlds compatible with the modal base which are best-ranked by
the ordering source (see Section 1.3.2). In other places, they appear to understand
the context of evaluation as a set of propositions somehow derived from the modal
base and ordering source (e.g. Giorgi and Pianesi 1997, ch. 5, fn. 21, 26). Given this
equivocation, and the absence of official definitions, it is quite hard to trace the
predictions of their analysis in detail. Crucially, though, they explicitly state that,
whenever a predicate uses a non-null ordering source, it creates a non-realistic
context of evaluation. We can see this assumption being made explicitly in (67).
They also state that, when a predicate has a non-realistic modal base, its context of
evaluation is non-realistic.
Giorgi and Pianesi propose that in French the subjunctive is sensitive to the
presence of a non-null ordering source; that is, French exemplifies the comparison-
based theory in its pure form. In Italian, in contrast, the subjunctive is chosen
whenever the context of evaluation is non-realistic, but not weakly realistic. Based
on this, they predict that any predicate which selects subjunctive in French will do
so in Italian as well, since a non-null ordering source is assumed to imply a non-
realistic context of evaluation; and in addition, Italian will select subjunctive in
other cases as well, specifically when the modal base is independently non-realistic

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

84 verbal mood

(but not weakly realistic). ‘Believe’ is the key case here: they claim that it has a null
ordering source and a non-realistic (doxastic) modal base, and for this reason that
it selects the subjunctive in Italian.
When Giorgi and Pianesi assume that the doxastic modal base is not weakly
realistic, they are saying that it is not presupposed to overlap the common ground.
This is a crucial point of contrast with ‘say’: they claim that ‘say’ takes a weakly
realistic modal base representing the content of the reported conversation (and
null ordering source), explaining why it takes the indicative in Italian. But this way
of explaining the difference between ‘say’ and ‘believe’ is a weak point in Giorgi
and Pianesi’s theory. There is no clear motivation for proposing that beliefs are
more likely to be disjoint from the common ground than reported conversations.
We sometimes report on the conversation of people we believe to be misinformed,
and conversely we sometimes report on the beliefs of individuals whom we believe
to be well-informed.
There are a number of other points to discuss with respect to Giorgi and Pianesi’s
analysis. They assume that a non-null ordering source always leads to a non-
realistic context of evaluation, but this is only the case if the ordering source is
itself non-realistic, and while they discuss a number of cases where the ordering
source is indeed non-realistic, nothing requires it to be so. With some verbs, it is
plausible that the ordering source would be realistic in some cases. For example,
with ‘regret’ it could be that the ordering source (a set of propositions the subject
disprefers) could all be presupposed to be true in the conversation.

(69) [In light of my dispreference for being so dirty and for being so tired:]
I regret going camping.

In this example, the ordering source of regret is at least weakly realistic. Indeed,
it’s likely that the “dispreference” ordering source is always weakly realistic, since
nearly everybody disprefers some of the things we all know to be true.
Like most theorists of mood, Giorgi and Pianesi do not merely wish to describe
the difference between contexts in which indicative is selected and those in which
subjunctive is selected, but they also want to explain why these two classes align
with the moods they do. And like most theorists, the central idea for explaining
this is that there is something in common between the indicative-selecting contexts
and the use of a root declarative sentence in conversation. Given the tools of modal
semantics which are central to their theory, Giorgi and Pianesi aim to do this by
drawing an analogy between the modal parameters of evaluation for a complement
clause and the context set of an actual conversation. The idea here is that, from
the perspective of the local evaluation of the clause, the two kinds of context play
the same or similar roles, and the indicative mood can be seen as indicating the

 Although Giorgi and Pianesi make a reasonable case that the modal base of ‘say’ would have to

share some propositions with a typical common ground, it seems just as reasonable to assume that
everyone believes some of the propositions in a typical common ground (for example, that itches feel
better if scratched or that the world changes from time to time). Giorgi and Pianesi suggest several
motivations for making the distinction between ‘believe’ and ‘say’ in the way they do.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 85

same property in both types of cases. However, there is a problem with simply
grouping context sets with contexts of evaluation and calling them all “contexts.”
The context set is defined as the intersection of a single set of propositions (the
common ground), while the context of evaluation is defined in terms of two
functions from worlds to sets of propositions (the modal base and ordering source).
Thus, the analogy is only approximate.
It would be natural for Giorgi and Pianesi’s theory to be presented using the
strategy of shifting parameters outlined earlier in the section, but they do not do
so. Instead, they propose that simple root sentences involve an implicit modal
operator. Specifically, they propose that the act of asserting a root declarative
should be modeled using a modal operator M with a modal base m defined to link
the modal to the context set in the following way: m(w) = cg, for every world w
(and a null ordering source). Root assertions would then involve a strongly realistic
context of evaluation, explaining the use of indicative mood for assertion. However,
although their idea is interesting, the particular model of assertion they propose
does not work; note the following derivation, which follows Giorgi and Pianesi
(1997, p.209):

(70) 1. Assertion of S =cg+[[ M(S) ]] m =

2. cg ∩ {w : ∀v ∈  m(w), v ∈[[ S ]] } =
3. cg ∩ {w : ∀v∈ cg, v ∈[[ S ]] } =
4. • cg ∩ ∅, if cg ⊆[[ S ]] ;

• cg ∩ the set of all worlds, if cg ⊆[[ S ]].

The problem is clearly seen between steps 2 and 3: because m does not depend on
w, but just returns cg no matter which world we consider, the proposition added to
the common ground in the end is either the set of all worlds or the empty set. (In
the terminology of dynamic semantics, assertion implements a test on the common
ground.) Now it is of course possible that a different definition of M would work
within Giorgi and Pianesi’s system, but we will not pursue alternatives here.
In addition to the issues raised so far, Giorgi and Pianesi’s analysis has the
problems which we identified earlier which hold generally for the comparison-
based theory, but they make interesting comments on some of the difficult cases.
One important discussion concerns ‘dream’ and other fiction predicates. They
note that these predicates are correctly predicted to select indicative in French,
in which subjunctive is associated with a non-null ordering source, but they see a
problem for Italian, which chooses subjunctive if the context of evaluation is not
weakly realistic, and German, which chooses subjunctive if it is not totally realistic.
I am actually not sure that Giorgi and Pianesi are correct to worry about Italian;
although the context of evaluation for ‘dream’ is non-realistic, it is also weakly
realistic since it is hard to imagine a dream where absolutely no facts in a typical
common ground hold. ‘Dream’ in a language like German poses a more significant
problem, and they say something quite interesting about it. They point out that its

 In addition, as noted above, it’s not clear whether contexts of evaluation are to be defined as sets

of worlds or sets of propositions.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

86 verbal mood

context of evaluation can persist in discourse in a way that many others do not
(based on Giorgi and Pianesi 1997, p.221; see also Portner 1997):

(71) (a) Mario dreamed that Carlo bought a house.

(b) It was beautiful and spacious.

Notice the anaphora between (71a–b). Even though dream is not repeated in (b), it
is possible for it to pick up the house which Mario dreamed about as its referent.
It seems that (71b) is implicitly modalized, with the same context of evaluation as
(a), and so means ‘Mario dreamed that it was beautiful and spacious.’ The fact that
the context of evaluation for ‘dream’ can be used by a root indicative in place of the
context set suggests that this context of evaluation is in some sense “similar to” the
common ground. It is for this reason, they suggest, that ‘dream’ licenses indicative
across all of the relevant languages.
Giorgi and Pianesi also have an interesting discussion of emotive and evalu-
ative factives. Recall that items like ‘regret’ and ‘be sad’ vary in mood selection
properties, taking indicative in Romanian but preferring subjunctive in Italian and
French. (Indicative with these predicates is also possible in French, and marginally
so in Italian as well, according to Giorgi and Pianesi.) While Giorgi and Pianesi’s
discussion of these items is not integrated into their explicit theory of mood
selection outlined above, it is clear that Italian and Romanian are problematical.
On the one hand, Romanian is said to select subjunctive whenever there is a non-
null ordering source, and Giorgi and Pianesi explicitly assume that these predicates
do have non-null ordering sources. (For example, they take the ordering source
with ‘odd’ to be totally realistic and the modal base to be a subset of the com-
mon ground expressing “what is strange.”) Thus, they should take subjunctive in
Romanian, contrary to fact. On the other hand, Italian is claimed to require indic-
ative whenever the context of evaluation is weakly realistic, and both the modal
base and ordering source here are weakly realistic—indeed, the ordering source is
realistic and the modal base totally realistic. Thus, in Italian these predicates ought
to take indicative.
Implicitly acknowledging these difficulties, Giorgi and Pianesi mention two
other factors which might explain these cases: The first factor is that these predic-
ates are factive, presupposing their complements to be true relative to the common
ground; for example, (72) presupposes that John wrote a letter to Mary. The second
factor is that the predicates can be seen as having a causative or conditional
semantics, so that (72) means something like ‘John writing the letter made me
surprised’ or ‘If John had not written the letter, I would not have been surprised’
(Giorgi and Pianesi 1997, p. 219).

 Though Giorgi and Pianesi describe Italian as if it falls under the pure comparison-based theory

(“the presence of a modal component with an ordering source forces the subjunctive in such a language”,
p. ), in the previous discussion they had made clear that Italian takes subjunctive when the predicate
uses a non-realistic, but not weakly realistic context of evaluation. The presence of a (non-null) ordering
source by itself triggers subjunctive in French and Romanian, according to their proposal, but not in
Italian.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 87

(72) It surprises me that John wrote that letter to Mary.

The intuitive idea is that the factivity of ‘surprise’ leads to its selecting indicative in
Romanian, while its causative meaning leads to its selecting subjunctive in Italian.
It seems to me that the first half of this could reasonably be incorporated into Giorgi
and Pianesi’s theory; factivity would be an additional factor, on top of the core
comparison-based theory, which affects mood choice. Later in this section, we will
see Farkas’s (1992b) more explicit attempt to build a theory of mood selection based
on multiple factors.
It seems less likely that factivity will solve the problem of Italian, and so Giorgi
and Pianesi point to the causative or conditional semantics to motivate the idea that
the relevant class of predicates has a non-null ordering source. However, as we’ve
seen, this would not be enough to lead a predicate to select subjunctive in Italian,
according to their analysis. The ordering source would have to be one which gives
rise to a non-realistic context of evaluation. Despite this issue, it is worth keeping
in mind the intuition that a causative or conditional meaning triggers the choice of
subjunctive in some cases, even though we do not yet understand how this feature
of meaning relates to others which contribute to mood selection.

Villalta. Villalta (2000, 2006, 2008) has two important ideas concerning the
analysis of the subjunctive mood. First, she proposes that the type of modal
comparison represented by Heim’s conditional semantics for desire predicates,
discussed in Section 2.1 above, is necessary for explaining why certain predicates
select the subjunctive mood in Spanish complement clauses. Specifically, we can
focus on the relation < in (73): Villalta thinks that this operator is in the semantic
representation for all subjunctive-selecting predicates (and no indicative-selecting
ones) and that its presence is why subjunctive is used in the predicate’s argument.
And second, she argues that Heim’s analysis must be modified because predicates
which select subjunctive mood in Spanish are focus-sensitive and compare mul-
tiple alternatives. She is motivated to modify Heim’s semantics by a problem with
the truth conditions Heim assigns to want. In order to understand how her analysis
develops, it is necessary to examine this problem in detail.
We begin with Heim’s analysis as represented in (73):

(73) [[ want ]] = [λpλaλs . a is the thinking participant in s and for every w ∈

Rdox (s): Simw (p) <s Simw (¬p)]

This definition implies that ‘a wants p’ is true in w∗ if and only if, for every one
of a’s belief worlds w (with respect to some situation s in w∗ in which a is the
thinking paricipant), the worlds most like w in which p is true are preferable to
the worlds most like w in which p is false. Notice that according to this analysis,
the compared worlds (namely the worlds in Simw (p) and Simw (¬p)) need not be
belief worlds. And this is good, because it is possible to want things one believes
certain or impossible (Heim’s 1992 examples).

(74) (a) (John hired a babysitter because) he wants to go to the movies tonight.
(b) I want this weekend to last forever.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

88 verbal mood

Despite the fact that the analysis (73) makes the correct predictions for (74),
Heim rejects it because it fails to explain some other facts having to do with
presupposition projection. In its place, she proposes another analysis, (75) (see also
Villalta 2008; Rubinstein 2012).

(75) [[ want ]] = [λpλaλs . a is the thinking participant in s and

for every w ∈ Rdox (s): Simw (Rdox (s) ∩ p) <s Simw (Rdox (s) ∩ ¬p)]

In this definition, Simw applies not to p and ¬p, but rather p and ¬p restricted
to belief worlds (i.e., Rdox (s) ∩ p). This restriction is useful for explaining certain
examples involving presupposition, but it does not capture the data in (74). To
account for those examples, we need the set of compared worlds to extend beyond
the subject’s belief worlds.
Villalta takes up the challenge which this set of problems poses for the com-
parative analysis of want. Her assumption is that using the set of belief worlds
or any other set to restrict the compared worlds in the semantics, as Heim does,
is a mistake. Rather, she thinks that context provides the compared alternatives,
subject only to the weak constraint that each such alternative be believed possible.
In order to see how this works, we will examine Villalta’s key example. Here is the
background scenario:

Sofía has promised to bring a dessert to the picnic. Victoria believes that there are three
possibilities for what she may actually do. She could prepare a chocolate cake, even though
Victoria considers that extremely unlikely because it represents far too much work. She
might bring an apple pie, which Victoria considers very likely since she can just buy it at
the bakery nearby. Or Sofía might bring ice-cream, which seems most likely to Victoria,
since she usually has some in her freezer. Victoria prefers the chocolate cake over the apple
pie and the apple pie over the ice-cream. (Villalta 2008, p. 476)

In this setting, (76a) is false and (76b) true. But although the example does not
seem to really be a counterexample to Heim’s analysis, it does provide a situation
in which we can clearly see how Villalta’s own analysis works.

(76) (a) Victoria wants Sofía to bring apple pie.

(b) Victoria wants Sofía to bring chocolate cake.

Villalta’s semantics for want is based on there being a contextually relevant set
of alternatives to the propositional argument of the verb; in the scenario, the
alternatives are Sofía bringing cake, apple pie, or ice cream. To be more precise, the
first alternative is not the simple proposition that Sofía brings chocolate cake, but
rather whatever is the contextually relevant version of this—presumably Victoria

 Heim suggests that we do this by using a different set than Rdox (s) in the slot X of Simw (X ∩ p).
 Villalta makes her argument at this point with wish rather than want, but the arguments are just as
applicable to want. Importantly, she argues that the falsity of (a) is incompatible with Heim’s analysis,
but as Rubinstein () shows, this is not the case.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 89

is not thinking about the possibility that Sofía brings a cake on a magic sleigh.
Rather, it is something like the proposition that Sofía brings in the expected way a
cake baked by her in the expected way.
The entry in (77) gives Villalta’s semantics (adjusted for our other assumptions),
where C is the contextually provided set of alternatives:

(77) [[ want ]] = [λpλaλs . a is the thinking participant in s and for all q ∈ C :

q = p → p <s q]

Given the definition of <, (77) amounts to saying that the best worlds among all
of the alternatives areall ones in which p is true; more precisely, for every world
in any alternative (= C), it’s either a p world, or there’s a p world better than it.
In this entry, a’s belief alternatives play no role. The range of worlds we look at in
the semantics of want is not delimited by a semantic component involving Rdox ,
but rather a contextual component, namely the range of worlds in the contextually
relevant alternatives. 
Because Villalta’s analysis boils down to saying that p worlds are best within C,
we can reformulate (77) as a standard necessity modal:

(78) [[ want ]] = [λpλaλs . a is the thinking participant in s and p is a necessity

with respect to mC and Rwant (s)]

Here, mC is a contextually provided modal base which picks out C and Rwant (s)
is an ordering source representing a’s desires in s. According to (78), a wants p is
true
iff all of the “best” (most highly-ranked by the buletic ordering source) worlds
in C are ones in which p is true. This is identical to (77), and we see that C defines
the space of relevant possibilities in a way which can be encoded by a modal base.
Villalta has another reason for formulating the semantics as in (77), namely her
hypothesis that ‘want’ is focus sensitive. Building on Rooth’s (1992) theory of focus,
she proposes that the identity of C is constrained by focus, and hence that the
relevance of contextual alternatives to the semantics of ‘want’ is connected to its
being focus sensitive. From there, she argues all subjunctive-selecting predicates
are focus sensitive. Based on this collection of hypotheses, Villalta then proposes
that the subjunctive encodes focus sensitivity, and that this fact explains why
subjunctive clauses are properly used with the particular class of predicates they
are. Essentially, subjunctive is used with comparative, focus-sensitive predicates.
Moreover, she claims that non-focus-sensitive predicates are strictly incompatible
with the kind of focus encoded by the subjunctive, and so they cannot take
subjunctive complements. It is important to see that, although Villalta adopts a
theory of verbal mood based on comparison, she does not take the common
route of trying to explain the distribution of mood forms in terms of concepts
from modal semantics (like “realistic modal base”) or discourse semantics (like
“assertion”). Rather, the key semantic property has to do with focus.

 Villalta defines the desirability relation < differently from Heim. For Villalta, p < q means that
s
p is a better possibility than q, in Kratzer’s sense of ‘better possibility.’

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

90 verbal mood

Villalta’s treatment of focus in subjunctive clauses has some technical problems

relating to the fact that focus can in general be interpreted in a wide variety
of ways, and it is difficult to set up a system in which the focus marked by a
subjunctive complement clause can only be interpreted by the embedding verb.
For example, why can’t the subjunctive mood be interpreted in connection with a
different focus-sensitive operator, like only, leaving the subjunctive-marked clause
as a whole free to embed under an indicative selector? Her analysis also faces a
problem with its prediction that all and only subjunctive-selecting predicates are
focus sensitive. Portner (2011c) and Harner (2016) raise questions about whether
directive predicates are focus sensitive in the way predicted by her theory, and
Harner additionally suggests that desire predicates are more complex than Villalta
accounts for.

Anand and Hacquard. Anand and Hacquard (2013) study the distribution of
modals in the complements of attitude verbs with the goal of explaining contrasts
such as the one observed in (79) (their (1a), (2a)):

(79) (a) John thinks that Paul has to be innocent. (epistemic reading ok)
(b) John wishes that Paul had to be innocent. (no epistemic reading)

As we see in (79a), when the modal has to is embedded under think, it can get
an epistemic interpretation; on that reading, the sentence means that it is certain,
given John’s beliefs, that Paul is innocent. In contrast, when embedded under wish
as in (79b), had to does not readily have a similar interpretation. As they develop
a theory of this contrast, Anand and Hacquard make use of close connections
between the distribution of epistemic readings of modals and the distribution
of indicative clauses in Romance languages, and so their analysis can be seen as
giving a partial theory of verbal mood. We will review their work with the goal
of understanding what it contributes to the development of the comparison-based
approach to the indicative/subjunctive contrast.
Anand and Hacquard’s analysis of (79) is based on the difference between two
classes of predicates which are closely linked to the selection of indicative and
subjunctive clauses in Romance. On the one hand, they describe the set of predic-
ates which express representational attitudes as “those which ‘convey a mental
picture’; that is, those that describe the content of a propositionally consistent
attitudinal state.” The category of representational attitudes was first proposed by
Bolinger (1968) to characterize the class of indicative-selecting predicates, and
while Anand and Hacquard agree with Farkas (1985, 1992b) that it does not ac-
curately do so, they argue that it is in fact the class which allows embedded modals
to receive an epistemic reading. On the other hand, they endorse a comparison-
based approach to the semantics of those predicates which select the subjunctive
as part of their explanation for why such predicates do not favor epistemic readings
for modals embedded in their complements. To be more precise, they describe the
complements of preference predicates (specifically desideratives except for ‘hope,’
because it can take indicative in French) and directives as “core subjunctives.” (They
do not talk about other predicate classes which tend to select subjunctive across

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 91

the languages they discuss.) Anand and Hacquard’s goal is to show that those
predicates which select subjunctive have a property which makes it very difficult
for an embedded epistemic modal to be interpreted. That property is, essentially, a
modal semantics based on comparison.
Anand and Hacquard adopt Villalta’s (2008) preference-based analysis of the
subjunctive, although they are not committed to the role Villalta assigns to focus.
Acknowledging that there are some problems with the preference-based analysis
(they mention Italian pensare ‘think,’ which selects subjunctive, and ‘promise’
across Romance, which selects indicative), they end up saying that “subjunctive
is an imperfect indicator of preferences.” Despite this imperfection, they use
the subjunctive to argue that desideratives and directives have a comparative
semantics.
Assuming that preference predicates and directives have a semantics based on
comparison, why would this disallow an embedded epistemic? As we noted in
Section 1.3.2, an embedded modal often has its modal parameters controlled by the
higher predicate. For example, in (21c), Ryan said it might be a salamander, Ryan
is the thinking participant in the situation which determines the set of accessible
worlds for might. What Anand and Hacquard are trying to do is explain why the
subject of some verbs (‘say,’ ‘think’) is able to be the thinking participant which
determines a set of accessible worlds for an embedded epistemic modal, while the
subject of other verbs (‘want,’ ‘order’) cannot. This is clearly a difficult problem,
since the subject of ‘want’ clearly is a thinking agent, and so this subject’s beliefs
could define a set of accessible worlds; yet had to in (79b) cannot receive the same
kind of epistemic interpretation.
Because of this problem, Anand and Hacquard develop a more sophisticated
theory which employs the method of shifting parameters outlined earlier in this
section to link the embedded modal to the propositional attitude verb in the
matrix clause. We have already sketched how this strategy works above; citing in
particular Veltman (1996) and Yalcin (2007), they propose the following meanings
for ‘imagine’ and ‘might’ (modified to fit our formalism):

(80) (a) [[ imagine φ ]] S = [λaλs . a is the thinking participant of s and for all
worlds w ∈ img(s), [[ φ ]] img(s) = 1]
(b) [[ might ]] S = [λp . ∃w ∈ S[w ∈ p]]

The parameter S in [[ ]] S is a contextually provided set of possible worlds used in

the interpretation of epistemic modals. We can think of it as the context set or as

 The problem is only present, of course, on the assumption that (b) does not have the interpret-

ation described. Anand and Hacquard acknowledge that it is not altogether impossible for (b)’s had
to to receive an epistemic reading, but they use survey results to show that there is a real difference, and
claim that the epistemic reading in this case is not based on the subject argument’s beliefs/knowledge.
I do not fully agree that had to in (b) can only be epistemic when it targets some body of knowledge
other than John’s. It can mean that John, the investigator, knows that he cannot rule Paul out as a suspect,
but wishes that he could. Anand and Hacquard might say that had to targets the evidence already
discovered by John in the investigation as the conversational background, and not John’s knowledge,
but that is a rather fine line to draw.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

92 verbal mood

the set of worlds epistemically accessible from w∗ . What should be noted is that the
semantic entry for imagine replaces S with a different set of possible worlds, img(s),
the set of worlds compatible with what a imagines in s. Put together, (80a–b) imply
that might in the complement of imagine will be interpreted relative to img(s). Thus,
the complement of (81) means ‘it is compatible with John’s past imaginings that Paul
is innocent.’

(81) John imagined that Paul might be innocent.

Anand and Hacquard assign semantic entries like (80a) to such verbs as ‘say’ and
‘think.’ These are the representational attitudes.
Non-representational predicates have a comparative semantics and combine
with their complements in a different way. Anand and Hacquard’s semantics for
want comes from Villalta, (77), repeated here:

(82) [[ want ]] S = [λpλaλs . a is the thinking participant of s and for all q ∈ C :

q = p → p <s q].

In (82), we don’t see the complement clause within semantic value brackets. There
is nothing like [[ φ ]] img(s) from (80a) in (82). Instead, (82) just works with the
proposition expressed by the complement, p, comparing it to the alternatives. This
means that an epistemic modal embedded in the complement of want can only
have access to the original, contextually provided accessibility relation S. Crucially,
it does not have access to an accessibility relation derived from the matrix predicate.
This fact is what makes it difficult to assign an epistemic interpretation to a modal
in the complement of want. (I’m not sure why it should be difficult to assign
embedded might an epistemic interpretation relative to the contextually provided
set S, but Anand and Hacquard assume that it is. They do agree that it has this
meaning, though.)
The lexical entries for ‘imagine’ and ‘want’ establish a correlation between two
properties: a comparative semantics, and a contextual shift for the complement.
As far as I can see, there’s no deep reason in the theory why these two things are
connected. It would be simple to write an entry which allows a verb like ‘want,’
with a comparative semantics, to perform a context shift. Consider the entry in
(83), which shifts the context for the embedded clause to r(s):

(83) [[ want φ ]] S = [λpλaλs . a is the thinking participant of s and for all q ∈ C :

q = [[ φ ]] r(s) → [[ φ ]] r(s) <s q].

One plausible candidate for r here would be a doxastic accessibility relation, what
the subject believes, and in that case might under want would mean “compatible
with the subject’s beliefs.” Of course, we have no reason to include r here, but the
context shift to img with ‘imagine’ is not motivated by any deep considerations
either; it is simply stipulated to account for the interpretation of an embed-
ded modal. A more explanatory theory would explain why a non-comparative

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 93

semantics goes along with the shifting of the modal background, while a compar-
ative semantics does not., 
Despite these problems, it is important to see the fundamental intuition of
Anand and Hacquard’s analysis. The idea is that one class of predicates utilizes
a set of possible worlds as part of its semantics, and that this set is provided to
the complement clause as a modal parameter. In contrast, another class compares
alternative possibilities and does nothing special to affect the way in which its com-
plement is interpreted. Although Anand and Hacquard do not propose their theory
as an analysis of verbal mood, to my mind it is among the most well-worked-out
versions of the comparison-based semantics of “non-representational” attitudes in
the semantics literature. For this reason, it gives an important perspective on the
comparison-based approach to verbal mood.

... Theories based on truth in a designated set of worlds The second

main idea which semanticists have worked with in an attempt to characterize
the distinction between indicative and subjunctive is that indicative clauses are
associated with grammatical contexts which guarantee the truth of the clause in
some relevant world or set of worlds. This intuition springs from the idea that
indicatives are somehow used with the goal of being true, while subjunctives are
used for some other sort of goal. This approach to verbal mood is closely connected
to the traditional idea that propositions may be distinguished as realis or irrealis,
with the indicative a marker of realis status.

Huntley and Farkas. The truth-based approach to verbal mood is nicely presen-
ted by Huntley (1984, p. 109):

there is intuitively a semantic difference between the indicative and non-indicative clauses.
Where the indicative is used a situation is represented as actual in a way that it is not where
the non-indicative is used, but it proves difficult to make the contrast more precise. One way
of putting it which I will elaborate on shortly is that an indicative clause, even in the future
tense, represents a situation (truly or falsely) as obtaining in the actual world (thought of
as historically extended), whereas the non-indicative clauses represent it as being merely
envisaged as a possibility, with no commitment as to whether it obtains, in past, present or
future, in the actual world.

Farkas (1992b) outlines two problems faced by a truth-based approach. First,

the factive emotives presuppose the truth of their complement, yet select the

 Anand and Hacquard argue that ‘hope’ has a comparative semantics but also makes the subject’s

belief set available to an embedded epistemic. It is thus somewhat like want but licenses embedded
epistemics.
 Anand and Hacquard point out various other constructions which correlate more or less with the

ability to embed epistemics, such as verb second in German and complement preposing in English.
Some of these might be used to argue for a difference in compositional mechanism between the two
classes of verbs, and moreover might allow a connection to the theory of clause types and sentence
mood.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

94 verbal mood

subjunctive in some languages, and second, predicates of fiction and mental

creation do not imply in any sense the truth of their complements, which neverthe-
less take the indicative form. To these points, we should add that belief predicates,
assertion predicates, and commissive predicates fail to imply the truth of their
complements in the actual world. The main effort in developing a truth-based
theory of mood is applied to addressing issues such as these.
Huntley and Farkas propose different approaches to solving these problems.
Huntley’s is presented in a rather sketchy way, but has an interesting intuition:
He proposes that indicative clauses involve indexical reference to a single world of
evaluation, identified as the actual world in root indicatives, and that because of this
reference to a particular world of evaluation, it makes sense to talk about indicatives
as being true or false. In contrast, non-indicatives do not make reference to a world
of evaluation; as a result, even though they technically do have truth conditions,
they cannot normally be said to be true or false.
It is not clear to me exactly how to understand Huntley’s ideas, and it is for
this reason that his analysis has not been very influential in subsequent theories of
verbal mood. It would fit with his explicit discussion to conclude that the indicative
mood is a possibility modal ♦indic which uses the totally realistic accessibility
relation which makes accessible, for any world w, only w itself ( {w, {w} : w ∈ W}).
However, this analysis would not lead to a semantic difference between φ and
♦indic (φ), and so it cannot be right. At this point we might be justified in moving
on, but I believe that there is an interesting way to reconceptualize Huntley’s ideas.
We can understand much of what he writes in terms of the idea that non-indicative
clauses, but not indicative clauses, express a diagonal proposition. Let me briefly
explain what diagonalization is and how it relates to Huntley’s comments about
verbal mood.
We ordinarily understand the proposition expressed by a sentence to be the set
of worlds at which it is true, given that it is interpreted in a context tied to the
world of evaluation w∗ . This conception of sentence meaning can be represented
as in (84a). Crucially, in (84a) the truth conditions are represented by the set of
worlds w, while the context parameter c(w∗ ) remains fixed. In contrast to this,
the proposition expressed by diagonalized clauses loses the separation between the
context c(w∗ ) and world of evaluation w. Instead, it uses an alternate context c(w)
linked to w. The diagonal proposition can be represented as in (84b).

(84) a. Ordinary proposition expressed by φ in c(w∗ ): {w : [[ φ ]] c(w∗ ),w = 1}

b. Diagonal proposition expressed by Dφ in c(w∗ ): {w : [[ φ ]] c(w),w = 1}

 See Stalnaker (, ) and Brown () for broader discussion of the significance of diagon-

alization. The definition in (b) leaves open what type of thing a “context” c(w∗ ) or c(w) is. If a context
is simply a world, then c(w∗ ) = w∗ and c(w) = w, and (b) is a proper definition of diagonalization.
But if a context is something richer, for example a sequence of indices world, time, speaker, . . ., we
face the issue of how the other parameters are affected. These issues become crucial in theories which
use diagonalization to analyze de se readings and shiftable indexicality (Anand and Nevins ;
Schlenker ).

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 95

Under this interpretation of Huntley’s ideas, the indicative indicates that the
context parameter is fixed (as c∗ ) to a particular world, and this (for some reason)
represents the idea that the sentence expresses a proposition which is being used
to say something true or false about that world. In contrast, a non-indicative is
interpreted relative to a context parameter (c(w)) which varies along with the world
of evaluation w, and this plausibly represents Huntley’s idea that the proposition
is not being used to say something true or false about a particular world. This way
of explaining Huntley’s ideas is certainly very sketchy, and in any case it goes well
beyond what he explicitly says. I mention it here only because it strikes me as a
reasonable and interesting way to understand his informal intuitions. See Section
3.1.3 for comments on Huntley’s main concern, the semantics of imperatives.
Farkas’s early work on verbal mood focuses on explaining the patterns of mood
selection in Romance, especially Romanian and French. She proposes that those
predicates which select the indicative introduce a single world at which their com-
plement is entailed to be true, while those which select the subjunctive introduce
a set of worlds (or futures of a world) for this function. When a single world is
introduced, it is called an extensional anchor for the embedded proposition;
when a set of worlds in introduced, it is an intensional anchor. The proposal
about verbal mood is that indicative is used whenever the sentence expresses a
proposition with an extensional anchor.
It is not very clear how to understand Farkas’s ideas in formal terms. One
might think that the anchors should be identified with the set of accessible worlds
provided by the matrix predicate, but this cannot be correct. The semantics of
‘believe’ cannot be given as in (85), since A’s beliefs could never be specific enough
to pick out a particular world:

(85) [[ A believes φ ]] w∗ = 1 iff [[ φ ]] w = 1 in the world w picked out by A’s beliefs.

Rather, as we know from the standard semantics of propositional attitude verbs as

strong modals, ‘believe’ quantifies over a set of possible worlds compatible with the
subject’s beliefs.
Later research in the tradition of Farkas, in particular Giannakidou (1997,
1999, 2009) and Quer (1998, 2001), aims to solve this problem. We will look at
Giannakidou’s and Quer’s work below. However, before getting to that, I would
like to sketch another way of making Farkas’s ideas precise, which may be more
fruitful. It seems to me that Farkas’s idea that indicatives have an extensional anchor
is actually quite similar to Huntley’s proposal that indicatives involve indexical
reference to a particular world at which their truth or falsity can be evaluated.
In combination with the suggestion made above that Huntley’s non-indicatives
denote a diagonal proposition, we reach the following thoughts:

 In previous overviews of Farkas’s ideas, I have understood her proposal in the way given by (),

and have therefore concluded that it cannot be correct as it stands (e.g. Portner c). Here, however,
I will consider another way of interpreting her proposals.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

96 verbal mood

• Indicatives
1. An indicative clause denotes the proposition corresponding to the set of
worlds at which it is true, given a particular context of evaluation c(w∗ ).
2. Predicates which select the indicative anchor the context parameter
for their complement clause to the world which contains the situation
described by the predicate.
• Subjunctives
1. A subjunctive clause denotes the proposition corresponding to the set of
worlds at which it is true, given that the context of evaluation is anchored
to the same world (the diagonal proposition).
2. Predicates which select the subjunctive do not anchor the context para-
meter for their complement clause. Instead, the context parameter is linked
to the world of evaluation by a diagonalization operator.

It seems to me that an analysis of this kind would to some extent capture Farkas’s
ideas about extensional and intensional anchoring.
An important worry about Huntley’s and Farkas’s proposals is that they do not
make deep predictions about which predicates govern indicative and which govern
subjunctive. Although Farkas goes through various classes of predicates and says
what the anchor for each should be (including whether the anchoring is extensional
or intensional), the identity of the anchor does not follow from any analysis of the
lexical semantics of the different classes. It is not at all clear why ‘believe’ has an
extensional anchor but ‘want’ an intensional anchor. Nor is there a fact about the
meaning of ‘dream’ or ‘order’ which would lead to their complements having the
particular type of anchoring they do.
Overall, then, we see in Huntley’s and especially Farkas’s work two points which
are, in fact, endemic to the truth-based approach to verbal mood. On the one hand,
this type of theory has insightful things to say about the meanings of the clauses
marked with indicative and subjunctive themselves. But on the other, they do not
have enough to say about why meanings of those types are used in the particular
semantic contexts that they are. In this way, its strengths and weaknesses are,
from a high-level perspective, the inverse of the strengths and weaknesses of the
comparison-based approach—comparison-based theories offer insightful ideas
about the lexical semantics of predicates which embed indicatives or subjunctives,
but say less about the mood-marked clauses themselves.

Giannakidou and Quer. Giannakidou (1997, 1999, 2009, 2011, 2016), Gian-
nakidou and Mari (2015), and Quer (1998, 2001) develop Farkas’s notion of an
“anchor” in an attempt to make Farkas’s ideas more precise and to apply them to
a wider range of data. The concept of anchor used by Giannakidou is that of an
individual model, a set of worlds provided by applying a special accessibility

 Brasoveanu () proposes that a specific Romanian verbal mood form, referred to as the

conditional-optative or Subjunctive B, encodes both temporal de se and “propositional de se” when

embedded under crede (‘believe’). See Section ... for discussion. Though he does not describe it in
this way, perhaps the propositional de se component could be analyzed in terms of diagonalization.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 97

relation to an individual. The idea can be expressed in terms of the method of

shifting parameters outlined earlier in Section 2.2.2. Suppose that the semantics
of a simple sentence with ‘think’ is computed as in (63), repeated here:

(86) [[ A thinks that φ ]] P0 = [[ thinks ]] P0 ([[ that φ ]] PΘ )([[ A ]] P0 )

Assuming that PΘ is the set of worlds accessible by the doxastic accessibility

relation used by ‘think,’ the meaning of (87) is the following:

(87) [[ Alice thinks that it is raining ]] P0 =

• {w : for some situation s < w: Alice is the thinking participant in s and
∀w [w ∈ Rdox (s) → w ∈ [[ that it is raining ]] Rdox (s) ]}

Giannakidou observes that, according to (87), ‘think’ is veridical:

(88) Veridicality: Suppose O is a propositional operator associated with indi-

vidual model PΘ . Then:
O is veridical iff, for every argument φ, O(φ) entails that PΘ ⊆[[ φ ]] .

‘Think’ is veridical because ‘A thinks that S’ entails that S is true in every world
compatible with what A thinks.
In contrast to ‘think’, Giannakidou gives a semantics for the verb ‘want’ which
makes it non-veridical. Its semantics is given by (89):

(89) [[ Alice wants that it be raining ]] P0 =

• {w : for some situation s < w: Alice is the thinking participant of s and
∀w [w ∈ Rwant (s) → w ∈ [[ that it is raining ]] RFdox (s) ]}

Note the individual model which serves as the shifted context for the complement
clause in (89). I call it RFdox and, according to Giannakidou, it represents the
future realizations of the actual world according to the subject. Of course ‘A wants
that S’ does not entail that S is true in every world which is a possible future
according to A, and so Giannakidou classifies ‘want’ as non-veridical. Based on
this understanding of the semantic difference between belief-type verbs and desire-
verbs, Giannakidou proposes that the indicative is associated with veridicality and
the subjunctive with non-veridicality. We can express this proposal in terms of
licensing conditions on indicative and subjunctive mood:

(90) a. A subjunctive clause Ssubj is licensed as the argument of a non-veridical

operator.
b. An indicative clause Sindic is licensed as the argument of a veridical
operator.

Giannakidou’s explanation of the difference between ‘think’ and ‘want’ relies on

their both anchoring their complements to the doxastic accessibility relation (or
something closely related, like RFdox ), and so, in order to extend this analysis to

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

98 verbal mood

other predicates, Giannakidou must specify, for each matrix predicate, what indi-
vidual model it assigns to its complement. Because ‘dream’ selects the indicative,
she cannot say that it assigns the doxastic or future-possibility accessibility relation.
(If it did, veridicality would not hold, since ‘A dreams S’ does not entail that
A believes S.) Instead, ‘dream’ assigns the “dreams” accessibility relation Rdream .
Similarly, ‘say’ assigns the “says” accessibility relation. But a subjunctive-selector
like ‘order’ does not assign an “orders” relation Rorder , because if it did, it would
be veridical. Rather, as with ‘want’, Giannakidou says that ‘order’ uses the future-
possibility relation RFdox .
Giannakidou’s theory ultimately suffers from the same explanatory gap as Far-
kas’s. Although veridicality captures in an interesting way the intuition behind the
truth-based approach to verbal mood, the fact that it is relativized to a lexically
specified parameter, the individual model, means that the theory does not explain
mood selection unless the particular individual model assigned by a given verb is
motivated by some deeper syntactic, semantic, or pragmatic factor. For example, if
we had a verb with the same core semantics as ‘dream’ but which assigned Rdox , it
would take subjunctive; if we had a verb like ‘want’ which assigned Rwant , it would
take the indicative. The analysis as given so far cannot explain why such verbs
do not exist. As we will see in Section 2.2.2.5, it may be possible to address this
issue by combining the truth-based approach to verbal mood with the comparison-
based approach. Indeed, some comments by Giannakidou point in this direction.
Quer (2001) uses Giannakidou’s concepts of the individual model and anchoring
in a slightly different way which avoids the problem of determining what the
individual model will be for a given predicate. According to Quer, all sentence-
embedding predicates provide their complement with an individual model based
on the predicate’s accessibility relation. The model for the complement of ‘dream’
is Rdream , the model for ‘want’ is Rwant , and so forth. But unlike Giannakidou’s
theory, which says that verbal mood is determined by veridicality assessed relative
to the individual model, Quer proposes that mood is determined based on whether
there is a model shift: subjunctive is chosen when the individual model for a
clause S is sufficiently different from the model of its matrix clause S . We can
see how this works by looking at (87). Quer assumes that the initial, contextually
provided parameter P0 with respect to which the whole sentence is interpreted is
the speaker’s belief model at the utterance time, Rdox (u). As we see, this parameter
has shifted to Rdox (s) in the embedded clause. The idea of the analysis is that this
shift from Rdox (u) to Rdox (s) does not amount to a significant change, because both
are based on Rdox (Quer 2001, p. 88).

 The way Giannakidou’s theory is presented, it might appear that she has a partial explanation. It

is certainly natural that ‘dream’ would use Rdream to derive the individual model—but the question is
why Rdream can be an individual model in the first place. In contrast, she assumes that Rwant is never
the basis for an individual model. In the end, it is just a premise of the theory that the only accessibility
relations which can be turned into an individual model are those for predicates of knowledge and belief,
predicates of assertion, and predicates of fiction and mental creation.
 Giannakidou and Quer call the sets of worlds accessible by a doxastic accessibility relation

“epistemic models,” but I find this choice potentially confusing given that “epistemic” modality is
normally understood to imply not just belief, but actual knowledge.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 99

Now consider sentences with ‘dream’ and ‘want’:

(91) a. [[ a dreams that it be raining ]] Rdox (u) =

• {w : for some situation s < w: a is the thinking participant of s and
∀w [w ∈ Rdream (s) → w ∈ [[ that it is raining ]] Rdream (s) ]}
b. [[ a wants that it be raining ]] Rdox (u) =
• {w : for some situation s < w: a is the thinking participant of s and
∀w [w ∈ Rwant (s) → w ∈ [[ that it is raining ]] Rwant (s) ]}

Quer assumes that the shift from Rdox (u) to Rdream (s) is not a significant model
shift, so indicative is used in the complement of ‘dream.’ In contrast, he takes
the shift from Rdox (u) to Rwant (s) to be significant, and as a result ‘want’ selects
subjunctive. In essence, Quer proposes to group the accessibility relations of all
sentence-embedding predicates into two classes, those which count as similar to
the speaker’s belief model Rdox (u), and those which are different. Subjunctive is
chosen whenever a predicate brings about a shift from the first class to the second.
I think that the most insightful way to look at Quer’s theory is not to focus on
the issue of model shift, but rather on this grouping of accessibility relations into
two sets, those which are similar to Rdox (u), and those which are different from
it. The function of verbal mood is to mark into which group the parameter falls.
Quer (2001) explores some interesting further extensions of his approach to verbal
mood choice in non-complement clauses, a topic we will touch on in Section 2.2.3.

Portner. Portner (1997) develops a theory of verbal mood which, like that of
Quer (2001), assigns to the indicative and subjunctive the function of marking the
modal parameters of interpretation used by a clause. He assigns explicit meanings
to complementizers and attitude verbs which shift the modal parameters for a
complement clause based on the meaning of the matrix predicate. Unlike Farkas,
Giannakidou, and Quer, however, he argues that the parameter is not just a set
of accessible worlds, but rather a pair consisting of an accessibility relation and a
modal force. Adjusting our current notation, meanings for ‘think and ‘want’ would
be as follows:

(92) a. [[ a thinks that it be raining ]] P0 =

• {w : for some situation s < w: a is the thinking participant of s and
∀w [w ∈ Rdox (s) → w ∈ [[ that it is raining ]] Rdox ,F∀ ]}
b. [[ a wants that it be raining ]] P0 =
• {w : for some situation s < w: a is the thinking participant of s and
∀w [w ∈ Rwant (s) → w ∈ [[ that it is raining ]] Rwant ,F∀ ]}

I have notated the force parameter in (92) as F∀ . It matches the universal quantifica-
tion over worlds which takes place within the lexical semantics of ‘think’ and ‘want’;
another way to put the same point would be to say that the verbs are necessity
modals and to refer to the force as ‘’.
It is also worth noting that, according to Portner, the quantification over possible
worlds in (92a)–(92b) is not introduced directly by the propositional attitude verb.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

100 verbal mood

Instead, he proposes that the complementizer of the complement clause triggers

quantification using the modal parameters (accessibility relation and force). That is,
the verb think in (92a) sets the contextual parameters to Rdox and F∀ , and then the
complementizer that introduces an appropriate modal operator C with universal
quantification over belief worlds. To put the idea differently, in Portner’s formaliz-
ation the CP complement is a modal functor whose arguments are provided by the
matrix verb (Portner 1997, p. 195). This system for introducing modal quantification
allows us to understand why parameter-shifting occurs: parameters are the “pipe”
through with the modal force and accessibility relation are transferred to the modal
operator in C.
Given lexical meanings like those given in (92), Portner proposes that verbal
mood is sensitive to the modal parameters of interpretation. In the case of Italian,
he proposes that indicative mood is only possible if the accessibility relation
parameter is classified as realistic (what he calls “factive”):

(93) An accessibility relation R is realistic iff, for every situation s in its domain,
ws ∈ R(s).

(The subjunctive mood is taken to be unmarked, and so it is licensed if the

indicative is not.) Portner’s analysis is similar to Quer’s in proposing that verbal
mood marks features of the modal parameters of the clause, but the two theories
are different in several important respects: First, Portner proposes that verbal mood
marks intrinsic properties of these modal parameters, such as realism, rather than
a “shift” of parameters, as on Quer’s theory. Second, the two scholars offer quite
different treatments of negation; we will return to this point in Section 2.2.3 below.
And third, Portner invokes an analogy between verbal mood and nominal gender
(or noun class marking). He sees the class of accessibility relations which trigger
the indicative (i.e. Rdox , Rdream , Rsay , etc.) as having a status similar to the set
of things referred to by nouns of a given noun class. The parallel is supposed
to go as follows: In some languages, noun classes have a strong semantic basis
but allow for exceptions (he mentions the fact that Yidiny birriny ‘salt water’
falls into the class for drinkable liquids; Dixon 1982). In other languages, noun
class is much more arbitrary (for example, gender in Indo-European languages
which extend natural genders to non-gendered objects). If verbal moods are like
nominal genders, we therefore expect that mood selection will show prototype
effects and some degree of lexical idiosyncrasy, and he claims to find such effects
in several places. One such case is the selection of indicative by ‘say’; he proposes
that while ‘say’ does not literally have the property of realism which licenses the
subjunctive, in “prototypical” contexts it does. Essentially, ‘say’ takes the indicative
because, in the prototypical case, what people say is true. Portner’s theory runs into
serious problems with predicates of imagination like ‘dream’, however. Although it
is reasonable to assume that what people say is prototypically true, it would not
be reasonable to think the same about what people dream (unless, that is, seers

 This treatment of modal quantification is similar to some other proposals based on parameter-

shifting, such as Yalcin () and especially Kratzer ().

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 101

are the prototypical dreamers). While Portner might counter that the selection of
indicative by ‘dream’ is a case of pure idiosyncrasy, such a response would not be
plausible if its mood selection is consistent across languages.

Schlenker. In a series of papers investigating the analysis of indexicality, logo-

phoricity, and de se interpretation, Schlenker (2003, 2005, 2011) makes several
interesting proposals concerning the semantics of verbal mood in French and Ger-
man. To understand his ideas, we have to begin with the phenomenon of shiftable
indexicality. Consider the following example from Amharic (Schlenker 2003):

(94) ȷ̌on jUgna nU-ññ yél-all. (Amharic)

Jon hero be.pf-1so 3m.say-aux.3m
‘John says that he is a hero.’

A morpheme-by-morpheme calque of (94) would be ‘John said that I be a hero.’

In English the I would have to refer to the person who spoke the sentence, but in
Amharic and other “shifting” languages, the first person pronoun can, in certain
circumstances, shift its reference to another individual. In Amharic, the verb ‘say’
allows the first person pronoun in its complement to shift to the referent of the
matrix subject. The intuition about what’s going on here is simple: ‘I’ refers to the
individual who utters a sentence; in English, only the actual utterance counts, but
in Amharic, the reported utterance (in this case, what John said) counts, and the
first person pronoun can thus refer to John.
Schlenker argues that the semantics of a shifted indexical is more subtle than
is captured by this simple description. Most importantly, the shifted indexical
inevitably receives a de se interpretation: For (94) to be true, John had to say
something like “I am a hero”; it would not be enough for him to have said “the
man who shot the General is a hero,” not knowing that he himself was the one who
shot the General. In light of the analysis of de se attitudes (see Section 2.1 above),
this fact implies that the embedded subject of (94) is treated more like a bound
variable than a referential expression.
Schlenker proposes that the shiftable indexical is bound by a special shifting
operator in the complementizer position of the embedded clause. He explores
several ideas for how best to formalize this operator, and while it is not worthwhile
to discuss them in any detail here, it is quite interesting to note that one option is
that the complementizer hosts a version of the diagonalization operator discussed
earlier in this section. We see two options in (95):

(95) a. John say λi, w, t Ii be(w, t) hero

b. John say D Ic be(wc , tc ) hero

 I now believe that Portner’s () proposal that indicative marks prototypical factivity in Italian

is incorrect. The analysis of ‘dream’ in English, based on “prototypical expandability,” is probably closer
to the mark for Italian as well.
 Other theories propose that the shiftable subject is bound by the verb (von Stechow , )

or a special context-shifting operator (Anand and Nevins ).

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

102 verbal mood

In (95a), the shifting operator is an ordinary λ which binds the subject, world,
and time variables in the complement clause. In (95b) we have a diagonaliza-
tion operator, and these same arguments are represented as context-dependent
elements. Since D shifts the context c of the complement clause, these elements
are, in effect, bound by it. It seems to me that there is a potentially productive
connection to be drawn between Schlenker’s analysis and the version of Farkas’s
and Huntley’s ideas reconceptualized in terms of diagonalization.
With this background, we are now ready to see what Schlenker says about verbal
mood. In his body of work, he makes two distinct proposals. The first, applied
to French, is that the indicative marks a clause whose world argument can be
presupposed to be in some accessible context, either the actual utterance context
or a reported context. The following condition on the indicative is slightly different
from Schlenker’s proposal, but captures the essential idea:

(96) For any world w, [[ φindic ]] PΘ (w) is only defined w ∈ PΘ .

The subjunctive is the default form which emerges when the presupposition of the
indicative is not met. This analysis is clearly a version of the family of theories
which associate indicative with truth in a designated set of worlds, and is in
fact quite similar to Giannakidou’s system. Schlenker’s second proposal, applied
mainly to German, is that the subjunctive marks a clause whose world argument is
obligatorily shifted (in other words, “logophoric”). Given the significant differences
between the French and German mood systems, these two proposals are not in
conflict.
Schlenker’s proposal for French is similar to Giannakidou’s analysis. Schlenker
identifies the accessible context used to check the presupposition of the indicative,
PΘ above, with the “context set” of some event in the logical form. Consider the
following semantics for a sentence with ‘think’:

(97) [[ a thinks that it is raining ]] cs∗ =

• {w : for some situation s < w: a is the thinking participant of s and
∀w [w ∈ Rdox (s) → w ∈ [[ that it is raining ]] cs(s) ]}

The root clause is interpreted with respect to the context set of the conversation,
cs∗ . Using the method of shifting parameters, in the complement clause cs∗ has
been replaced by cs(s), the “context set” of the thinking situation s. Because of the
indicative mood, the condition w ∈ [[ that it is raining ]] cs(s) is only defined if w is
in cs(s). Schlenker assumes that in this case, where s is a belief situation, the context
set is the set of doxastically accessible worlds. In an obvious way, Schlenker’s cs(s)
plays the same role as Rdox in Giannakidou’s and Portner’s theories. Similarly, if the
matrix verb is ‘dream,’ s is a dreaming situation, and the context set cs(s) is the set
of worlds in which the dream actually happens (i.e. cs(s) in this case is Rdream ). But

 In reality, Schlenker assumes that the subjunctive competes against both the indicative and

infinitive; this allows him to capture the fact that subjunctive subjects normally cannot corefer with
the matrix subject (the obviation effect; see e.g. Farkas a; Kempchinsky  and Section ...
below). He notes that Portner () and Siegel () also treat the subjunctive as the default form.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 103

if s is a wanting situation, cs(s) is identified with the set of doxastically accessible

worlds, and crucially not with a’s buletically accessible worlds; hence, the condition
in (96) is not met. In this way, Schlenker’s analysis manipulates the context set in a
way very similar to what Giannakidou’s does with the individual model.
One way in which Schlenker’s proposal differs from Giannakidou’s is that his
theory is set up to suggest that cs is a linguistically natural function which assigns
accessibility relations to situations, with the context set of the conversation cs∗
as a prototypical case. Whether this feature of the proposal contributes to a real
explanation depends, in large measure, on our assumptions about what situations
are like. Why does cs return a doxastically accessible set of worlds both when its
argument is a thinking situation and when its argument is a wanting situation—but
not when its argument is a dreaming or promising situation? The idea seems to be
that a wanting situation should be closely connected to a set of beliefs, so that it can
identify the same set of belief worlds, but a dreaming or a promising situation is not
related to a set of belief worlds in the same way. This idea may well be plausible, but
it relies on some assumptions which are not fully clear about the nature of mental
states of various kinds.

... Further issues for theories based on comparison or truth in a designated

set We have organized the discussion of theories of verbal mood in terms of two
main approaches: theories based on comparison and theories based on truth in a
designated world or set of worlds. In this section, we will consider three challenges
to this way of dividing up the field. In turn, we discuss the following points:

1. There is reason to think that the two approaches might be non-distinct.

Some proposals which explicitly fall under the truth-based approach make
reference to the comparative semantics of predicates like ‘want,’ suggesting
that these fail to entail the complement’s truth in the designated set because
they are comparative. On this way of looking at things, a predicate fails the
truth-based condition for indicative mood if it meets the comparison-based
condition for subjunctive mood.
2. Farkas’s (2003) proposal that both comparativity and truth in a designated set
of worlds are relevant to mood selection; they are independent factors which
in combination explain the mood selection properties of a given language.
3. Portner and Rubinstein (2013) argue that neither comparativity nor truth in
a designated set is responsible for mood selection. Instead, they argue for the
importance of contextual commitment.

The possibility that the two approaches are not distinct is suggested by a close
reading of the proposals of authors like Farkas (1992b), Giannakidou (1997, 1999),
Quer (2001), and Smirnova (2011, 2012). As has been mentioned several times, the
main problem which faces theories based on the truth of the complement clause is
that it is difficult to independently motivate the choice of a particular world or set
of worlds against which to check the truth of the complement. The choice of this set
(which goes by such names as “anchor,” “individual model,” “modal context,” and
“context set (of a situation)”) is typically just stipulated as part of the meaning of

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

104 verbal mood

the matrix predicate. However, comments like the following (Giannakidou 1999)
suggest a connection to the comparison-based theory:

The semantics of want-type attitudes proposed here presumes that what one desires is
connected with what one believes, a connection prevailing in the classical treatments of
desire reports, see Stalnaker (1984), Asher (1987), Heim (1992). The connection is done in
terms of preference. (Giannakidou 1999, p.391)

Giannakidou is pointing out here that the only set of accessible worlds mentioned
in Heim’s (1992) analysis of desire predicates is the subject’s belief worlds. The
desirability ordering <s used in giving the meaning of ‘want’ in (14), earlier in
the chapter, is not an accessibility relation, and so according to Giannakidou it
is not the right sort of thing to serve as our PΘ . As a result, non-comparative
predicates will select the indicative, because they quantify over PΘ , while compar-
ative predicates typically will not, because they are not simple universal quantifiers
over PΘ .
Smirnova (2012) gives an analysis of mood selection in Bulgarian which most
clearly combines the comparison-based and truth-based approaches. Her reas-
oning is not quite the same as that sketched above because of the way she
integrates comparativity into the semantics of predicates like ‘want,’ but the key
insight is the same: a comparative semantics breaks the truth-based property
which triggers the indicative. Smirnova’s analysis is interesting from a technical
point of view because she models comparativity using a standard Kratzerian
ordering semantics, rather than the combination of similarity and preference used
by Stalnaker and Heim or the comparison of contextual alternatives proposed
by Villalta. Though it is not quite stated that way, Smirnova’s analysis boils
down to the claim that the indicative is chosen when the ordering source is
empty (or entirely absent), because when it is empty the sentence entails that the
complement is true throughout the modal base. In giving crucial importance to
whether the ordering source is empty or not, her theory is quite similar to that of
Giorgi and Pianesi (1997).
Next we turn to the idea that comparison and truth in a designated set are
both relevant to verbal mood. Farkas (2003) develops the truth-based analysis
of verbal mood in a way that makes clear that the importance of comparativity
is explaining why certain predicates do not select the indicative, but in addition
she makes two important innovations. The first is that she adopts the method of
derived contexts. Thus, the meaning of ‘think’ is represented roughly as in (65),
repeated here:

(98) c+A thinks that φ = c ∩ {w : w + φ = w }

(Recall that w is the modal parameters associated with ‘think’, and for our
purposes can be thought of as Rdox (w), but it is being treated internally to the
semantics as a derived context.) Farkas’s claim is that ‘think’ selects the indicative
because its complement φ updates w using ‘+’ (w +φ), just as a root declarative
S is indicative because it updates the discourse context c (c + S). Farkas describes

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 105

this situation by saying that the complement of ‘think’ has an assertive context
change potential, and she believes that most classes which select indicative,
including predicates of assertion, fiction, and mental creation, have meanings given
in this same assertion-like update.
Farkas proposes that desire predicates and other subjunctive-selecting predic-
ates do not perform an assertive context change, but rather an “evaluative” one. She
does not define this type of context change; the idea seems to be that the meaning
of ‘want’ should look something like the following:

(99) c+A wants φ = c ∩ {w : . . . w φ . . . }

That is, somewhere in the semantics for ‘want’, the complement φ will be used to
update in the evaluative way a derived context w . We use to indicate evaluative
update, and compared to (98), the key point is that replaces +. The evaluative
update triggers the choice of subjunctive mood.
Although her intuitions are intriguing (and I will try to build on them further in
Chapter 4), in fact Farkas does not give a meaning for ‘want’ which fits the pattern
(99). Farkas refers to Heim’s (1992) analysis in which ‘want’ has the meaning that
does not fit well with her proposal (100):

(100) c+A wants that φ = c ∩

{w : ∀w ∈ w [Simw (w + φ) <A,w Simw (w + ¬φ)]}

Within (100), φ performs an assertive update (w + φ), and it seems to me this
would be expected to trigger the indicative. I think the following definitions better
capture Farkas’s ideas:

(101) a. A context c is a pair csc , <c  of a context set csc and an ordering
relation <c .
b. • For any context c and atomic sentence φ, c + φ =
csc ∩[[ φ ]] , <c 
• For any context c and sentence φ, c φ =
{w ∈ csc : Simw (c + φ) <c Simw (c + ¬φ)}, <c 
c. c+A wants φ = csc ∩ {w : w φ = w }, <c 

We assume that a context involves both a set of worlds and an ordering relation and
then define an evaluative update . In the complement of ‘want’, φ evaluates φ
relative to the derived context w . This derived context must give a set of worlds
csΩw and an ordering <Ωw . The set of worlds should be the subject’s doxastically
accessible worlds and the ordering should represent the subject’s desires. With all
of this in place, the ‘want’ sentence means that the subject prefers φ to ¬φ, given
her beliefs.
 There are probably better ways to define , but this one reflects Farkas’s reliance on Heim’s ()

paper. I believe that one of the dynamic updates proposed for imperatives and deontics by the likes of
van der Torre and Tan (), Portner (), or Starr () would work better. See Sections .. and
. for detailed discussion.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

106 verbal mood

The second main contribution of Farkas (2003) is her proposal that verbal
mood is not only sensitive to whether the update is assertive or evaluative, but
also to another factor, decidedness. As we have seen, the assertive/evaluative
contrast corresponds to the difference between non-comparative and comparative
meanings, and so that factor falls under the comparison-based approach to verbal
mood. Decidedness, in contrast, reflects the intuition that the indicative marks the
truth of a clause in a designated set of worlds. We can see the two factors at play
in the mood-selection facts relating to emotive factives like regret and odd (that).
Emotive factives are clearly evaluative, but their complement is true in the derived
context. In the case of ‘regret’(102), the derived context for φ would be doxastic, as
with ‘want’, and the ordering relation < would encode dispreference.

(102) A regrets that φ.

‘Regret’ performs an evaluative update, but because of factivity, the sentence as a

whole guarantees that φ is true throughout the doxastically accessible worlds. This
means that the complement of ‘regret’ counts as decided.
With emotive factives, the comparison-based component (evaluativity) suggests
that the complement should be subjunctive, but the truth-based component (de-
cidedness) suggests indicative. Farkas proposes that, when the two factors are in
conflict, languages may differ in which one takes precedence. In Romanian, we find
that emotive factives select subjunctive; in this language, evaluativity wins out. But
in French, they select indicative, indicating the opposite ranking.
Finally we turn to the skeptical note raised by Portner and Rubinstein (2013).
Based on a study of several less-studied classes of predicates in French, they argue
that neither comparison nor truth in a designated set is sufficient to explain verbal
mood selection. To begin with, they point out that the classic modal adjectives
‘possible’ and ‘necessary’ typically select the subjunctive, but it is not clear that they
always involve ordering in their meanings.

(103) Il est possible que cet échantillon soit dissout dans l’eau.
It is possible that this sample be-subj dissolved in the water
‘It is possible that this sample dissolves in water.’

Here, ‘possible’ seems to indicate nothing more than that there is some circum-
stantially accessible world where the sample dissolves in water. There is no obvious
sense in which the meaning is comparative or evaluative.
Several other predicates raise the opposite issue. Portner and Rubinstein point
out that ‘promise,’ ‘hope,’ and in some cases ‘probable’ take indicative complements,
though their meanings certainly involve comparison or an ordering of worlds.
With none of them is their complement decided, so Farkas’s analysis of emotive
factives does not seem applicable. One option for dealing with these cases is
to consider them as providing examples of prototype effects, as assumed by

 Some French speakers prefer subjunctive with espérer ‘hope,’ but many only allow indicative. See

Schlenker () and Anand and Hacquard () for other ideas about ‘hope.’

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 107

Portner (1997). We could say that, while people don’t always keep their promises,
in prototypical cases they do, and this is good enough to make the complement or
‘promise’ count as decided. Portner and Rubinstein pursue a different idea, though:
they propose that the indicative mood indicates contextual commitment on the
part of the arguments of the matrix predicate.
As defined by Portner and Rubinstein, an individual is contextually committed
to a modal background (a modal base or ordering source) if he is prepared
to defend it, in context, as a good basis for action. The concept is easy to see
with ‘promise’: If I promise to do P, I imply that P, together with my other
promises, is a contextually defensible priority. Portner and Rubinstein argue that
the difference between ‘hope’ and ‘want’ is that hopes must be based on preferences
to which one is contextually committed, but ‘want’ need not be. They say that
‘want’ expresses a “visceral” desire and ‘hope’ an “intellectual” one. With regard
to the modal adjectives, they argue that there are two versions of ‘probable,’ a
normal subjunctive-selecting variant and an indicative-selecting one that requires
the speaker’s contextual commitment towards its ordering source. With ‘possible,’
whether or not it has an ordering source is unimportant, since it never implies
contextual commitment on the part of the speaker. Their analysis receives some
support from the fact that the indicative-marking version of ‘probable,’ for speakers
that allow it, seems to imply a more “objective,” less “subjective,” judgment.

2.2.3 Clauses which are not complements to a selecting predicate

Semantic theories of verbal mood have been developed with the primary aim of
explaining the selection of indicative and subjunctive complement clauses by a
matrix predicate, but at times semanticists have paid attention to verbal mood in
other contexts, and in this section we will outline some of their main ideas and
findings. We will not, however, be able to give anything like an overview of the
distribution of verbal mood forms in non-complement positions, for the simple
reason that the field of semantics has not attempted to produce a general theory
of mood which encompasses all such contexts, or even a significant range of them.
Rather, within the semantics literature we only find sporadic discussions of specific
constructions. In this section, we will consider work on purpose clauses, relative
clauses, if clauses (conditional antecedents), and root clauses. The field could
certainly benefit from more rigorous attempts to extend theories of verbal mood
to a wider range of adjuncts, and in this connection I can point out that Thieroff
and Rothstein (2010), a collection of overview articles on the mood systems of
European languages, includes a great deal of useful information on mood choice
in non-complement clauses.

Adjunct purpose clauses. In a brief discussion, Quer (2001) notes that purpose
clauses select subjunctive in Catalan and aims to explain this fact in terms of his
theory of model shift (Section 2.2.2.4; example Quer 2001, (14)).

 Their idea is closely related to Bolinger’s () claim that ‘want’ expresses a “glandular”
preference.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

108 verbal mood

(104) L’ha convidada perquè no s’enfadi/∗ enfada.

her-have.3sg invited so that not se-get.angry.subj/indic.3sg
‘S/he has invited her so that she will not get angry.’

According to Quer, the purpose clause ‘so that she will not get angry’ must
be subjunctive because it is interpreted relative to a buletic individual model,
specifically here a model representing the desires of the individual referred to by
the matrix subject. In Quer’s theory, the subjunctive is chosen because this buletic
individual model is different from the speaker’s doxastic model Rdox (u) in which
the entire sentence is interpreted. In his theory, the subjunctive is chosen for the
same reason that it would be required in the scope of ‘want.’
Quer is not explicit about how the buletic individual model enters the sentence’s
logical form or what role it plays in the compositional semantics. In the case of
(104), we can speculate that it is contributed by perquè ‘so that.’ We might associate
‘so that’ with a meaning like ‘because x wants and intends p,’ with x bound by the
matrix subject. Given such a logical form, it seems that Quer’s explanation of mood
selection by ‘want’ can indeed be extended to (104).

Relative clauses. In many languages, both indicative and subjunctive verbal

mood occur in relative clauses, and in contexts in which both forms are possible,
the choice between them relates to a difference in meaning. Work by such scholars
as Farkas (1985), Quer (1998, 2001), and Giannakidou (2014) reveals that relative
clauses have subjunctive mood when they are interpreted in the immediate scope of
an attitude verb or other operator which would trigger subjunctive in its argument.
Since relative clauses typically take the same scope as the noun phrase whose head
noun they modify, this pattern can often be described in terms of quantifier scope:
the use of a subjunctive relative indicates that the noun phrase of which it forms a
part is given narrow scope relative to a subjunctive-selecting, intensional operator,
and the use of an indicative relative as part of a noun phrase which appears to be
within the scope of such an operator indicates that it is to be assigned wide scope in
logical form. These generalizations can be illustrated with the following (examples
from Quer 1998, pp. 105, 108):

(105) a. Vull enviar-li regals que el facin

want.1sg send.inf-him presents that him make.subj
content. (Catalan)
happy
‘I want to send him presents that make him happy.’
b. Vull enviar-li regals que el fa content.
want.1sg send.inf-him presents that him make.indic happy
‘I want to send him presents that make him happy.’

In example (105a), the phrase regals que el facin content (‘presents that make him
happy’), with subjunctive mood, is interpreted within the scope of ‘want,’ and as a
result it means roughly that, if the speaker’s desires are satisfied in world w, then

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 109

he gives the recipient presents in w which make the recipient happy in w. In (105b)
the corresponding phrase regals que el fa content, with indicative mood, receives
wide scope, and so means that there are some presents that make this individual
happy in the actual world, and if the speaker’s desires are satisfied in world w, the
speaker gives him those presents in w.
It is also common to describe the contrast between (105a) and (105b) in terms of
specificity rather than scope, saying that the indefinite ‘gifts that make him happy’
is non-specific when it contains the subjunctive and specific when it contains the
indicative. Such terminology is even more natural in examples like (106a)–(106b)
(Quer 1998, p. 121):

(106) a. Estem buscant una intèrprete que sap oromo.

be.1pl seeking an interpreter that know.indic.3sg Oromo
‘We are looking for an interpreter who knows Oromo.’
b. Estem buscant una intèrprete que sàpiga oromo.
be.1pl seeking an interpreter that know.subj.3sg Oromo
‘We are looking for an interpreter who knows Oromo.’

Intuitively, (106a) says that our search will only be satisfied by a specific interpreter,
while (106b) means that it can be satisfied by any interpreter who meets our criteria.
In light of this way of thinking about these examples, subjunctive relatives are
sometimes described as triggering a non-specific interpretation. As Quer points
out, the subjunctive is only determined when the relative clause is in the scope of
an operator with the type of meaning that governs subjunctive. The relative clause
on the definite in (107a) must have indicative mood, because it is in the complement
of the indicative-selecting ‘say’ (Quer 1998, p. 116):

(107) a. El Pau diu que el noi que va trucar ahir

the Pau says that the guy that aux.indic.3sg called yesterday
era grec.
be.indic.3sg Greek
‘Pau says that the guy that called yesterday was Greek.’
b. ∗ El Pau diu que el noi que truqués ahir
the Pau says that the guy that call.pst.subj.3sg yesterday
era grec.
be.indic.3sg Greek
‘Pau says that the guy that called yesterday was Greek.’

The concept of specificity is especially salient when the noun phrase containing
the relative clause is an argument of the matrix predicate, as in the following (Quer
1998, p. 119):

 In a very different way, Baker and Travis () propose that verbal mood in Mohawk encodes

definiteness.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

110 verbal mood

(108) El que hagi assassinat tants d’algerians és

the that have.subj.3sg murdered so many of Algerians be.indic.3sg
una bèstia infame.
a beast infamous
‘The one that murdered so many Algerians is an infamous monster.’

With the subjunctive relative, the phrase El que hagi assassinat tants d’algerians (‘the
one who murdered so many Algerians’) is not understood to refer to a particular
individual whom the speaker has in mind; rather, the sentence can be paraphrased
as something like “the person who murdered so many Algerians, whoever he may
be, is an infamous monster.” Following Farkas, Quer connects this semantic effect
to Donnellan’s (1966) attributive reading and proposes that the sentence contains
an operator (in this case, a covert modal ‘must’) to produce the intensional context
which licenses the subjunctive. He thus takes the sentence to be equivalent to “the
person who murdered so many Algerians must be an infamous monster,” with
‘must’ having wider scope than the subject.
Among the theoretical tools designed for the analysis of verbal mood which have
been discussed in this chapter, the method of shifting parameters appears to be
well-suited to developing an analysis of mood choice in relative clauses. Consider
the meaning ‘want’ in (92b) above:

(109) a. Paul wants to find an interpreter who knows Oromo.

b. [[ Paul wants (Paul) find an interpreter who knows Oromo ]] P0 =
• {w : for some situation s < w:
Paul is the thinking participant in s and
∀w [w ∈ Rwant (s) → w ∈ [[ Paul finds … ]] Rwant , F∀ ]}

In (109b), the relative clause (an interpreter) who knows Oromo is interpreted within
the scope of ‘want,’ and as a result its modal parameter of interpretation has shifted
from the original P0 to Rwant . If we understand the mood of the relative clause
to be sensitive to this modal parameter (as it is according to several of the theories
discussed in Section 2.2.2), it is correctly predicted to be subjunctive in Catalan and
other languages with similar patterns of mood choice in relative clauses. However,
if the phrase an interpreter who knows Oromo were assigned wider scope than
‘want,’ its modal parameter would be P0 , and indicative would be the correct verbal
mood for the relative clause. It seems likely that the examples in (105)–(107) could
be explained by a theory along these lines. It is much trickier to explain the use
of the subjunctive in the root subject position, (108). Just saying that it contains
an unpronounced ‘must’ is not sufficient; the modal would have to shift the modal
parameter to one which triggers subjunctive.

Negation. Subjunctive mood often appears in subordinate clauses when the

matrix is non-affirmative, for example when it is negated or interrogative. The
following examples show the subjunctive as the complement of verbs which would
select indicative in an affirmative sentence (from Mulder 2010, p.170; Quer 2010,
p.227):

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 111

(110) a. Pierre ne croit pas que tout le monde soit venu. (French)
Peter neg believe neg that everybody be.subj come
‘Peter does not believe that everybody has come.’
b. Recordes que els hagin donat mai un premi? (Catalan)
remember that them have.subj given never a prize
‘Do you remember if they have ever given them a prize?’

Quer (1998, 2010) classifies these cases as examples of the polarity subjunctive (or
“dubitative” subjunctive), which he also identifies as occurring under predicates
like ‘doubt.’ He argues that they have very different properties from the inten-
sional or “volitional” subjunctive selected by ‘want’ and ‘order.’ In particular, Quer
points out that the dubitative contexts license subjunctive in multiple levels of
embedding, as in (111), and that they do not show the same sequence of tense and
subject disjoint reference properties as volitional subjunctives (Catalan example
from Quer 2010, p. 228):

(111) Dubto que creguin que ens hagin convençut amb això.
Doubt.1sg that believe.subj that us have.subj convinced with this
‘I doubt that they believe that they have convinced us with this.’

We should not, however, understand Quer to be proposing that there are two
distinct but homophonous subjunctive forms, since in his work on model shift he
tries to explain cases like (110a) using the same theory as he applies to subjunctives
selected by a volitional predicate. This analysis is problematical, however. Quer
(2001, p. 91) proposes that subjunctive is triggered under negation because the
individual component of the model shifts; in the case of (110a), it shifts from the
speaker to Pierre. I do not see how this proposal can fit into his overall theory,
though, since even in the non-negative version of this example, the same model
shift would occur.
Giannakidou offers a straightforward analysis of the subjunctive under negation.
Recall that her theory is based on the idea that the subjunctive is sensitive to
non-veridical contexts, and negation is clearly a non-veridical operator. Thus it
appears easy to explain (112) by saying that dhen licenses the subjunctive (from
Giannakidou (2011, (39)):

(112) Dhen pistevo na irthe o Janis. (Greek)

neg believe.1sg subj came the John
‘I don’t believe that John came.’

It remains unclear, however, why negation cannot trigger the subjunctive in its own
clause. Perhaps negation is not in the right structural position to do so.

 See Quer’s work for additional examples of these properties. Note that subjunctive relatives do

not consistently show the properties which Quer associates with either type of subjunctive, and so it is
difficult to assign them to either class (Quer ).

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

112 verbal mood

Portner (1997) treats cases like (110a) and (112) in a way which explains the fact
that negation only leads to the subjunctive in a subordinate clause. In intuitive
terms, he proposes that negation and the verb it negates combine to yield a
complex modal meaning, and that the resulting modality licenses the subjunctive.
Recall that Portner follows the method of shifting parameters, but he utilizes two
such parameters: an accessibility relation and a modal force. In the affirmative
sentence I believe that John came, the modal parameters for the embedded clause
are the doxastic accessibility relation Rdox and the force of necessity F∀ . But when
the matrix clause is negated, as in (112), he argues that negation is integrated into the
modal force, resulting in the force of non-necessity F¬∀ . (The sentence is therefore
true if not all of the speaker’s belief worlds are ones in which John came.) Portner
proposes that the indicative is incompatible with the force of non-necessity, and so
the subjunctive is selected in the negative environment.

Conditionals. In many languages, conditional ‘if ’ clauses can have the subjunct-
ive mood. Consider the following examples from Catalan (Quer 2010, p.232):

(113) a. Si té gana, agafarà una galeta.

if have.prs.indic hunger, take.indic a cookie
‘If s/he is hungry, s/he will take a cookie.’
b. Si tingués gana, agafaria una galeta.
if have.pst.subj hunger, take.cond a cookie
‘If s/he were hungry, s/he would take a cookie.’
c. Si hagués tingut gana, hauria agafat una galeta.
if have.pst.subj had hunger, have.cond taken a cookie
‘If s/he had been hungry, s/he would have taken a cookie.’
d. Si tenia gana, agafaria una galeta.
if have.pst.indic hunger, take.cond a cookie
‘If s/he were hungry, s/he would take a cookie.’

Following Quer, we can describe the differences in meaning among these examples
in terms of tense and the contrast between real and unreal conditionals. With the
present indicative if clause, (113a) is a real conditional: it leaves open whether the
referent of the subject will, in fact, be hungry. The past subjunctive (113b) is an
unreal conditional about the present: it suggests that this individual is not in fact
hungry. Example (113c), with the past perfect subjunctive, is an unreal conditional
about the past: it suggests that he or she was not hungry.
In philosophical work on conditionals, it is common for unreal conditionals
to be described as “subjunctive conditionals,” and this terminology makes sense
given the correlation between subjunctive mood and the unreal interpretation in
(113a–c). However, as pointed out by Iatridou (2000), crosslinguistic investigation
reveals that the unreal interpretation actually depends on the past marking of the

 Such examples are often called “counterfactuals,” but as pointed out by Anderson () and many

subsequent authors, they are compatible with certain contexts in which the antecedent clause is true.
For this reason, the term “unreal” is preferable. See Portner (, ch. ) for further discussion.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

indicative and subjunctive 113

if clause. Subjunctive form may be (and sometimes must be) used if the language
has an appropriate past subjunctive form, but it seems better to understand the
subjunctive as licensed by the unreal interpretation, rather than triggering it. In
this context, note that Quer’s example (113d), with the past (imperfect) indicative,
has a meaning very close to (113b); this fact further suggests that the subjunctive
mood has only a marginal role in deriving the unreal interpretations.
Looking beyond the if conditionals, mood choice seems to play a role in less
well-studied types of adverbial clauses with conditional-like meaning. For example,
French employs subjunctive in clauses introduced by à condition que (‘on condition
that’), avant que (‘before’), de sorte que (‘so that’), quoique (‘although’) (see Mulder
2010 for examples). Quer (1998, 2001) examines mood choice in concessive clauses.
He points out that a clause introduced by encara que (‘although’) in Catalan means
‘although’ with indicative mood, but ‘even if ’ with subjunctive. This variation
seems similar to that observed with German wenn (‘when/if ’) clauses, where
indicative marking tends to lead to the meaning ‘when’ and subjunctive to ‘if ’
(Thieroff 2010, pp.141–4). In many of these cases, it is natural to think of the role
of subjunctive as reflecting or contributing some kind of modal meaning and, in
his study of concessives, Quer (1998, 2001) develops this intuition in a plausible
but relatively informal way. The field could certainly benefit from more precise
and compositionally detailed theories of verbal mood in conditionals and other
adverbial clauses.

Root clauses. The canonical verbal mood of root declaratives and interrogatives
is, of course, the indicative. Certain theories of sentence mood assign a crucial
role to the indicative verbal mood (for example Truckenbrodt 2006a, Lohnstein
2007; see Chapter 3). Still, we do find root subjunctive clauses with various non-
declarative, non-interrogative discourse functions. Portner (1997, pp.192–3) gives
examples of the Italian subjunctive expressing an order, wish, supposition, and
astonishment:

(114) a. Tenga le mani a posto.

hold.subj.3sg the hand in place
‘Keep your hands in place!’
b. Il Signore ci protegga.
the Lord us protect.subj.3sg
‘May the Lord protect us!’
c. L’avesse anche detto lui.
it-have.subj also said him
‘Suppose he had said it too.’
d. Che sia nel bagno?
that be.subj in-the bath
‘She is in the bath?!’
 See the works cited by Iatridou for many important points on this issue, especially Steele ().
 Examples are from Moretti and Orvieto (). Examples (a,b,d) are literary examples; the
original sources are C. Cassola, La ragazza di Bube; G.T. di Lampedusa, Il Gattopardo; C. Cassola, Una
relazione.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

114 verbal mood

The first example can be considered a subtype of the imperative sentence mood
in which the subjunctive is used in place of the imperative form. This so-called
suppletive imperative results from the choice of third person morphology to
express the lack of familiarity between the speaker and the addressee (informally,
“politeness,” though that term is particularly inappropriate for this example).
Example (114b) has a similar meaning, and can be described as an optative or third
person imperative.
There has been very little research in semantic theory on these root uses
of the subjunctive. Han (1998) explains the choice of subjunctive in suppletive
imperatives by proposing that they have an [irrealis] feature appropriate to the
imperative sentence mood (see Section 3.1.3 for discussion). Grosz (2011, 2014)
develops a detailed analysis of German optatives in which subjunctive mood
plays a critical role. (It is interesting that Grosz’s analysis of optatives ties them
closely to conditionals; see Section 3.3.4 for some more discussion of Grosz’s
ideas.) Portner (1997) proposes that the modal parameters of a root clause can
sometimes be shifted to ones which would trigger the subjunctive. His idea is that
an example like (114d) can be interpreted relative to the same modal parameters
which would be introduced by ‘be surprised that,’ and therefore mean that the
speaker is surprised that she is in the bath. Ambar (2016) discusses root uses of the
Portuguese subjunctive in some detail and relates them to a broader theory based
on the idea that subjunctive marks evaluation and indicative marks assertion.

2.3 Beyond verbal core mood

As has been mentioned above, the most influential semantically-based theories of
verbal mood focus on explaining the selection of indicative and subjunctive forms
by a matrix predicate. In Section 2.2, I hope to have given both a useful presentation
of those theories and some insights into the valuable intuitions which drive them.
But the discussion has not yet touched on a number of areas of research which are
deeply relevant to the study of verbal mood. In this section, I would like to give
some pointers into the body of work on those topics.

2.3.1 Other mood-indicating forms

Thus far, we have focused on the contrast between the core verbal moods, indicative
and subjunctive. Many languages have distinct types of clauses which, in addition
to or instead of these, seem to meet the criteria for being considered verbal moods.
The most obvious and well-studied examples of such forms are infinitives and
dependent modals.

... Infinitives The study of infinitives is closely connected to the analysis of

verbal mood. In many languages, we find infinitives in contexts which are very

 The use of the term “suppletive” here is not meant to imply that tenga is to be analyzed as a form

based in one paradigm (the subjunctive) replacing a form within another (the imperative).

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

beyond verbal core mood 115

similar to those in which the subjunctive is used. We see this point in English with
the fact that infinitives are typically used in the complement of desiderative and
directive verbs, as in (115), and can also be used to express meanings similar to root
subjunctives, (116):

(115) a. Ben wants for John to be successful.

b. She ordered me to leave.
(116) Oh, to some day live in Paris!

Not all infinitives in English are associated with subjunctive-like contexts. For
example, ‘know’ and ‘believe’ do not typically have an affinity for the subjunctive,
yet we use infinitives in examples like the following:

(117) Ben knows/believes John to be hiding in the basement.

There is a crucial difference in temporal interpretation between the infinitives in

(115) and (117). The former examples—that is, the ones which appear in subjunctive-
like contexts—are future-oriented, while the latter ones describe a situation which
overlaps temporally with the time of Ben’s knowing/believing. Bresnan (1972),
Stowell (1982), Pesetsky (1992), and Portner (1992) argue that there is a semantic-
ally distinct subclass of future-oriented infinitives, sometimes known as “for-
infinitives” due to the fact that the presence of the complementizer for, as in (115a),
signals future orientation. Portner (1997) discusses for-infinitives from the point of
view of the theory of mood.
In languages which make robust use of the subjunctive, infinitives and sub-
junctives occur in similar contexts. In the complement of verbs like ‘want’ and
‘order,’ the infinitive is typically required when the subject of the embedded clause
is semantically bound by (and hence, in an intuitive sense, coreferential with) the
matrix subject (from Quer 2009, p.1789).

(118) a. No vull que saludis absolutament ningú. (Catalan)

neg want that say.hello.subj absolutely anyone
‘I don’t want you to say hello to anyone at all.’
b. No vull saludar absolutament ningú.
neg want say.hello.inf absolutely anyone
‘I don’t want to say hello to anyone at all.’

See Ruwet (1984), Kempchinsky (1985, 2009), Picallo (1985), Raposo (1986a), Suñer
(1986), Farkas (1992a), Tsoulas (1996), Avrutin and Babyonyshev (1997), Quer
(1998), Dobrovie-Sorin (2001), Schlenker (2005, 2011), Costantini (2006), Ambar
(2016), and Zu (2016) for some of the many analyses of this obviation effect.
Farkas (1992a) proposes that the infinitive represents both “world-dependency”
and “subject-dependency,” while the subjunctive marks only world-dependency,
and the former blocks the latter when the subject is responsible for the action
described by the clause (the “canonical control relation”). Schlenker, Costantini,

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

116 verbal mood

and Zu relate the obviation effect to a preference for a de se reading to be expressed

in an unambiguous way if possible. (Zu also takes into account Farkas’s canonical
control relation.) Pearson (2013) is an important recent study of de se interpretation
which incorporates a theory of control of infinitival subjects.

The temporal interpretation of subjunctives and infinitives. The close con-

nection between infinitives and subjunctives highlights some important features
of the temporal interpretation of subjunctive clauses. As illustrated in (115), the
complements of desiderative and directive predicates are future-oriented. When
these predicates take a subjunctive complement, there are severe restrictions on the
tense morphology in the subordinate clause. For example, the following contrast
shows that the past subjunctive is not allowed under matrix ‘want’ in the present
tense (from Quer 1998, p.33):

(119) a. Vull que acabi la tesi. (Catalan)

want.pres.1sg that finish.subj.pres.3sg the dissertation
‘I want her/him to finish the dissertation.’
b. ∗ Vull que acabés la tesi.
want.pres.1sg that finish.subj.pst.3sg the dissertation
‘I want her/him to finish the dissertation.’

Facts like this have led to the idea that the tense of subjunctives is dependent on
that of the matrix clause. This dependency can be understood in different ways.
The subjunctive tense is sometimes analyzed as anaphoric (Picallo 1984, 1985; Luján
1979, 1980; Pica 1984, among others) or as a semantically vacuous copy of the matrix
tense (Progovac 1993; von Stechow 1995). Such analyses minimize the distinction
in temporal interpretation between infinitives and subjunctives, and therefore lead
naturally to the view that the two types of clauses differ mainly in whether the
subject is controlled by the matrix subject or not (as seen above in (118)). Related
ideas about the temporal dependency of subjunctives and infinitives are pursued
by Ambar (2016) and Wiltschko (2016).
While these intuitions are important and probably provide essential clues to
the full analysis of verbal mood, their applicability is limited by the fact that
subjunctives in adjunct clauses and those triggered by negation are freer in their
temporal interpretation (Raposo 1986b; Suñer and Padilla-Rivera 1987; Suñer 1986;
Quer 1998). Moreover, as pointed out by Portner (2011c), merely identifying the
temporal properties of (selected) subjunctives does not explain why clauses with
those properties occur in precisely the contexts they do. Does the ungrammatical-
ity of (119b), for instance, follow from some incoherence in the meaning which
would be expressed if its complement had an independent past tense, or is it
fundamentally a syntactic fact?
In order to properly analyze the temporal dependency of subjunctives and
their relation to infinitives, it will be necessary to have a better understanding of
the temporal semantics of subjunctives and infinitives than we currently do. The
semantics literature on these topics is quite fragmentary, and so while there are

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

beyond verbal core mood 117

important insights, they are difficult to combine into a single coherent theory. It
may prove useful to point out a few key observations:

1. Von Stechow (1995, 2004) and Giannakidou (2009) propose that the temporal
argument of a subjunctive clause is bound by a λ-operator in C, creating a
phrase which denotes a property of times. We might then assume that verbs
which select for the subjunctive take a temporal property as their semantic
argument, while those which select for indicatives take temporally saturated
propositions. However, this would probably not be correct, as standard
theories of tense treat all embedded clauses as denoting temporal abstracts
(e.g. Ogihara 1996, 2007; Abusch 1997).
2. Portner (1992) points out the following contrast (examples from Portner 1992,
pp.260–1):

(120) a. Jackie remembers reading Jude the Obscure.

b. Jackie remembers that she read Jude the Obscure.

In order for (120a) to be true, Jackie has to be able to place herself in the
moment of reading Jude the Obscure; in contrast (120b) only requires that
she be able to recall a fact. Portner says that (120a) describes Jackie as having
“internal perspective” on the event she remembers, while (120b) describes an
“external perspective.” Higginbotham (2003) proposes that (120a) represents
a de se attitude towards the event of reading, and Schlenker (2011) suggests
infinitival complements are also used to express event-related de se attitudes.
3. Portner (1997) and Abusch (2004) provide explicit semantic theories of the
future-orientation of for-infinitives. Both postulate an operator which intro-
duces futurity in a way that is very different from standard tenses. (In other
respects their analyses are quite different. Portner uses a situation semantics
and introduces futurity within the infinitival clause, while Abusch employs a
more standard tense semantics, and introduces futurity in the matrix VP.)
4. Schlenker (2005) notes that some temporal expressions can receive a de se in-
terpretation, and Brasoveanu (2006) argues that a particular Romanian verbal
mood form (the conditional-optative or Subjunctive B) directly encodes a
temporal de se meaning (Brasoveanu 2006, (1)):

(121) Maria crede cǎ ar fi în pericol. (Romanian)

Mary believes that subj.3sg be in danger
‘Maria believes that she is in danger.’

Putting these points together, it seems plausible to assume that the temporal prop-
erties of temporally-dependent subjunctives and infinitives should be explained
on the basis of the particular varieties of de se meaning which they encode. We
might guess that ‘want’ expresses an attitude which is always de se with respect
to the complement’s time or event argument, and that the only way to produce
a time- or event-based property which can serve as its argument is to use a
subjunctive or infinitival clause. This way of viewing things connects nicely to the

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

118 verbal mood

idea that the obviation contrast between infinitives and subjunctives has to do with
the preference to express subject de se with an infinitive. Against this approach,
though, is the fact that indicative clauses can also have de se meanings, both in their
subject position and their temporal argument. It is thus unclear why indicatives
cannot be used in place of subjunctives and infinitives (under their de se logical
forms) and why subjunctives and infinitives cannot be used in place of indicatives
(in contexts where a de se reading is to be derived).

... Mood-indicating modals Building on the presentation by Palmer

(1990), Portner (1997) proposes that the modal auxiliaries may and would are
“mood-indicators” in sentences like the following (his (42c,d) and fn. 11; (a) is from
Palmer):

(122) a. I pray that God may bless you.

b. It is possible that Sue may win the race.
(123) If you would come, I would be happy.

The clauses in which may occurs in (122) would be subjunctive in many languages.
Moreover, the meanings of the sentences do not seem to involve two separate
modal operators; for example, (122b) does not mean ‘It is possible that it is possible
that Sue wins the race.’ Given these facts, Portner proposes that may in the
subordinate clause is dependent on the modality of the matrix predicate, and is
therefore a mood-indicator within his theory. Similarly, he suggests that the first
would in (123) is a mood-indicator, though he does not give a detailed analysis of
this case.
Other linguists have not pursued Portner’s idea, but several have pointed out the
similarity between examples like (122) and modal concord. Modal concord is the
phenomenon whereby a modal adverb and a modal auxiliary or verb in the same
clause seem to represent a single modal element at logical form (example from
Zeijlstra 2008, (3)):

(124) You may possibly have read my little monograph upon the subject.

Zeijlstra (2008) in fact classifies patterns like (122), in which a modal can be thought
of as in concord with a matrix predicate, as modal concord, although the analysis
he proposes focuses on the clause-mate cases.
There is a small but interesting body of research on modal concord, including
work by Geurts and Huitink (2006), Zeijlstra (2008), Huitink (2012), Grosz (2010),
and Anand and Brasoveanu (2010). Cui (2015) presents a detailed corpus investig-
ation of combinations like (122) in Mandarin and applies rigorous criteria to de-
termine in which cases the embedded modal makes no detectable contribution to

 The term “mood-indicator” suggests that there is an inventory of moods on which languages

draw, and that they may be indicated or marked by various grammatical means. In the terminology
of this book, Portner’s idea might be better expressed by saying that these modals are verbal moods or
components of verbal moods.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

beyond verbal core mood 119

the overall sentence’s meaning. She identifies a significant number of verb–modal

combinations which can be said to be in a concord relation, such as huaiyi …keneng
(‘suspect …might’) in examples like the following (based on Cui 2015, 4.11, 4.26):

(125) a. Lao Li huaiyi xiongshou keneng shi Xiao Wang. (Mandarin)

Lao Li suspect culprit might be Xiao Wang
‘Lao Li suspects that the culprit might be Xiao Wang.’
b. Lao Li pizhun Xiao Wang keyi likai.
Lao Li approve Xiao Wang may leave
‘Lao Li approved Xiao Wang’s optional departure.’

Cui’s theory of these combinations of modal concord builds on Yalcin’s (2007) and
Anand and Hacquard’s (2013) method of shifting parameters (see Section 2.2.2).
Her analysis of epistemic examples like (125a) ends up being similar to Yalcin’s pro-
posal about English might, but her treatment of examples with priority modality,
like (125b), is novel and based on a proposal involving reported speech acts.

2.3.2 The roles of semantics, syntax, and non-grammatical factors

The goal of this book is to introduce and evaluate semantic theories of mood, and
we have accordingly focused on proposals which use the tools of semantic theory.
As we have seen, the best such theories are quite successful in explaining certain
properties of verbal mood systems, in particular the distribution of indicative and
subjunctive in complement clauses. In this section, however, I would like to briefly
consider to what extent we should expect to fully explain the distribution and
linguistically-relevant properties of verbal mood forms solely within a semantic
analysis. Do other components of linguistic theory, such as syntax and morphology,
play independent roles? And beyond grammar, what are the contributions of usage-
based, social, psychological, and historical factors?
We should certainly expect syntax to make an important contribution to an
overall theory of verbal mood. In order to explain the patterns of indicative and
subjunctive in a given language, we will have to take into account what logical
forms can be syntactically derived; we would not expect a subjunctive relative in
a nominal phrase which obligatorily takes widest scope, for example. Similarly,
morphology will certainly play an important role, since a language may simply not
allow for a choice of verbal mood forms under certain conditions. The relevance
of these components of the grammatical system follows from the architectural
relationship between morphosyntax and semantics more generally.
The issues are less clear when we consider the role of non-grammatical factors.
As mentioned in Section 2.2.2, Portner (1997) proposes that verbal mood systems
are similar to gender systems, and that as a result we should expect a certain
amount of idiosyncrasy in the distribution of verbal mood forms. He suggests
that semantic properties are the basis for mood choice (as biological sex might be
for gender systems), but that in specific cases the choice of verbal mood might
be made on the basis of similarity to a prototype; for example, if ‘want’ is the
prototypical subjunctive-selector, another verb might also trigger subjunctive if it

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

120 verbal mood

is similar to ‘want’, even if it does not have the key property which makes ‘want’
itself such a strong trigger for subjunctive. A theory which allows mood choice
to be determined on this basis will clearly incorporate a strong psychological
component.
It is also important to address the role of sociolinguistic factors. It is easily
imaginable that a particular verbal form could take on associations with a salient
aspect of social identity, for example a form which is understood as indicating that a
speaker is highly educated. Based on this, speakers might use it to index that social
identity, resulting in a disconnect from its semantic basis. Usage-based factors
could have a similar effect. We can imagine a particular pairing of matrix verb and
mood form coming to be so highly associated in usage that the form is used in the
presence of that verb, even if the verb is used in a meaning or structural position
which would not be expected to trigger the mood form by semantic criteria.
Much of the discussion points to the need for semanticists to be attentive to
historical developments as they study verbal mood. As a language’s mood system
changes over time, there may be points at which it cannot be fully understood
in terms of any single, coherent analysis. There may be stages in the historical
development at which the semantics of mood is in flux, with two or more systems
in play. The morphological and syntactic systems might likewise not be entirely
complete or consistent. The effects of psychological, social, and usage-based factors
may lead to further complexity. In this type of situation, the essential goal from a
semanticist’s point of view is to understand all of the complexities sufficiently well
to continue the pursuit of a better understanding of the semantic factors which, it
seems, are the most powerful in explaining the properties of verbal mood in the
long run.

i i

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

i i

3
Sentence mood

In syntax, speech act theory, dynamic semantics, and other areas of linguistics and
philosophy which study how functions of language like asserting a fact or asking a
question are achieved using language, one idea often comes up: that one can read
off from the form of a sentence which of these important functions it has, or at
least gain information which will probably tell, in combination with other things
you know, which it has. The sentence moods of a language are the particular forms
which carry information of this kind. Our goal in this chapter is to develop a better
understanding of how sentence moods fit into theoretical conceptions of the basic
functions of language and the architecture of grammatical theory.
This chapter will begin in Section 3.1 with an examination of some fundamental
topics, both foundational and empirical, related to sentence mood. We will try to
situate the phenomenon of sentence mood properly within syntax, semantics, and
pragmatics, and we will especially try to understand the relation between sentence
mood and the concepts of clause type and force. Then we will survey some work
within the two main theoretical approaches to sentence mood, those embedded
within speech act theory (Section 3.2) and the dynamic approach to meaning
(Section 3.3). These surveys form the core of the chapter, as it is there that I will
explain and evaluate some of the important recent research on particular sentence
moods. Finally, in Section 3.4 I will take the discussion into a broader perspective
by asking what insights each major theory can offer concerning the overall systems
of clause types and sentence mood in natural language. As we will see, although
linguists have yet to develop a grand theory of sentence mood which explains all
relevant syntactic, semantic, pragmatic, and typological facts, the recent literature
does offer a great many insights that should be maintained and developed as the
field works towards such a theory.

3.1 Sentence mood, clause type, and sentential force

The term “mood” is not always employed in the same way in discussions of the
relation between the syntactic/semantic forms of sentences and their conversa-
tional functions. It is sometimes used as a semantic or pragmatic category, for
example by Wilson and Sperber (1988). In this sense, we often hear of some
particular grammatical feature being a “mood-indicator” or “mood-indicating
device,” and in such cases it is not clear whether mood itself is understood to be
a semantic category or an abstract grammatical category which groups together
sentences which do not share any specific element in common. “Mood” is also used

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

122 sentence mood

to pick out a particular paradigm of linguistic form; this use is more in parallel with
that of “verbal mood,” and because a broad goal of this book is to better understand
how different notions of mood are related, we will use it in this way here. This
usage fits with the working definition of mood from Chapter 1, Section 1.1.2: Mood
is an aspect of linguistic form which indicates how a proposition is used in the
expression of modal meaning. Building on this, I would like to define sentence
mood as follows:

(1) Sentence mood is an aspect of linguistic form conventionally linked to the

fundamental conversational functions within semantic/pragmatic theory.

Sometimes we will use the term “sentence mood” in a slightly sloppy way to refer
to a set of sentences which share the same sentence mood in the sense of (1).
The aspects of linguistic form which define sentence mood could be as simple as
a single piece of morphology or as complex as a combination of several construc-
tional patterns of lexical choice and word order. This definition leaves open several
crucial matters at the very foundation of the concept of sentence mood:

1. It does not say what “the fundamental conversational functions” are. We

cannot decide in advance what those conversational functions are, as different
theories will identify them differently. These functions are given a label,
“sentential forces,” just below.
2. It leaves open how to identify the linguistic forms which get linked to
the functions; different approaches to morphology and syntax can motivate
different ideas about sentence mood; conversely, different assumptions about
how to classify conversational functions into fundamental classes can motiv-
ate different models of morphology and syntax.
3. It presupposes that there is a set of linguistic forms conventionally linked to
these functions, but it leaves open how this conventional link between form
and function is established. We will consider several approaches to the issue
of how this form–function relation is established in Section 3.1.3.

As we will see, different theories settle these points in different ways.

It will also be helpful to define two other terms, clause type and sentential
force.

(2) Clause types are grammatically defined classes of sentences which corres-
pond closely with sentence moods.

 Note that the use of the term “conventional” is not meant to imply that the relation is a matter of

social convention. Davidson () and Harnish (), among others, critique the idea that sentence
mood is a matter of convention. It could be a matter of literal meaning produced by innate brain circuits.
Nor does it imply that any social convention which associates a form with a function would count as
a sentence mood; a code which uses Are you thirsty? to mean ‘I am going outside to smoke a cigarette’
does not establish a new sentence mood.

i i

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

i i

sentence mood, clause type, and sentential force 123

The difference between the concepts of clause type and sentence mood is that the
former are primarily identified in terms of grammatical form, while the latter are
primarily identified in terms of meaning. Traditional grammatical descriptions
make use of such clause types as declarative, interrogative, and imperative. In
many languages, sets of sentences can be associated with these types strictly in
terms of their morphosyntactic properties; for example, the clause types of Korean
can be defined in terms of sets of sentence-final particles, while the clause types of
English can be defined in terms of matters like the form of the verb, the position
of the verb, the presence or absence of wh-phrases in clause-initial position, and
the choice of complementizers. Such differences are thought of as central matters
for syntactic theory to explain.
A good example of the difference between sentence mood and clause type
concerns subordinate clauses. The embedded clause in (3) is an interrogative in
terms of its clause type, but may not have a sentence mood at all, since it does not
perform the function of asking a question.

(3) Everybody knows who won the race.

It can be confusing that clause types often go by the same names as the sentence
moods, including the names ‘declarative,’ ‘interrogative,’ and ‘imperative.’ Many
discussions use the term sentence types to think about the systems that I am
describing as clause types and sentence moods (e.g. Sadock and Zwicky 1985; Reis
1999). The notion of sentence type has the syntactic orientation of clause type but
it is typically not applied to subordinate clauses on the grounds that they do not
have an independent conversational function. In a terminology like this, ‘sentence
type’ is very similar to ‘sentence mood’ as used here; specifically, a sentence type is
a set of sentences of the same clause type which share the same sentence mood.
Next we define the concept of sentential force:

(4) Sentential forces are the fundamental conversational functions with

which sentence moods are associated.

There are many different views on the nature of sentential forces. They might be
many and specific or few and general. According to the literal meaning hypothesis
(as discussed in Section 1.4.2), the sentential forces are the illocutionary forces of
classical speech act theory. According to dynamic semantics and pragmatics, there
might be as few as two or three sentential forces (distinct ones for declarative,
interrogative, and possibly also imperative sentences; see Section 1.4.1). The version
of speech act theory which invokes the notion of illocutionary point puts the
number at five (Searle 1975b; Searle and Vanderveken 1985; Vanderveken 1990; see
Sections 1.4.2 and 3.2.2). In our reading of some scholars, we will have to determine

 One could propose that the embedded clause in () is associated with the function of asking, but

that the association does not lead to its actually asking a question in this case. (Using the terminology
about to be introduced we might say that it has the sentential force of asking which is somehow
neutralized.) But the very fact that we can ask whether it has a sentence mood, while still being clear
that it is an interrogative, shows the difference between the two concepts.

i i

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

i i

124 sentence mood

Table . Terminology for sentence moods

Our terms Alternative terminology

Forms associated with functions sentence mood sentence type

Those functions sentential force mood (or sentence mood)
Related clauses containing those forms clause type (none)

the best way to align their understanding of conversational function with our
concept of sentential force.
Our notion of sentential force covers the semantic use of the term “mood” in
theories which might say that clause type is a “mood-indicating device.” Table 3.1
lays out correspondences between the terminology used in this book and other
ways scholars talk about similar categories. Though the alternative terms can be
used in a clear and precise way, I believe it is important to have a name for the
aspects of form which are linked to fundamental conversational functions (= the
sentence moods) as distinct from the syntactically defined types which incorporate
those aspects of form (= clause types). This allows us to talk, for instance, about the
subordinate clause in (3) being of the interrogative clause type and having many of
the grammatical features which comprise the interrogative sentence mood; from
such a vantage point, we can then ask in a clear way how it is different from root
examples of the same clause type and why these differences lead to it not having a
sentential force.
Theories of sentence mood focus on a number of basic questions:

1. What grammatical properties constitute sentence moods, and how do they

relate to the syntax of clause types?
2. What are sentential forces?
3. How are sentential forces linked to sentence moods?

We will investigate how they are answered by approaches based on speech act
theory in Section 3.2 and by theories which follow the dynamic approach in Section
3.3. In the remainder of this section, we will discuss some important background
concerning clause types, sentential forces, and sentence mood. First we will de-
scribe the basic properties of clause type systems (Section 3.1.1), then review the
intuitive functions of the most widely discussed sentential forces (Section 3.1.2),
and finally categorize at a general level the approaches to sentence mood within
syntax (Section 3.1.3).

3.1.1 Clause types as grammatical categories

The concepts of clause type and sentence type are used to take a syntactic, or more
generally a structural, approach to understanding the relation between grammar

 This determination is sometimes difficult. For example, Gazdar () and Kaufmann ()

assume an austere version of dynamic pragmatics while also accepting the notion of illocutionary force
from classical speech act theory. See Section .. for discussion.

i i

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

i i

sentence mood, clause type, and sentential force 125

and sentential force. As I use the terms, clause types are categories defined properly
within syntax which are observed to have a close relation to sentential forces, but
they are distinct from sentence moods because they do not have to align precisely
with sentential forces. We have already seen in the opening part of this section
that an embedded interrogative clause may be seen as lacking sentence mood,
even though it is a member of the interrogative clause type. Similarly, there is a
case to be made that sentence moods sometimes do not have simple definitions in
terms of syntactic properties, even if we restrict ourselves to unembedded clauses.
For example, Zanuttini and Portner (2003) argue that there is no natural syntactic
property which groups together the following three subtypes of exclamatives in
Paduan (data from Zanuttini and Portner 2003):

(5) a. Che roba che l magna! (Paduan)

what stuff that he eat
‘The things he eats!’
b. Quanto è alto!
how much is tall
‘How tall s/he is!’
c. No ga-lo magnà tuto!
neg has-sg.cl eaten everything
‘He ate everything!’

The wh-phrase che roba in (5a) cannot occur in interrogatives, and it occupies a
different position in the clausal structure than the quanto in (5b),which can occur
in an interrogative. The exclamative in (5c) does not seem to have a wh-phrase at
all, and its verb is in a higher position than the subjects of (5a–b). If Portner and
Zanuttini are correct that sentences showing exclamative sentence mood in Paduan
do not share any single syntactic property, it might still be possible to group them
into a single clause type with a disjunctive definition which references the two types
of wh-phrase and the position of the verb, but this ersatz clause type would not be
a natural object for syntactic analysis. Nevertheless, because they all share the same
sentential force, they would constitute a natural sentence mood, and be a proper
subject for study at the syntax/semantics/pragmatics interface.

The traditional conception of sentence types. Sentence types are syntactically

defined classes of sentences which are of interest because they are associated in a
regular way with specific conversational functions. For example, (5a) can be seen as
a representative of a sentence type defined by the association between the pattern
of structure in (6a) and the function (6b):

(6) a. Special wh-phrase in initial position (che+N), followed by the comple-

mentizer che; verb in post-subject position
b. Exclamation about a property of an object or group of objects

The concept of sentence type is most often used in linguistics research where the
primary goal is to provide systematic descriptions of the variety of sentences in

i i

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

i i

126 sentence mood

a language and to identify the specific conversational functions which a given

sentence could be associated with. The descriptions are typically pattern-oriented,
based on surface properties, and non-generative. Research of this kind represents
what I call the “traditional conception of sentence types.”
We can get a sense of the range of research which builds on the traditional
conception of sentence types by considering two important (and otherwise very
different) works: Sadock and Zwicky (1985) and Altmann (1987). Sadock and
Zwicky define the concept of sentence types with the goal of creating a single con-
ceptual framework which can be applied across a range of typologically different
languages. Their work makes a contribution by emphasizing the fact that sentence
types form a closed system, in the following sense:

(1) “The sentence types of a language form a system, in at least two senses:
(a) there are sets of corresponding sentences, the members of which differ
only in belonging to different types, and . . .
(b) the types are mutually exclusive, no sentence being simultaneously of
two different types.”
(2) “Sentence types show certain characteristic forms across languages.”

This definition of a system of sentence types implies that certain syntax–pragmatics

correspondences should not be understood as sentence types. For example, the
cleft structure, while it is syntactically definable and has a characteristic discourse
function, is not a sentence type because it cross-cuts the classes of declaratives and
interrogatives.
In a comprehensive study of sentence types in German, Altmann (1987) gives
a fine-grained and precise description of many specific pairings between gram-
matical patterns and conversational use. His work shows how much valuable
knowledge about a language’s grammar can be uncovered and systematized using
the traditional concept of sentence type. But as shown by Reis (1999), his work
also makes apparent the conception’s weaknesses. As Reis argues, because sentence
types are understood as specific pairings of form and function, it forces us to
subdivide some of the most intuitively important categories. For example, the three
kinds of exclamatives seen in (5) would not count as a sentence type, beause they do
not share a single grammatical form. Similarly, Reis argues that Altmann’s theory
does not allow for a single broad category of imperative sentences in German. In
addition, Reis highlights the tendency of work which uses the concept of sentence
types to assume a view of grammatical description based on surface patterns. These
“form types” (as Reis calls them) are defined as collective patterns, with no indi-
vidual compositional contribution made by the component features. For example,
when we talk about the form type (6a) and its function (6b), we do not assign any
particular significance to the presence of che+N or the position of the verb.
These two problems for the traditional conception of sentence types are quite
critical, and they lead me to avoid using the term in this work. However, it is likely

 Other useful discussions are found in Palmer (), Allan (), and König and Siemund ().

i i

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

i i

sentence mood, clause type, and sentential force 127

that they are really due not to the concept of sentence types itself, but rather to
the simple pattern-based method of defining form types. As we will see in Section
3.1.3, a more flexible pattern-based syntactic theory like Construction Grammar
may be able to solve these problems. But even if this is the case, the concept
of sentence types still suffers from the fact that it does not allow us to separate
clearly when we are talking about syntactic categories and when we are talking
about semantic/pragmatic categories. It forces us to treat all syntactic categories as
semantic categories, and vice versa, and so it may be misleading in cases when they
do not align in a simple way.

Basic properties of clause type systems. Although it is important to distinguish

the concepts of sentence mood and clause type, it is impossible to develop a theory
of sentence moods which does not pay close attention to the syntax of individual
clause types and to the structure of clause type systems. By studying the syntax of
a clause type or a set of clause types whose sentences all have the same sentence
mood, we learn a great deal about semantics and ultimately the discourse function
of that sentence mood. For example, it is obvious that clauses introduced by wh-
words can have interrogative or exclamative sentence mood, but not declarative
sentence mood. In order to explain this fact, we will need to understand the kinds of
contribution which wh-words can make to the meaning of a clause and the relation
between those meanings and sentential force.
In the remainder of this section, I will outline some of the most important prop-
erties of clause type systems. Our perspective here will be that of identifying some
important properties of clause type systems as a whole, rather than studying each
clause type in detail. We will discuss individual clause types in Sections 3.2–3.3, as
part of a detailed survey of semantic theories of individual sentence moods. We
will then return to a system-oriented view in Section 3.4.

Basic and minor clause types. Linguists often assume that the declarative, in-
terrogative, and imperative clause types can be recognized in all languages, and
that this is so because the sentential forces which these types realize are somehow
fundamental to the communicative function of human language. Sadock and
Zwicky (1985) refer to these three as the basic types in recognition of both their
ubiquity and function. In addition, linguists commonly recognize minor types,
including exclamatives, optatives, and prohibitives.
Various traditions within linguistics come with different, and often unexamined,
assumptions concerning the right way to identify types and describe the relations
among them. We can understand the issues involved by examining some of the
types which may be related to imperatives: prohibitives, hortatives/exhortatives,
promissives, jussives, and optatives.
Many linguists would describe the English examples in (7a) and (7b) as positive
and negative imperatives, respectively:

(7) a. Open the window!

b. Don’t open the window!

i i

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

i i

128 sentence mood

Calling both of these “imperative” implies that they are members of a single type,
differing only in the presence or absence of sentential negation. In contrast, in
her discussion of realis marking in Teiwa, Klamer (2012) treats the sentences
in (8) as exemplifying two different types. She calls (8a) an imperative, but (8b)
a prohibitive (her (2b), (19)).

(8) a. yo, iqaan ba tewar

yes dark seq walk
‘Yes [it’s getting] dark so go!’
b. wat wrer gaxai!
coconut climb do.not
‘Don’t climb the coconut [tree]!’

Klamer’s perspective on (8b) is motivated by the fact that prohibitives are marked
by the negative verb gaxai rather than a general-purpose negative marker. Van der
Auwera and Devos (2012) label the English (7b) a prohibitive, and while they are
clear that this term is equivalent to “negative imperative,” they are not explicit
whether positive imperatives and prohibitives are separate types, with imperative
a higher-level grouping of similar types, or whether the more general imperative
is the sentence type, with positive and negative imperatives being variants that do
not count as types on their own.
In this way, discussions of clause types (and sentence types) are not often
particularly explicit on how basic and minor types relate to one another within
the overall system of types. They could be members of the same type or members
of distinct (but pragmatically similar) types. The distinction matters as we try
to analyze the sentence moods, because (for example) if one decides to treat
prohibitives as a separate type, there will be the possibility—and in practice, the
tendency—to treat their function, “prohibition,” as a sentential force distinct from
that of imperatives. Thus, in the context of developing and evaluating theories of
sentence mood, the decision about classification will have consequences for our
theories of semantics and pragmatics.
Similar points can be made about the other clause types related to imperatives.
Exhortatives and promissives are illustrated with data from Korean in (9), from
Zanuttini et al. (2012, (2)):

(9) a. Cemsim-ul sa-la. (imperative)

lunch-acc buy-imp
‘Buy lunch!’
b. Cemsim-ul sa-ma. (promissive)
lunch-acc buy-prm
‘I will buy lunch.’
c. Cemsim-ul sa-ca. (exhortative)
lunch-acc buy-exh
‘Let’s buy lunch.’

The imperative type is marked in this example by the sentence-final particle -la.
The promissive type, marked by -ma, is used to make a promise; the exhortative,

i i

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

i i

sentence mood, clause type, and sentential force 129

marked by -ca, is used to urge a joint action by the speaker and addressee. The
differences between telling someone to do something, promising to do it, and
urging to do it together might suggest that these three comprise distinct clause
types. However, according to Portner (2004) and Zanuttini et al. (2012), these three
types actually share a common function: they are all used to create a commitment
on the part of one or more discourse participants to take a certain action. For
this reason they can be thought of as members of a broader class, jussives
(e.g. Pak et al. 2008a,b). The idea that they belong to a common, general type
implies that they differ only in grammatical features not directly tied to sentential
force. (For Pak et al., this feature is grammatical person. )
Optative sentences like (10) are used to express a wish of the speaker (from
Grosz 2014, (2)):

(10) If I’d only listened to my parents!

As with promissives and exhortatives, the status of optatives as a clause type is

not clear. The normal use of an optative implies that the speaker prefers one
alternative to another, and in this way it is similar to an imperative. On this basis
it is reasonable to think of some optatives as third person imperatives (or third
person members of a jussive type including imperatives), as suggested by König
and Siemund (2007). Grosz (2011, 2014), however, relates optatives to exclamatives
rather than imperatives and Biezma (2011) treats them as conditional antecedents
with elided (declarative) consequents; thus, it remains unclear whether optatives
and imperatives share any property as fundamental as their sentential force.
The other basic clause types have subtypes as well. Among interrogatives,
we can distinguish yes/no (or polar), information (or wh-), alternative, biased,
confirmative, echo, and rhetorical interrogatives. There is little question that all
of these show the same interrogative clause type. On our view here, they belong to
a common clause type because they can be described as forming a class in syntactic
terms which is, as a whole, associated with one sentential force. By these criteria,
certain rhetorical questions might have a kind of split status: interrogative in clause
type but not sentence mood. For example, an utterance of (11) is typically associated
with the suggestion that no answer exists (Obenauer 2004, (37)):

(11) Who can you trust, nowadays?

On the assumption that (11) does not have the sentential force of asking, it would
not have the interrogative sentence mood. Nevertheless, it would be a member of
the interrogative clause type. The interrogative clause type, in other words, could
not be seen as an unambiguous exponent of interrogative sentence mood, though

 Note that there is an important distinction to be made between promissives and first person

imperatives (cf. Aikhenvald ; König and Siemund ), in that promissives lead to the speaker’s
commitment to do something, while first person imperatives aim to get the addressee to allow or make
the speaker to do something.

i i

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

i i

130 sentence mood

we would of course want to explain why interrogative clause type almost always
leads to the sentential force of asking a question.
Linguists do not normally talk about subtypes of declarative clauses from
the perspective of sentence mood or clause type. They are used to considering
differences in function among declaratives as arising from independent factors like
negation, information structure, modality, and evidentiality. Sadock and Zwicky
(1985, p. 170) briefly mention the non-affirmative mode of Blackfoot as likely
representing a separate dubitative sentence type which is used both for expressing
uncertainty and asking yes/no questions. They treat it as a type separate from that
of other declarative-like sentences, the affirmative.
Gunlogson (2001) discusses the examples in (12a–b), distinguished by rising vs.
falling intonation, as members of the declarative clause type (her (1)–(2)):

(12) a. It’s raining?

b. It’s raining.
c. Is it raining?

In her analysis, clause type and intonation contribute compositionally to the

discourse function of the sentence, and so we cannot say whether (12a–b) have
the same sentence mood without determining which levels of function count as
fundamental for the purposes of identifying sentential force. It is possible to un-
derstand her as suggesting that (12a) and (12b) each have two sentence moods: (12a)
would have the declarative sentence mood (marked by clause type) and “addressee-
oriented” sentence mood (marked by rising intonation); (12b) would have the
declarative sentence mood and the “speaker-oriented” sentence mood (marked
by falling intonation). Such a classification would be quite novel, and though we
cannot confidently attribute this perspective to Gunlogson herself, it is worthwhile
to bring it out as an innovative possibility. We will discuss Gunlogson’s ideas in
more detail in Section 3.3.1.
There is one minor clause type with a fairly uncertain status, exclamatives.

(13) a. What a nice guy he is!

b. How very tall you are!
c. He is such a nice guy!
d. You are so very tall!

Unlike the other minor types discussed above, the sentential force of exclamatives
is not in any obvious way related to that of a major type. As a result, exclam-
atives are normally considered as forming their own type. In terms of syntax,
they show obvious connections to interrogatives (and sometimes declaratives)
across languages. For example, (13a–b) are similar to interrogatives in including

 It is also possible that rhetorical questions do have the sentential force of asking (which is somehow

neutralized), and so have interrogative sentence mood. The interpretation of their status given in the
text is raised at this point only for illustration.
 Gunlogson uses the term “sentence type,” but her usage is close enough to our “clause type.”

i i

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

i i

sentence mood, clause type, and sentential force 131

a wh-movement structure (also (5a–b) above). In terms of semantics, some have

argued that exclamatives have a type of meaning closely related to that of in-
terrogatives as well, with the difference between interrogatives and exclamatives
residing in pragmatic factors (Zanuttini and Portner 2003). Of course, there are
syntactic differences as well, such as the particular forms of the wh-phrases in (13a)
and (5a). The uncertain status of exclamatives deepens further if one also considers
the sentence in (13c) to be an exclamative; the only visible aspect of syntax which
indicates that this sentence is exclamative is the correspondence between what an
X and such an X. Overall, exclamatives seem to represent a case where a clearly
distinct sentential force goes along with a messy characterization in terms of clause
type. This makes the type quite important for evaluating theories of sentence mood
at the syntax/semantics interface.

Marking of clause types. In discussions of clause type, sentence type, and sen-
tence mood, linguists often talk about some grammatical element or combination
of grammatical features as “marking” a sentence as belonging to a particular
category. The idea behind this way of speaking seems to be that the range of possible
types (or moods) is known in advance, and the task of the speaker is to indicate—
and the task of the hearer to determine—the type of each sentence uttered. As we
have seen, I think this way of thinking runs together a grammatical issue, what
the forms of the clause types are, with an interface issue, how the relationship is
established between sentences with a particular sentence mood and their sentential
force. Still, setting aside the nature of the relation between clause type and sentence
mood for discussion later (Section 3.1.3), it is useful to describe some patterns of
clausal syntax associated with sentence mood. In other words, we will review the
most common ways in which clause types are marked across languages.
I would like to make a distinction between clause types which are explicitly
marked (in a particular language) and ones which are inexplicitly marked. In not
very precise terms, we can say that a clause type is explicitly marked when there is
a grammatical feature or set of features which unambiguously identifies the type. It
is inexplicitly marked when there is no such form. As an example, we can point to
the uses of how and how very seen in (14)–(15). Clauses of the form (14a) cannot be
anything but an exclamative, as we see in their root and embedded occurrences
in (14b)–(14d). How very explicitly marks an English clause as exclamative. In
contrast, we might think of the form in (15a) as inexplicitly marked, since it can
serve as an exclamative or as an embedded interrogative.

(14) a. how very tall she is

b. How very tall she is!
c. I’m amazed how very tall she is.
d. ∗ I wonder how very tall she is.

(15) a. how tall she is

b. How tall she is!
c. I’m amazed how tall she is.
d. I wonder how tall she is.

i i

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

i i

132 sentence mood

The contrast between explicit and inexplicit marking is of course theory-bound,

since one might argue that (15a) is structurally ambiguous and that the type is
always explicitly marked (though the marking may be covert, in the sense of not
being reflected in the phonological output). Cheng (1991), for example, proposes
the Clause Typing Hypothesis, which states that all clauses must be syntactically
typed. Be that as it may, we need the concept of inexplicit marking so as to be
able to talk about cases where there is an issue of whether the types are marked
explicitly or not.
Clause types can be explicitly marked by grammatical structure, by the morpho-
logical form of the verb, and by independent particles. I will briefly illustrate each
of these possibilities. When we say that clause type can be marked by grammatical
structure, this is meant to encompass differences in word order, the absence of
elements which are normally obligatory, or the presence of elements which are
normally absent. For example, English interrogatives require either the fronting
of the auxiliary (in most root questions) or an interrogative word of some sort.
In (16a) the auxiliary is at a position before the subject; in the subordinate clause of
(16b) there is a wh-word in clause-initial position, and in the subordinate clause
of (16c) there is an interrogative complementizer:

(16) a. Is it raining?
b. I wonder who that is.
c. I wonder if/whether it is raining.

Looking back at (15a), it is reasonable to say that, as an interrogative, its type is

marked by the presence of a wh-word, in the same way as (16b). But of course,
given the ambiguity of (15a), this would be a case of inexplicit marking. In order
to maintain the idea that all clause marking is explicit, one would have to claim
that the wh-element does not mark the type on its own; rather it would be marked
by some other grammatical property (such as a particular syntactic structure or
abstract element), possibly in combination with the wh-element. It is for reasons
like this that most linguists who aim to develop a syntactic and semantic theory
of clause types would rather think of this case as one where the type is marked
by grammatical structure, and would call the wh-word a “marker” only in an
informal sense.
Clause type can be marked by the morphological form of the verb in various
ways. The verb may have a special affix, as in the Panyjima imperative (Dench 1991,
cited in Aikhenvald 2010):

(17) minyma panti-ma

still sit-imp
‘Sit still!’

 In fact, her claim is stronger than merely that all clauses are explicitly marked. The hypothesis states

that all clauses are marked by S-structure, implying that marking is always accomplished by syntactic
means.

i i

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

i i

sentence mood, clause type, and sentential force 133

The verb might employ a particular verbal mood form. Here the terminology gets
a little bit tricky. In (18a), we see that a basic imperative in Italian uses a reduced
form of the verb (compared to root declaratives and interrogatives). This canonical
form is used when the subject is second person singular familiar and the clause is
non-negative. In other cases, the verb takes a different form. When negative, it is
the infinitive; when the subject is plural, it is indicative; and when the subject refers
to an addressee towards whom the speaker marks politeness, it is subjunctive.

(18) a. Siediti!
Sit.imp-you
‘Sit!’
b. Non sedersi!
neg sit.inf-self
‘Don’t sit!’
c. Sedetevi!
sit.indic-you.pl
‘Sit!’ (plural)
d. Si sieda!
self sit.subj
‘Have a seat!’ (polite)

Once again, the verb forms in (18b)–(18d) are not in themselves explicit markers;
infinitive, indicative, and subjunctive verbs have other uses. If we believe in explicit
marking of clause type, the marker must be a feature of the abstract syntactic
representation. In many cases, there is good evidence that they do have a distinct
syntax from other uses of infinitives and subjunctives. For example, the verb
precedes the object clitic in (18b) and (18c), whereas infinitives and indicatives
would normally follow the clitic. But in (18d), there is no direct evidence for a
syntax distinct from other subjunctives. If we wish to maintain the idea that (18d)
is explicitly marked as an imperative, we would need to seek indirect arguments or
just take it on faith. The alternatives are to consider it to be inexplicitly marked or
to decline to categorize it as imperative clause type at all.
As mentioned above, Korean marks clause types through the use of sentence-
final particles. The following illustrate the simplest forms of the declarative, inter-
rogative, and imperative types (from Pak 2008).

(19) a. Cemsim-ul mek-ess-ta

lunch-acc eat-past-decl
‘I ate lunch.’
b. Cemsim-ul mek-ess-ni
lunch-acc eat-past-int
‘Did you eat lunch?’
c. Cemsim-ul mek-e-la
lunch-acc eat-imp
‘Eat lunch!’

i i

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

i i

134 sentence mood

The particles in (19) are unambiguous and occur in both root and embedded
clauses. They can be seen as fully explicit markers of clause type. But other
sentence-final particles which express gradations of formality and speech style
are not in themselves explicit markers of clause type. For example, the marker of
intimate style -e is compatible with both declarative and interrogative functions in
root clauses.

(20) Cemsim-ul mek-ess-e

lunch-acc eat-past-intimate
‘I ate lunch.’/‘Did you eat lunch?’

We might categorize -e as an inexplicit marker of clause type, or we might assume

there are abstract (explicit) markers of type which determine whether (20) is
declarative or interrogative.
Korean illustrates another point which is important when describing how
languages mark their clause types. Although across languages clause types are
marked in a variety of ways, there are patterns to the combinations of ways
which a particular language will employ. For example, Korean is consistent in
using the sentence-final particles to mark type; it does not use particles for some
types and word order for others. This fact suggests that the differences among
types can be condensed into the choice of a single morpheme. But in contrast,
many languages will use grammatical properties such as word order to mark the
distinction between declaratives and interrogatives, but a form of the verb (perhaps
together with word order, as in Italian (18a)) for imperatives. This fact suggests
that declaratives and interrogatives share something which separates them from
imperatives. Then again, in many languages, imperatives (at least in specific cases)
have a form more similar to declaratives than to interrogatives (for example, Italian
polite imperatives, (18d)), suggesting that interrogatives have some property which
distinguishes them from imperatives and declaratives. Furthermore, as mentioned
above, minor types tend to be more closely allied with some basic types than others.
A theory of clause types should seek to explain not only how individual types are
marked within and across languages, but also what patterns of marking are used
for entire clause type systems.
Next we turn to some more cases in which the type of a clause might be seen
as inexplicitly marked. Scholars who are pursuing a syntactic or semantic analysis
of clause types or sentence types are generally unwilling to treat any case as truly
involving the inexplicit marking of the type. Rather, they propose some covert
explicit marking. A relatively simple case for discussion purposes concerns the
ways of expressing a directive speech act in Hebrew (from Sadock 1974, pp. 92–3):

(21) a. šev
sit.2m.sg
‘Sit!’
b. tišev
fut-sit.2m.sg
‘Sit!’ or ‘You will sit.’

i i

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

i i

sentence mood, clause type, and sentential force 135

The form in (21a) can only be used with an imperative-type meaning, but (21b) is
a more common way to issue a directive and is the normal future declarative form
as well. Based on several factors, Sadock argues that the directive meaning is not
pragmatically derived, but rather that the sentence has two distinct Logical Forms
(realized within the Generative Semantics framework at deep structure).
While linguists are often quite willing to propose a covert explicit marking
of clause type in cases where there is no marking visible on the surface, they
are sometimes puzzled by cases in which a clause type seems to be marked in a
descriptive sense, but not in a way which makes sense according to the syntactic or
semantic theory being developed. We can describe the situation by saying that the
type is indicated or determined, but not necessarily “marked.”
One relatively well-studied case where type is indicated but maybe not marked
involves sentences with declarative word order but rising interrogative intonation,
as in (12). Clearly (12c) is an interrogative and (12b) is a declarative. But what about
(12a), with the word order of (12b), the intonation of (12c), and a discourse function
which seeks a response from the addressee? Given its function, it is very natural
to classify (12a) as an interrogative, and in that case, the only visible marking of its
interrogative status is the intonation. This fact leads us to the following possibilities:

1. Sentences like (12a) are explicitly marked as interrogatives by rising

intonation.
2. Sentences like (12a) are explicitly marked as interrogatives by some hidden
grammatical structure which correlates with rising intonation.
3. Sentences like (12a) are inexplicitly marked as interrogatives; rising intonation
is merely a clue to their type.
4. Sentences like (12a) are explicitly marked as declaratives by their word order.
Their function, which involves seeking a response from the addressee, is
atypical for declaratives. We account for this atypicality by saying one of the
following things:
(a) The sentence has the clause type of declarative, but the sentence mood of
interrogative.
(b) The sentence has both the clause type and sentence mood of declarative,
and its function is compatible with declarative sentence mood.

Option 1 requires a framework of grammatical analysis in which intonation,

morphology, and syntax all contribute to defining grammatical categories like
clause type. Option 2 requires some abstract syntax and a grammatical framework
in which syntax can determine intonation. Option 3 requires that we give up on
thinking of clause types as grammatical categories which can be defined in terms
of more primitive grammatical properties. And option 4 requires us to think more
carefully about the way in which clause types are associated with conversational
functions.
Gunlogson (2001) in fact argues for option 4. While (12a) seeks a response,
it is biased towards the response being “yes, it’s raining.” Building on this point,
Gunlogson argues that the declarative clause type conveys a commitment towards
the propositional content of the clause (here, that it’s raining). The difference

i i

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

i i

136 sentence mood

between falling and rising intonation is that, in the former cases, the speaker
indicates her own commitment, while in the latter, she indicates the addressee’s
commitment.
As pointed out earlier in this section, Gunlogson’s analysis raises some difficult
questions for our overall understanding of clause types. The role she assigns to
rising and falling intonation is so central to the discourse model that it is difficult to
see the traditional declarative/interrogative opposition as more relevant to sentence
mood than the opposition between rising and falling intonation. This may mean,
on the one hand, that clause type systems are more complex than we traditionally
assume, or, on the other, that we have failed to identify the fundamental contribu-
tion of the declarative clause type. As we will see in Section 3.3.1, I believe the latter
to be the case; Gunlogson’s analysis of rising and falling intonation is essential, but
more must be added to her analysis of declarative type in order to give a proper
account of the sentential forces of these examples.

3.1.2 Sentential forces as pragmatic categories

The sentential forces are the fundamental conversational functions within
semantic/pragmatic theory. As such, we cannot identify them a priori, but rather
must determine what they are through empirical investigation and a scientific
process of theory-building. Within linguistics and philosophy, there are two main
theoretical approaches to conversational function which have served as a basis
for thinking about sentence mood: speech act theory and the dynamic approach
to meaning. Our goal in this subsection is to understand the range of sentential
forces which scholars assume as the basis for the analysis of sentence mood
and how different theoretical perspectives influence their understanding of which
conversational functions count as sentential forces.
Within speech act theory, the most obvious set of functions to stand in the
role of sentential forces are the illocutionary forces. We have seen in Section 1.4.2
that the existence of indirect speech acts leads to serious problems for the simple
idea that sentence moods can be defined in terms of illocutionary force and that
one would at least have to say that sentential forces are the illocutionary forces
conventionally associated with utterances of a sentence. In other words, we would
identify sentential force with the illocutionary force of a sentence used in a direct
speech act. If one adopts such an approach, one is committed to the claim that the
distinction between direct and indirect speech acts (in particular, the distinction
between conventional and inferred illocutionary forces) makes sense within the
overall understanding of communication within speech act theory.
Let us assume that one can identify direct speech acts and, based on this, define
the conventionally determined illocutionary forces. Even so, since its earliest stages
speech act theory has identified the illocutionary force of an explicit performative
with the act-type named by its main verb. For example, (22a) has the force of
claiming and (22b) has the force of apologizing:

(22) a. I claim that Bill will win the election.

b. I apologize for ruining the stew.

i i

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

i i

sentence mood, clause type, and sentential force 137

Obviously, the idea that performative verbs identify the illocutionary forces of their
sentences leads to a theory with a huge number of illocutionary forces—all of which
are conventionally associated with the performative sentences in question. So, we
end up with a huge number of sentential forces. But normally we think of (22a)–
(22b) as declaratives (both in clause type and sentence mood). This route towards
defining sentential force will lead to the conclusion that declarative sentences, at
least, are associated with a great number of sentential forces and so either (i) the
sentence mood–sentential force relation is not one-to-one, or (ii) the declarative
clause type actually participates in a large number of different sentence moods.
As discussed in Section 1.4.2, this problem has led many scholars of speech act
theory to assume that the sentential force of an explicit performative is not to be
identified with the illocutionary force it seems to have; rather, they claim that the
examples in (22) have the same sentential force, either assertion (the force of non-
performative declaratives) or declaration (the special force of performatives), with
the specific, intended illocutionary forces of claiming or apologizing derived from
the sentential force.
Setting aside explicit performatives, we also find variation in sentential force
within the classes of sentences we normally assume to comprise a sentence mood.
For example, (23a) and (23b) are both imperative, but one is naturally thought
of as having the same force as an explicit performative with order, while the
other is naturally thought of as having the same force as an explicit performative
with invite.

(23) a. Don’t you move!

b. Please have a glass of wine!

In fact, all clause types can be used in diverse ways paraphrasable by a variety of
performative verbs. If we apply speech act theory to the analysis of sentence mood
by identifying illocutionary force with sentential force, examples like those in (22)–
(23) suggest that all clause types will be associated with very many sentential forces.
There are two main strategies available for developing a concept of sentential
force within speech act theory. One of them identifies sentential force with a set
of illocutionary forces. According to Harnish (1994), an illocutionary force
potential is the set of illocutionary forces with which a particular clause type
is compatible. For example, sentences of the declarative clause type are compat-
ible with any illocutionary force which presupposes the speaker’s belief in the
propositional content of the sentence. The set of such forces thus constitutes the
illocutionary force potential of each declarative clause, and we can generalize to
the thesis that they constitute the illocutionary force potential of the declarative
sentence mood.
Another strategy for creating a concept of sentential force within speech act
theory is to group the illocutionary forces in terms of some more fundamental

 It is difficult to express the relation between clause type and illocutionary force with complete

accuracy without laying out the entire framework of Bach and Harnish () and Harnish ().
See the works cited for details.

i i

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

i i

138 sentence mood

property that they share and identify this common property as the sentential force.
As discussed in Section 1.4.2, this is the approach of Searle and Vanderveken
(1985) and related work. Searle and Vanderveken propose five illocutionary points,
the assertive, directive, commissive, declaratory, and expressive points, which
combine with other factors including degree of strength and felicity conditions of
various sorts to produce fully realized illocutionary forces. Within this framework,
sentential forces are naturally identified with the illocutionary points, yielding a
much simpler relation between clause types and sentential forces. However, as
pointed out in Section 1.4.2, this perspective still does not lead to a theory of
sentence mood which treats the basic clause types as fundamental in the way that
linguists traditionally assume that they are. Notably, it groups interrogatives (which
have the directive point—specifically, they direct the addressee to answer) with
imperatives, even though in grammatical terms interrogatives and imperatives
each share more with declaratives than they do with each other. It also fails to
explain why promissives (with the commissive point) are so much rarer than
imperatives (with the closely related directive point) and how exclamatives (which
presumably have the expressive point) are related to interrogatives. The version
of speech act theory which uses the concept of illocutionary point contains an
interesting intuition about the relation between explicit performatives and other
declaratives (see Vanderveken 1991, pp. 140–1), namely that the declarative point
entails the assertive point and therefore assertion is marked by adding something
onto the basic declarative sentence mood, but it is still not clear why declarations
across languages are never realized with a distinct clause type of their own.
There have been other attempts to reconcile speech act theory with a
linguistically informed notion of sentence mood, including Zaefferer (1986, 2001,
2007) and work which follows the dynamic force hypothesis (Section 1.4.3). We
will discuss the analysis of sentence mood in speech act theory in more detail in
Section 3.2 below.
In comparison with speech act theory, the dynamic approach appears to be
much more readily suited to play a role in the theory of sentence mood. The
reason for this advantage is simply that it developed in its earliest stages with
the goal of giving an idealized model of conversation which emphasizes those
functions, like assertion and asking, which are needed for the analysis of sentence
mood (for example, Hamblin 1971); it has downplayed highly specific functions
like urging or warning, which have been of interest to speech act theory. It is not
clear whether this focus on fewer, more general conversational functions initially
resulted from an unarticulated goal of explaining the pragmatics of clause types, or
whether it is the result of an intuition which sees as fundamental the same functions
which are fortuitously also treated as important by grammar. Eventually, many
scholars working within the dynamic approach have taken on the explicit goal of
explaining sentential force within a theory of sentence mood which focuses on

 Perhaps exclamatives are related to interrogatives because the two share certain propositional con-

tent conditions. That is, perhaps the interrogative-like structures which we often see with exclamatives
are due to the fact that the expressive point needs to operate on a semantic meaning which is similar to
that which the directive forces related to asking do.

i i

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

i i

sentence mood, clause type, and sentential force 139

the basic clause types (e.g. Portner 2004; Roberts 2004; Mastop 2005; Starr 2013).
These scholars try to provide a theory of update potentials which treats asserting a
proposition, asking a question, and directing someone to do something as simple
and fundamental because of the fact that they seem fundamental when we look at
clause type systems. The challenge for the dynamic approach is more in explaining
the functions of language other than these basic sentential forces. We will return
to the analysis of sentence mood within the dynamic approach in Section 3.3.
Although the dynamic approach may provide a more natural basis for develop-
ing a theory of sentence mood than classical speech act theory, we should not over-
state the importance of this advantage. It is quite possible that a version of speech
act theory could be developed which better matches basic illocutionary forces or
illocutionary points with clause types. What is not clear, though, is whether the goal
of contributing to the theory of sentence mood is sufficiently compelling to justify
adjusting the classification of communicative acts embedded in classical speech
act theory. It might just be, for example, that there is a truth to the idea that the
function of interrogatives and the function of imperatives are closely related which
needs to be incorporated into the philosophy of language even at the cost of not
participating in the explanation of empirical patterns related to sentence mood.

3.1.3 The syntax/sentence mood interface

Sentence moods are conventionally linked to conversational functions. Either
speech act theory or the dynamic approach would lead us to say that the declarative
in (24a) is linked to the function of assertion, specifically the speaker asserting that
Ben played horn in the concert. Likewise, (24b) is linked to asking (whether he
played horn) and (24c) to directing (the addressee to play horn).

(24) a. Ben played horn in the concert.

b. Did Ben play horn in the concert?
c. Ben, play horn in the concert!

Because of obvious associations like these, we conclude that the declarative form
of (24a) is linked to the function of asserting, and so on for the other types. Since
each of the sentences in (24) can be used for other functions, though, we think of
sentence mood in terms of some kind of special, conventional association between
form and conversational functions. We have called the function with which a
sentence mood is associated (in the right, “conventional” way) its sentential force.
Scholars of pragmatics have long struggled with the issue of how to define the
sentential force of a sentence. Although we have an intuition that (24a), used in its
most “basic” way, is used to make an assertion, we have no simple way of identifying
basic uses from non-basic uses, and, as pointed out in discussion of the literal force
hypothesis in Section 1.4.2, there are sentences which are not so naturally used with
the force which we wish to specify as their sentential force. It seems that, if we

 A good example is Levinson’s () May I remind you that your account is overdue. Because this

sentence accomplishes the reminder, it cannot possibly be asking permission to remind.

i i

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

i i

140 sentence mood

believe in sentential force, we must give up on defining the force of a sentence based
on its basic use, considered in isolation from the rest of the language. We must,
rather, identify the sentential forces of sentences through a more holistic process
of theory-building, whereby sentence moods and sentential forces are linked even
though some sentences may not as a practical matter ever have the exact function
afforded by their sentential force.
Besides the descriptive question of how to identify the sentential force of a
particular sentence, linguists have addressed an important issue concerning the
syntax/meaning interface: How does the grammatical form of a sentence contrib-
ute to its sentential force? That is, starting with a grammatical structure assigned to
a sequence of morphemes (and maybe other meaningful components, like inton-
ation), what semantic and pragmatic processes apply and how do they determine,
constrain, or influence the sentential force which is eventually associated with the
sentence? We find three main types of response to this question in the literature:

1. The operator approach. A distinct element in the sentence’s representation

determines its sentential force. This operator might be identified as a morph-
eme in the grammatical representation or be treated as a more abstract
element of logical form, and it might determine the sentential force either
through its semantics or by triggering an interpretive rule at the discourse
level. In canonical cases like (24), we can say that the operator marks the
clause type and encodes the sentential force.
2. The construction-based approach. Sentence moods are patterns of grammat-
ical form, or constructions, which are translated into sentential force by
(often implicit) interpretive rules. Looking at the standard paradigm (24), we
can say that the construction as a whole marks the clause type and encodes
the sentential force.
3. The compositional approach. Sentential force is not itself encoded in the
grammatical form at all; rather, the grammatical form is compositionally
interpreted, with the force arising from the combination or interplay of all
of the particular meaning-components which are derived through the com-
positional processes. The analysis might also involve a model of the discourse
context and a pragmatic theory of how sentence-meanings interact with that
context. From this point of view, sentence moods are epiphenomenal patterns
which arise from more fundamental processes.

These three approaches are not entirely distinct. For example, the first can be
combined with either of the other two. We might say, in accordance with the second
approach, that the construction includes an operator as part of the pattern. Or we
might say, in connection with the third, that while there is no force-encoding
operator in the grammatical form, there is an operator with a more standard
semantics which effectively (though indirectly) determines the sentential force.

 For example, Rawlins () defines two operators Q and A which have no meaning except to

place restrictions on the denotations of their arguments. He assumes that they should also “capture”
the speech acts of asking and assertion (presumably by encoding force), but he leaves this out of his

i i

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

i i

sentence mood, clause type, and sentential force 141

Before going through these approaches, it might be helpful to point out a pair
of ideas which are present in many analyses across all three. The first of these is
that an important role is played by the position (or set of positions) in which we
find complementizers like that and whether and to which the tensed verb moves
in languages which show verb-second order in root declaratives. This position
is labeled C in many generative systems and the phrase it heads is CP. We will
repeatedly find versions of the intuition that the root CP is associated with a
sentential force on the basis of the material in C. The same fundamental intuition
can be seen in theories which do not have a single CP, but rather “split” it into a
larger functional field (e.g. Rizzi 1997), and in construction-based theories (such
as Ginzburg and Sag 2001). The second theme which will emerge is the idea that
sentence mood is fundamentally a matter of which syntactic features a clause
has. For example, we see features like [+wh] and [imp(erative)]. Since syntactic
features have syntactic effects, like driving movement or requiring the insertion
of phonological material, these features give rise to the overt patterns which we
typically think of as marking clause types. In addition, these syntactic features have
meanings which influence or determine the sentential force of the clauses in which
they occur. The generalization that the C position is relevant to sentence mood and
the strategy of marking clause type through the use of syntactic features go together
in a natural way: many analyses place some kind of syntactic feature relevant to
sentence mood in C.
The importance of the C position (or more generally, the syntax involving
complementizers and verbs at the periphery of the clause) leads to another way in
which we may classify approaches to the type/force relation. Do they use the syntax
of CP primarily to explain clause type, sentential force, or some other syntactic
generalization? We see three approaches. For a given clause type, it is possible that:

1. All members of the clause type share a crucial property of CP syntax. For
example, if we assume that clause type is marked by the presence of a
particular syntactic feature in C, this approach would claim that all members
of the clause type—whether root or embedded—have the feature. We would
then probably say that, depending on the syntactic context, the feature can
either determine a sentential force or contribute to the semantic processes
involved in subordination.
2. Only members of the clause type with independent force share the crucial
property of CP syntax. Again, on the assumption that this property is the
presence of a specific feature in C, this way of thinking would say that
members of that clause type which lack force (for example, complement
clauses) lack the feature. The clause type itself would then need to be defined
by other syntactic properties.
3. Members of the clause type with independent force, as well as sentences
lacking independent force but with a CP-level syntax like those which do,

analysis because it is not crucial to the points he wants to make. However, within the compositional
approach, an operator which restricts the sentence’s compositional semantic value could be sufficient
to determine its sentential force.

i i

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

i i

142 sentence mood

have the crucial property. In terms of syntactic features, we would say that a
specific feature in C is shared by all root clauses with independent force as
well as some subordinate, non-force-bearing clauses.

A prime empirical domain for observing these three approaches concerns

Germanic V2 declarative clauses. For the most part, root declarative clauses have
the tensed verb in C, while embedded declarative clauses have the verb in a lower
position. However, it has long been noted that certain subordinate clauses allow
the verb in C. We exemplify this with Swedish examples based on Wechsler (1991).

(25) a. Hon köpte inte en ny bil idag. (Swedish)

she bought not a new car today
b. Johan vet att [hon inte köpte en ny bil].
John knows that she not bought a new car
c. Hugo påstod att [hon köpte inte en ny bil].
Hugo claimed that she bought not a new car

In these examples, the verb is understood to be in C when it precedes the negation

inte. A theory focused on clause-typing would place a declarative feature in C
in (25a–c), and the feature would mark clause type. A theory focused on force
would have the feature in C for (25a) only, and it would be understood to encode
or otherwise determine force. And one which aims to explain the behavior of
the tensed verb would have it in C for (25a,c) but not (25b); in this case, the
feature would then be seen as representing something which root sentences and
sentences embedded under ‘claim’ have in common, but which is not shared
by sentences embedded under ‘know.’

The operator approach. We can begin this discussion with the concept of a force-
indicating operator. The idea that something like force is represented via a primitive
operator has its origins at least as far back as Frege’s judgment stroke, long before
linguistic theory had advanced to a point which would allow syntacticians and
semanticists to formulate precise theories of sentence mood. Building on a long-
standing tradition, Stenius (1967) gives a very influential version of the idea at
just about the point in time when speech act theory was becoming the dominant
perspective on how linguistic meaning is used for communication. According to
Stenius, the logical form of a sentence is M(p), with a modal element M as the
force indicator and sentence radical (p) expressing the descriptive content.
Stenius specifically identifies three force indicators, I, O, and ?, for the indicative,
imperative, and interrogative mood, respectively. Each is associated with specific

 See Huntley () for a brief summary, including an interesting discussion of Frege’s ambiguous

position. Dudman (, ) also provides helpful background on Frege.

 Note that the parentheses are part of the notation for the sentence radical. Stenius actually says

that M signifies the sentence’s mood, but this clashes with our terminology. We would say it is the mood
(in the logical language) and therefore signifies the sentential force.

i i

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

i i

sentence mood, clause type, and sentential force 143

rules which determine how the descriptive content can be used in a conversation.
(Stenius’s broadly Wittgensteinian ideas cannot be readily classified within either
speech act theory or the dynamic approach.) Thus, Stenius builds into his proposal
the suggestion that the conversational functions which he discusses should be
understood as associated with the major clause types. In this way, I think, he
implies that they are in fact the most fundamental conversational functions, that
is sentential forces. What he does not do, though, is address the interface question
of how a logical form with I, O, or ? is derived.
Many linguists and philosophers have followed Stenius in taking on an implicit
goal of explaining sentence mood without being completely explicit about how
grammar and force are related. For example, Bierwisch (1980) proposes that declar-
atives, imperatives, and interrogatives have logical forms (or “semantic structures”)
p, Imp, p, and Qu, p, respectively. The logical form of a sentence determines
its utterance meaning in the context in which it is used; an utterance meaning is a
pair of an attitude and a propositional content. Attitudes are “pre-reflexive,” funda-
mental stances which one can take towards a proposition, and among the attitudes
are three crucial ones D, I, and Q (roughly ‘taking it to be the case that,’ ‘intending
that,’ and ‘intending to know’). Specifically, the utterance meaning of a declarative is
D, [[ p ]] , that of an imperative I, [[ p ]] , and that of an interrogative Q, [[ p ]] .
These utterance meanings then influence the determination of which speech act is
ultimately performed by the use of the sentence in context.
Bierwisch’s paper elegantly distinguishes the syntax/semantics notion of a force-
marker (Imp, Qu, and the absence of either in declaratives) from the components
of utterance meaning (I, Q, and D) which help determine the type of speech act
performed. Bierwisch does not, however, help us understand what Imp and Qu
precisely are. What kind of thing could Imp be which “determines” the attitude I
but which is distinct from I? Moreover, he does not present an analysis of how the
syntactic structure of each clause type leads to the right force-marker being present
in logical form.
Another operator-based analysis which proceeds at a similar level of abstraction
is given by Krifka (2001). He uses the “illocutionary operators” assert, command,
and quest for declaratives, imperatives, and interrogatives, and he means to
imply, I think, that the speech acts these define are sentential forces—fundamental
conversational functions linked to particular grammatical forms. In Krifka (2001)
he does not develop a theory of the forces themselves, instead referring to Wittgen-
stein (1953), Stalnaker (1974), Heim (1982), and Merin (1991). (In later work, such as
Krifka 2014 discussed later in this section, he develops a dynamic semantics system
with illocutionary operators.) In (26), for example, Krifka (2001, p. 21) assigns types
to assert (type pa, functions from propositions to speech acts) and ˆraining (type
p), composing them by function application.

(26) a. It is raining.
b. assert(ˆraining)

 Bierwisch actually allows the utterance meaning of a declarative to include an attitude other than

D, though D is the unmarked case.

i i

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

i i

144 sentence mood

Yet, while Krifka gives derivations of speech act-type meanings, he does not
describe a morpheme or propose a semantic rule which introduces the crucial
operators like assert.
Stenius, Bierwisch, Krifka, and others following them do not make any specific
claims about the grammar/sentential force relation; as Reis (1999, pp. 198–9)
says, works of this kind presuppose a notion of sentence types, but “skirt the
crucial grammar-internal issue, which should be uppermost in a linguist’s mind.”
Other linguists have, however, proposed that similar force-indicating operators are
directly represented in syntax. An important early analysis along these lines is given
by Rivero and Terzi (1995). They propose that the syntactic C position “preferably”
hosts the sentence’s modal element, “the formal ingredient that contributes to the
determination of illocutionary force” (p. 305; they also call this element “logical
mood”). Specifically citing Stenius (1967) as explanation for how they intend for
this element to be interpreted, they then propose that, in the case of Modern Greek
imperatives, the modal element should be understood as having the syntactic status
of a grammatical feature which drives verb movement to C. The result of verb
movement is a structure which represents sentential force in C. Rivero and Terzi are
not, however, clear about the syntax of the force-marking element in imperatives in
those other languages, like Ancient Greek, in which the imperative modal element
is not in C, nor is it clear how sentential force is represented in other clause types.
Rizzi (1997) goes a step further than Rivero and Terzi and proposes a functional
projection ForceP which identifies the sentence’s clause type and serves as its
interface with higher grammatical structure or the discourse context. While Rizzi’s
proposal is focused on the syntax, and so does not incorporate any specific
idea about the nature of sentential force, it is natural to interpret his analysis as
representing in the ForceP an element which (in addition to identifying its clause
type directly) determines the sentence’s sentential force, if it has one (e.g. if it is not
embedded). In this way, we can see Rizzi placing into syntax a version of the force-
indicating operator. One thing that is left mysterious by Rizzi and his followers,
however, is exactly what Force does in an embedded clause and how this is related
to the assignment of illocutionary force in root contexts.
The idea of a force-indicating operator heading a syntactic phrase has become
quite common. Platzack and Rosengren (1994) offer a brief but interesting dis-
cussion of clause type and sentential force using the ForceP. According to them,
clause typing is accomplished through the presence of syntactic features in ForceP;
specifically [−wh] in Force0 (the head of ForceP) marks declarative, [+wh] marks
interrogative, and [imp] marks imperative. These features do not, however, in-
troduce the characteristic meanings of the clause types. In addition to the features,
interrogatives contain a Q operator and imperatives an N operator (for “necessity”).
Declaratives contain no extra operator. In their analysis, it is not entirely clear what

 Within syntax, ‘operator’ sometimes has a technical meaning which we should set aside here. The

point is that we see strong similarities between the kind of element which we are calling the force-
indicating operator and the material in ForceP in Rizzi’s syntactic theory.
 Note that a [+wh] feature can be present without a wh-word, and a wh-word can co-occur with

a [−wh] feature. See Platzack and Rosengren (, pp. –).

i i

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

i i

sentence mood, clause type, and sentential force 145

aspects of meaning are accounted for by the features and which by the operators. In
the discussion of Q, they refer to “the rather complicated semantics of interrogative
clauses” and say that it “signalizes” slightly different things in yes/no and wh-
interrogative clauses. In the case of imperatives, it is more clear that N introduces a
sentential force: “The operator so to speak directly sets or creates what we may call a
norm” (Platzack and Rosengren 1994, p. 188). The operators Q and N are associated
with the features [+wh] and [imp] in ForceP, but Platzack and Rosengren are
not explicit about the syntactic relation between them, nor do they delve into
other important issues like whether the operators are present only in clauses with
independent force.
Another influential version of the idea that force is encoded in an operator is de-
veloped by Han (1998). Han proposes that imperative clauses contain an imperative
operator consisting of two features, [directive] and [irrealis], the former of which
encodes directive illocutionary force. Han clearly identifies illocutionary force with
sentential force here (adopting the literal force hypothesis for imperatives), so her
account is that imperatives contain a clause-typing operator which includes, as
one component, a feature which encodes sentential force. (The feature [irrealis]
also occurs in subjunctives and infinitives.) Han differs from many proponents
of the force-indicating operator, however, in that she does not apply it uniformly
across the grammar. She allows that in many cases, sentential force is not directly
encoded. She briefly indicates that declarative and interrogative clauses receive
their illocutionary force by pragmatic reasoning. More importantly, she argues that
imperatives which are not expressed with the canonical imperative morphosyntax
receive their force by pragmatic inference as well. In other words, she divides the
clause type of imperative sentences into those in which force is encoded and others
for which it is inferred. Because Han develops in an explicit way a line of reasoning
which many linguists seem to assume implicitly, we will briefly review her reasons
for dividing up the clause type of imperatives in this way.
In both Platzack and Rosengren’s and Han’s analyses, the crucial feature
(i.e. [imp] or [directive]) not only encodes sentential force; it also serves the
purely syntactic function, in many of the languages they discuss, of driving verb-
movement to the left periphery of the clause. We can see this in Rivero and Terzi’s
(1995) example cited by Han, (27):

(27) Lée lo! (Spanish)

read.imp it
‘Read it!’

However, in certain cases imperative meaning cannot be expressed with the verb
form and syntactic construction seen in simple imperatives like (27). Han focuses
on one very common type of such case, negative imperatives. In Spanish, negative
imperatives are expressed with a verb form from a different paradigm, such as
subjunctive or infinitival. We see this in Han’s examples (28):

 The subjunctive is the normal way of expressing the negative imperative meaning in Spanish, but

the infinitive is standard in other languages like Italian.

i i

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

i i

146 sentence mood

(28) a. ∗ No lée lo! (Spanish)

neg read-2sg.imp it
‘Don’t read it!’
b. No lo leas!
neg it read-2sg.subj
‘Don’t read it!’
c. No leer lo!
neg read-inf it
‘Don’t read it!’

Notice that the verb follows the object in (28b), indicating that [directive] is not
present. As a result, according to Han force is not encoded and therefore must
be determined pragmatically. Han argues that subjunctives contain [irrealis] and
that its meaning is sufficient to allow hearers to identify the sentence’s directive
force. Han gives the same explanation to the infinitival negative imperative (28c),
even though the verb precedes the object clitic in this case. We therefore see
in Han’s theory a combination of the first (force-indicating operator) and third
(compositional determination of force) approaches outlined earlier in this section.
It is worth asking whether there is any reason to accept that [directive] encodes
force, if [irrealis] by itself is enough to guarantee that (27) is associated with
directive force. (There is in fact a reason within Han’s theory; on the assumption
that force cannot be encoded within the scope of another operator, the idea that
[directive] encodes force explains why (27) cannot be negated or embedded.)
Another theory which straddles the line between the operator approach and
the compositional approach is due to Truckenbrodt (2006a), and his theory also
has some properties which are worth discussing here. Building on the important
tradition of research on the interaction between verbal mood and verb movement
in Germanic, he proposes that C contains a “context index” which determines
sentential force. The context index is a rather complex object, with the form
DEONTS (, x)1 (, EPIST)2 . The details of the theory are complex, but it is built
from three basic ideas:

1. All speech acts can be paraphrased as “so-and-so wants p.” The attitude of
wanting is represented by DEONT.
2. With speech acts of assertion and asking, what is wanted can be paraphrased
as “so-and-so knows p.” The attitude of knowing is represented by EPIST.
3. Sometimes the speech act involves the speaker wanting someone else to be
responsible for bringing about what he wants. This situation is represented by
including the variable x and setting its reference to that individual.

 We encounter this tradition throughout this chapter. Representative works include Altmann ;

Meibauer ; Wechsler ; Platzack and Rosengren ; Brandt et al. ; Reis , ;
Lohnstein , ; Gärtner ; Brandner ; Meinunger ; and the papers in Rosengren
a,b.

i i

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

i i

sentence mood, clause type, and sentential force 147

Truckenbrodt describes syntactic principles which determine the form of the

context index. The first concerns DEONT: root C contains DEONT but embedded
C does not. The second concerns EPIST: EPIST is present when either indicative
mood or [+wh] is in C. This leads, roughly, to the following possibilities:

(29) a. Imperatives have DEONT but not EPIST, and only occur as root clauses.
They express the meaning that the speaker wants x (= the addressee) to
do something.
b. Root declaratives and interrogatives have DEONT and EPIST. They mean
that the speaker wants p to be common ground, either on the basis of
the speaker’s contribution (declaratives, where x = the speaker) or the
addressee’s (interrogatives, where x = the addressee).
c. Embedded clauses lack DEONT, but with V2 contain EPIST. Embedded
V2 is only possible when EPIST is compatible with the semantics of the
embedding predicate.

Truckenbrodt (2006a, p. 267) introduces a rule which assigns a context change

potential to any CP with a context index containing DEONT. In this way, DEONT
makes the whole index into a force-indicating operator.
Although Truckenbrodt’s article proposes a version of the operator-based ap-
proach, it should be noted that in a response to commentaries on his ideas
(Truckenbrodt 2006b) he revises the analysis just outlined. Specifically, he removes
DEONT from the context indices and achieves the same effect by means of
a pragmatic rule which applies to root clauses. This revised theory no longer
explicitly represents sentential force and is more in line with the compositional
approach to the relation between clause type and sentential force.
All of the operator-based approaches we have discussed so far have an important
assumption in common. If they take a clear stand on the issue at all, they assume
that their operator only introduces sentential or illocutionary force into the clause’s
meaning when the clause is unembedded. This assumption is not accepted by
one recent analysis, however. Building on his earlier work discussed above, Krifka
(2014) proposes that illocutionary operators can be present in the logical forms
of both root and embedded clauses. A sentence-embedding verb will only be able
to combine with a clause whose meaning is built with such an operator if it has a
meaning which can take a speech act (a meaning built from a force and a content)
as argument.
In order to see the issues at play, it is worthwhile to compare Truckenbrodt’s and
Krifka’s analyses of embedded V2 clauses in German. The pattern in German is
very similar to the Swedish one outlined in (25). In (30), we see that embedded

 Embedded V is one example of what are often called main clause phenomena, morphemes

or constructional patterns which for the most part occur only in root clauses. Emonds’ () root
transformations are classical examples: negative inversion, topicalization, VP-preposing, and others.
Hooper and Thompson () critique Emonds’ classification, showing that many root constructions
can occur in embedded clauses, and argue that they are actually restricted to “asserted” environments,

i i

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

i i

148 sentence mood

V2 clauses naturally occur in German as the complement of ‘say’ and ‘believe,’ but
not ‘want’ (examples from Truckenbrodt 2006a):

(30) a. Maria glaubt [Peter geht nach Hause].

Maria believes Peter goes to home
b. Maria sagt [Peter geht nach Hause].
Maria says Peter goes to home.
c. ∗ Maria will, [sie ist in diesem Fall in Berlin].
Maria wants she is in that case in Berlin.

The explanation of embedded V2 should aim to relate V2 to a property shared by

both root assertions and the meaning of verbs like ‘say’ and ‘believe,’ as opposed
to ‘want.’ Building on Hooper and Thompson (1973), Meinunger (2004) says that
V2 clauses have “assertive potential” and that only predicates which are compatible
with this meaning may embed them.
Meinunger’s semantic analysis is sketchy, but Truckenbrodt and Krifka give
more content to the notion of “assertive potential.” According to Truckenbrodt,
embedded CPs with a verb in C contain EPIST, the semantics of which implies that
the referent of the matrix subject believes the content of the V2 clause. According
to Truckenbrodt, this implication is compatible with ‘believes’ but not ‘wants’ (he
says that ‘believe’ “absorbs” the entailment of EPIST), so the sentence is acceptable.
The idea of absorption is intriguing, but we would want to operationalize it more
precisely before judging whether it captures a useful notion of “assertive potential.”
Krifka (2014) is not explicit in the same way as Truckenbrodt concerning his
assumptions about the interface between clause type and sentential force, but he
does think about the pattern in ways which are crucially different from Meinunger
and Truckenbrodt. He argues that V2 introduces assertive force (not merely
assertive potential) even in some embedded clauses. So, for example, in (30b) the
embedded clause would have the logical form assert(p), just as the same structure
would as a root clause. In this case, the predicate ‘say’ would not have a meaning
which takes a proposition as argument, but rather it would be a function which
takes an act-type argument, together with its individual arguments, and expresses
the proposition that a speech act of the kind denoted by the complement took place.
Though Krifka does not take a stand on the full range of embedded V2 cases, in
order to extend his analysis to (30a,c), one would have to say that ‘believe’ can
embed an assertive act, but ‘want’ cannot.

The construction-based approach. As we saw in Section 3.1.1, the traditional

conception of sentence types incorporates a broadly construction-oriented per-
spective on the status of clause types in grammar. This conception leads to two

whether syntactically root or embedded. For more on the syntactic perspective on main clause
phenomena, see Haegeman () and the papers in Aelbrecht et al. ().
 See Wiklund et al. () and references therein for more on the semantic distinction between

predicates which allow embedded V and those which do not.

i i

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

i i

sentence mood, clause type, and sentential force 149

serious problems: first, the classes of sentences we intuitively think of as clause

types can be non-uniform in their morphology and syntax, and these differences
typically correspond to differences in meaning. For example, as we saw in our
discussion of Italian, Spanish, and Greek imperatives, different subtypes of im-
peratives are often expressed with different grammatical patterns. This situation
pushes one towards the view that there are actually two or more distinct clause
types (positive imperatives, prohibitives, polite imperatives), and unfortunately
this leads, within the traditional view, to the conclusion that imperatives as a group
have no theoretical status at all.
The second problem concerns the arbitrariness of the mapping between form
and sentential force. The traditional conception assigns sentential force holistically.
For example, it would map the Spanish negative imperative pattern seen in (28b)
(negation plus clitic preceding the subjunctive verb) to the force of prohibition.
The approach does not offer an explanation for why a particular combination of
grammatical properties is associated with a particular force. But clearly (28b) has
the meaning it does largely in virtue of the meanings of its characteristic parts, such
as the negation no and the subjunctive verbal mood.
One way of addressing these problems is to define the relevant patterns only in
terms of a very restricted set of grammatical properties. For example, Wechsler
(1991) proposes that the Swedish CP is assigned sentential force (what he calls
“illocutionary potential”) if it contains a finite verb in C, with the particular
force determined by two simple properties: whether it is root or embedded and
whether the verb is preceded by another element in CP. (Technically, the second
property is whether the maximal projection of CP is C or C .) For example, a
verb-initial root CP will be assigned the force of a polarity question, while an
embedded verb-second clause with the tensed verb in C is assigned the force
of (in)direct assertion. In Wechsler’s system, we do not have to worry about
distinguishing sentence moods too finely, since the system can only define a small
number of them. And on the assumption that such elements as negation and
subjunctive have their normal semantic contribution, it also reduces the problem
of arbitrariness greatly. We can see Wechsler as proposing a very modest version
of constructionism which focuses only on sentence mood. More recent work
within the construction grammar approach has not been so modest, though, and
seeks to solve the problems in a different way. The key idea of such work is that
constructions form an inheritance network.

 We should not assume that Wechsler () would see himself as advocating a construction-based

approach. Perhaps certain details of the analysis were not essential to his thinking. However, because it
is given in precise terms, we can see that it does in fact have the characteristics of that approach.
 The specification “(in)direct” apparently indicates two alternatives: either direct assertion for

examples like (i) or indirect assertion for ones like (ii) (his (b), (a)):
(i) . . . eftersom Johan har inte kommit
since John has not come
(ii) Hugo påstod att du kommer aldrig att läsa den här boken.
Hugo claimed that you will never read this book
Indirect assertion is the illocutionary potential of embedded V clauses.

i i

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

i i

150 sentence mood

In construction grammar, sentences instantiate abstract grammatical patterns,

or constructions, which include specifications of both form and meaning. A
construction can be understood as an abstract type and particular phrases and
sentences as tokens of the types. The constructions form a hierarchy in which pairs
of constructions stand in inheritance relations: If construction c1 inherits from c2 ,
then c1 has all of the properties of type c2 , plus some more. For example, specific
constructions like ‘polar interrogative,’ ‘fronted-wh interrogative,’ and ‘in-situ-wh
interrogative’ could all inherit from a general ‘interrogative’ construction; mean-
while, the two wh-interrogative constructions could inherit from a general ‘wh-
clause’ construction. A particular sentence like Who left? would then instantiate a
specific construction (‘fronted-wh-interrogative’) and thereby have the properties
specified by both of the constructions from which that one inherits (interrogative
and wh-clause).
The constructional framework has promise for solving the problems of the
traditional approach to clause types. The first problem, that of overly specific clause
types, will be solved by having both specific and general constructions. In the
case of interrogatives, for example, the goal would be to group all members of
the clause type under ‘interrogative’ even while assigning specific sentences to
constructions which inherit from it. Each sentence will involve a more specific
form–force pairing, but will share those aspects of the pairing specified by the
overall type. The second problem, the arbitrariness of the form–function pairing,
might also be solvable. Certainly a simple association like that between a negative
morpheme and sentential negation could be captured using inheritance (from a
‘sentential negation’ construction).
Michaelis and Lambrecht (1996) give one of the first construction-based analyses
of a clause type. They focus on a sentence pattern which they label the nominal
extraposition construction illustrated in (31) (from Michaelis and Lambrecht
1996, (1)):

(31) It’s amazing the people you see here.

They propose that the nominal extraposition construction inherits from a gen-
eral abstract exclamative construction which brings in several semantic/
pragmatic components. They are not very specific about how this construction
should be related to the concepts of sentence mood and clause type. It seems that
they treat exclamation as a type of assertion, namely “assertion that something
manifests a property to an unusually high extent” (Michaelis and Lambrecht 1996,
p. 238). It’s not clear, however, whether they would consider this type of assertion
to be a sentential force in a sense which would qualify it as a sentence mood. While
their article gives a good idea of how construction-based theories approach the
analysis of clause types, it does not go into enough detail to allow us to discern a
theory of sentence mood.
The most fully developed theory of sentence mood based on the constructional
approach is given by Ginzburg and Sag (2001). They present a precisely formalized
analysis of constructions corresponding to the traditional clause types declarative,
interrogative, imperative, and exclamative. They discuss the semantic values of

i i

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

i i

sentence mood, clause type, and sentential force 151

each type as well as the mechanisms by which each is assigned a sentential

force when unembedded. It is worthwhile to study their theory with the goal of
understanding the potential of the construction-based approach.
Ginzburg and Sag’s theory of sentence mood is based on two central claims:

1. Each clause type is associated with a distinct type of semantic value:

(a) The declarative construction specifies a semantic type of proposition.
(b) The interrogative construction specifies a semantic type of question.
(c) The imperative construction specifies a semantic type of outcome.
(d) The exclamative construction specifies a semantic type of fact.
2. The root construction is associated with the semantic type of proposition
together with an appropriate illocutionary relation towards the semantic
value of the clause it embeds:
(a) When the root construction embeds a declarative clause, the illocution-
ary relation is assertion.
(b) When the root construction embeds an interrogative clause, the illocu-
tionary relation is asking.
(c) When the root construction embeds an imperative clause, the illocution-
ary relation is ordering.
(d) When the root construction embeds an exclamative clause, the illocu-
tionary relation is exclaiming.

According to this view, the root interrogative in (32a) has the semantic value (32b):

(32) a. Who left?

b. the proposition that the speaker asks the addressee the question who left.

This semantic value is derived by combining the interrogative construction, which

gives a semantic value of question (namely, the question who left), with the root
construction, which causes the sentence as a whole to denote the proposition that
the speaker asks the addressee this question.
Ginzburg and Sag do not make clear how making an utterance with the meaning
(32b) amounts to asking who left. As they formalize it, the root construction
denotes a proposition built with a performative modal operator, and thus in its
understanding of semantics and pragmatics, their theory seems little different
from the performative hypothesis together with the assertion-based theory of
performatives (see Section 1.4.2). Apart from this important issue, Ginzburg
and Sag’s work shows that a construction-based theory can be very effective at

 Ginzburg and Sag (, p. ) argue that their analysis does not suffer from certain problems

which affect the performative hypothesis, but I do not understand their claims. First, they state that
information about illocutionary force is not syntactically represented, but it does occur within the
feature structures which are the syntactic representations within the HPSG theory. And second, they
state that “the way in which CMT [conversational move type] information enters into the content of
a sign does not affect the assignment of (non-illocutionary) content.” However, simply labeling the
illocutionary relation an “illocutionary relation” does not make it special, and since the content of the
root clause is a proposition, it seems that the illocutionary relation is nothing but a modal operator.

i i

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

i i

152 sentence mood

making precise the concept of clause type and defining the clause types in a
particular grammar. It improves upon the traditional conception of sentence types
by allowing for types to be defined at multiple levels, including at the high level of
the common-sense declarative, interrogative, imperative, and exclamative types.
Though Ginzburg and Sag develop a traditional theory of clause types and
sentential force based on the performative hypothesis, their system does open
up some other possibilities which are worth remarking on. The way they assign
distinct semantic values to the clause types actually lends itself naturally to the
compositional approach to the assignment of sentential force. They might have
said something like this: the root construction does nothing but add a [root]
feature to the clause. Principles of pragmatics assign a dynamic update function
(a context change potential) to root clauses, selecting the appropriate update
type based on the clause’s content (proposition, question, outcome, or fact). For
example, the natural update type for propositions is assertion, while the natural
update type for questions is asking, so declaratives are assigned an assertive update
while interrogatives are assigned an asking update. Such an analysis would be
developed within a construction-based theory of grammar, but it would not be
a constructional analysis of how sentential force is assigned. I am not certain,
however, whether such an approach would have been compatible with Ginzburg
and Sag’s broader views on semantics and pragmatics.

The compositional approach. Under both the operator-based approach and the
construction-based approach, some grammatical element or pattern represents a
sentence’s sentential force. This force then influences or determines the ultimate
communicative effect of the utterance. For example, in Stenius’s (1967) theory,
grammatical form determines whether the modal operator is I, O, or ?, and then
this operator determines which rules of use apply. When a sentence M(p) is
actually used, we implicitly invoke a principle like this: “when someone utters a
sentence with logical form M(p), apply the instructions associated with M to p
in order to determine what counts as rule-abiding behavior going forward.” In
contrast, the compositional approach to sentential force recommends that we take
the instructions for how p is used out of the logical form, and instead derive the
instructions from p itself. We invoke a principle like this:

(33) Someone who utters a sentence with meaning p uses p according to the rules
which are appropriate given the semantic properties of p, under such-and-
such normal conditions.

An analogy goes as follows. Suppose you work in a metal shop with two big
machines. One takes a bar of metal and turns it into a blade, and the other takes a
bar of metal and turns it into a frying pan. From time to time I send you some metal
to work with. Stenius’s theory is like a situation where I attach a red or blue label; the

It might be possible to solve the second problem by taking a dynamic approach and interpreting the
illocutionary relation as a context change potential.

i i

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

i i

sentence mood, clause type, and sentential force 153

red label indicates that you’re to make a blade while a blue one indicates that you’re
to make a frying pan. By contrast, the compositional approach is like the scenario
in which the metal is either steel, appropriate for making a blade, or aluminum,
appropriate for making a frying pan, and you are expected to use your judgment
to put it into the right machine. Continuing the analogy (and perhaps overdoing
it), a theory in which sentence meanings are full speech acts, like Krifka’s, is like
my loading the bar into one or the other machine for you, and only expecting you
to turn it on. A theory in which sentence meanings are modal propositions based
on distinct semantic types at the content level, like Ginzburg and Sag’s, is like the
situation where the blade might be steel or aluminum, and I attach a note telling you
which machine to put it in, since you can’t be trusted to figure it out on your own.
The compositional approach is immediately ruled out by certain standard ways
of presenting the theory of speech acts. For example, Searle (1965) states that all of
the following have the same semantic content, the proposition that John will leave
the room (his (1)–(5)):

(34) a. Will John leave the room?

b. John will leave the room.
c. John, leave the room!
d. Would that John left the room.
e. If [John will leave the room], I will leave also.

Anyone who wishes to pursue the compositional approach will have to deny Searle’s
assumption that all of the sentences in (34) have the same content; most likely, she
will say at least that (34a–c) differ in their narrow semantics.
Two early proponents of the compositional approach are Hausser (1980) and
Pendlebury (1986). Their proposals are somewhat similar, and we will illustrate
their potential by examining Hausser’s. He proposes that declaratives, imperatives,
and interrogatives have distinct semantic types: declaratives denote propositions,
imperatives properties of the hearer, and interrogatives properties of possible
answers, and he hypothesizes that this is all there is to the semantics of clause types.
From one side, he argues that there are no grammatical features which could be
interpreted as force-indicating operators, and so it would be unduly abstract to
introduce them. From the other, he argues that it suffices to differentiate sentence
moods by their semantic types. He then makes the proposal that the conversa-
tional uses of each sentence can be determined on the basis of its semantics.
The implication seems to be that the semantic type of each clause type strongly
influences the uses to which sentences of that type can be put, and in this way
he raises the idea that the contribution of grammatical structure to conversational
function is exhausted by its contribution to narrow semantic content of the kind
he proposes for each clause type. A sentence’s clause type (and more generally, its
syntactic form) only affects its conversational use indirectly, through the mediation
of semantics. There are no force-indicating operators or constructions which
introduce force into sentences’ meanings.
Although Hausser thinks that semantic type can explain the conversational
uses to which declarative, interrogative, and imperative clauses are put, he does

i i

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

i i

154 sentence mood

not present a theory of how it does so. What is lacking in his version of the
compositional approach is a specification of what conversational uses are and
what principles lead from denotations in the types mentioned above to the uses
that we see. In particular, we would want an explanation of sentential force:
why declaratives are characteristically used to assert the proposition they denote,
why imperatives are characteristically used to direct the addressee to have the
property they denote, and why interrogatives are characteristically used to ask for
an assertion of the true answer. Later work in the compositional approach attempts
to answer these questions.
Two other early advocates of the compositional approach are Huntley (1984) and
Wilson and Sperber (1988). Focusing on the analysis of imperatives and the ways
they differ from declaratives, Huntley argues that both types have propositions
as their denotations, but that these propositions have different characteristics.
The propositions denoted by declaratives (and other indicatives) involve indexical
reference to the actual world (or another world of evaluation), while those denoted
by imperatives (and other non-indicatives, in particular infinitives) do not. His
intuition is that both declaratives and imperatives are evaluated as true or false at
a world, but only declaratives indexically identify the world at which they are to be
evaluated. Imperatives lack indexical reference to a world of evaluation. See Section
2.2.2 for some discussion of Huntley’s ideas about the semantic difference between
indicatives and non-indicatives.
However we evaluate Huntley’s ideas about that contrast, he does not present any
concrete ideas about how the difference in indexicality between declaratives and
imperatives explains their difference in conversational function. In fact, it may be
reading a bit too much into his discussion to see it as advocating the compositional
approach as I’ve defined it here. Still, comments like the following suggest that he
favors the compositional approach:

I have identified a minimal respect in which declaratives and imperatives differ semantically,
but it remains to be seen if the particular proposal I have offered is compatible with an
adequate pragmatic theory of the use in context of these different sentence types.
(Huntley 1984, p. 131)

Wilson and Sperber (1988) understand Huntley’s proposal as following what we

have called the compositional approach but argue that the semantic values he
proposes for declaratives and imperatives cannot explain their conversational
functions. They focus in particular on the fact that Huntley assigns the same
non-deictic denotation to infinitives as he does to imperatives, pointing out that
infinitives are not as tightly tied to directive uses as imperatives are:

 Infinitives often do receive a directive, imperative-like force, as noted by Palmer (), for

example. Sperber and Wilson’s argument is weakened by the fact that the infinitive and imperative have
subjects with different properties; the directive meaning of the imperative might also be connected to
its second person subject.

i i

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

i i

sentence mood, clause type, and sentential force 155

(35) a. To meet the president of the United States. Hmm! (Wilson and Sperber
1988, (9b))
b. Meet the president of the United States!

Sperber and Wilson argue that imperatives describe worlds not merely as possible,
but as achievable and desirable. More precisely, imperatives describe states of affairs
in worlds which the speaker regards both as achievable and as desirable to some
individual z. (Different uses of imperatives are distinguished by the identity of z. In
requests, orders, and other cases, z = the speaker, while in advice and permission,
z = the addressee.) On the basis of this meaning, the force of an imperative
should be pragmatically derived “by manifest contextual assumptions” using the
ideas of Relevance Theory (Sperber and Wilson 1986). Unfortunately, they do not
give a detailed account of this derivation or the semantics of imperatives which
underlies it.
The compositional approach has been taken up again in recent years by such
authors as Portner (2004) and Lohnstein (2007). Portner makes a set of semantic
assumptions very similar to Hausser’s (1980), but pairs it with a dynamic prag-
matics model of conversational update. He proposes an articulated model of the
discourse context which extends Stalnaker’s common ground with two additional
components, the question set (based on Roberts 2012) and the to-do list function:

1. Common ground: a set of propositions

2. Question set: a set of questions
3. To-do list function: a function from participants in the conversation to to-do
lists. The to-do list of an individual a is a set of properties restricted to a.

We will examine this type of model of context in more detail in Section 3.3.
For now it is enough to see how he uses the framework to give an explanation
for the assignment of sentential force. Because declaratives denote propositions,
they are naturally used to update the common ground. This makes the operation
of adding to the common ground, assertion, the sentential force of declaratives.
Likewise, questions are naturally used to update the question set and imperatives
are naturally used to update the to-do list of the addressee. Generally, the sentential
force of a sentence S is the update operation Cx ∪{ [[ S ]] }, for that component Cx
of the context which is a set of semantic values matching [[ S ]] in type (and in the
case of imperatives, domain-restriction). Portner’s analysis shows that the com-
positional approach can assign appropriate sentential forces to root occurrences
of the basic clause types; he does not, however, consider Wilson and Sperber’s
puzzle with (35) or the considerations which led other scholars (like Wechsler 1991;

 Recently, Jary and Kissine () have presented a DRT analysis of the semantics of imperatives

closely related to Sperber and Wilson’s view. On their theory, imperatives introduce “potentialities”
which cannot be evaluated for truth in the main DRS, and directive force results “from pragmatic
considerations: the hearer seeks to identify the point of a nonassertoric utterance which presents him
as the agent of an action, and a reasonable hypothesis is that the utterance is offered as a reason for him
to take that action.”

i i

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

i i

156 sentence mood

Gärtner 2002; Truckenbrodt 2006a) to propose that certain embedded clauses are
assigned sentential force.
Lohnstein (2007) addresses some of these issues left open by Portner while
developing the compositional approach in a slightly different direction. Building
on the tradition of research on German V2, he assigns importance to several
grammatical factors: the presence or absence of a [+wh] or [−wh] element in
CP, the presence or absence of a tensed verb in C, and the mood of the tensed
verb. The role of [±wh] is to differentiate declarative and interrogative clauses.
Lohnstein proposes that the denotation of a yes/no question is the semantic core
for both of these clause types. A yes/no question is a bipartition of the context set
C, as in (36):

(36) a. Is it raining?
b. {C ∩ {w : it’s raining in w}, C ∩ {w : it’s not raining in w}}

From (36), a [−wh] element reduces the bipartition to a singleton, so that declar-
atives denote propositions, while a [+wh] word differentiates the partition, so that
each possible answer to the question is represented by a separate element of the
partition.
Lohnstein’s ideas about the imperative are not as fully developed. He proposes
that there are two “conversational backgrounds,” the factive and the epistemic.
Indicatives and some subjunctives are restricted to the epistemic background,
while imperatives and other subjunctives are restricted to the factive background. It
is not clear, though, what role this distinction plays in explaining the conversational
functions of the clause types.
The position of the finite verb also plays an important role in Lohnstein’s theory.
Only when the finite verb is in C should the semantic object denoted by the
clause be “anchored” to the discourse context; otherwise it should be anchored to
the grammatical context. This is a way of addressing the issue raised by Wilson
and Sperber (1988) with regard to (35)—infinitives would not be anchored to the
discourse context, and so would not have a sentential force.
It is my judgment that the compositional approach is best suited to explaining
the relation between syntax and sentential force. Theories based on a force-
indicating operator find it difficult to explain the diversity of structures which
constitute some sentence moods. Operators and features in C may be important
in the analysis of clause types, because clause types are syntactically coherent, but

 In Portner (), I present a theory which aims to explain the connections between infinitives

and imperatives, but not the absence of directive meaning in examples like (a).
 More precisely, Lohnstein builds on Groenendijk and Stokhof () and treats the partition as

index-dependent: {C ∩ {w : the fact about whether it’s raining in w is the same as it is in v}, C ∩ {w : the
fact about whether it’s raining in w is not the same as it is in v}}.
 Of course, other principles might determine a function for phrases which would not get a

sentential force in the normal way. For example, Lohnstein might say that root V-final clauses in German
cannot have a sentential force, since the principles of sentential force assignment do not apply, but
nevertheless can be used to perform an illocutionary act with the help of Gricean reasoning or language-
specific convention. Portner might say something similar about root infinitives in English.

i i

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

i i

sentence mood in speech act theory 157

a good analysis of a clause type does not amount to a full analysis of sentence
mood except in very simple cases. The construction-based approach also seems
unable to explain why particular constructional patterns are associated with the
sentential forces they are (unless it is combined with the operator or compositional
approach). The compositional approach explains a sentence’s force in terms of its
semantic content, and its semantic content in terms of standard compositional
principles. In this way, it promises to explain why a given range of sentences have
the same force: they have the same force if they derive meanings which are similar
in the relevant ways.

3.2 Sentence mood in speech act theory

In Section 1.4.2, we reviewed some of the ideas in speech act theory which have
allowed it to serve as an important basis for thinking about the nature of sentence
mood. As we now trace some of the more linguistically informed uses of speech
act theory in the analysis of sentence mood, we will notice that there is no
one simple and direct way to apply even the most basic tenets of speech act
theory to the grammatical categories of clause types and sentence moods or the
semantic/pragmatic notion of sentential force. And, it must be said, the literature
on the relation between speech acts and grammar is quite extensive, going far
beyond what I can review in a book of this one’s scope; if we were to attempt a more
thorough study of sentence mood in speech act theory, we would find even more
diversity in approaches, hypotheses, and assumptions. My goal in this section is
to highlight some of the ideas which are most important for understanding recent
research with the aim of developing a better understanding of sentence mood.
As we saw in Chapter 1, the ideas of speech act theory have become so funda-
mental to thinking about information flow in discourse that it is almost impossible
to talk about sentence mood without invoking at least its most basic concepts like
“speech act” and “illocutionary force.” In the interests of keeping the discussion
organized, however, we need to have a somewhat narrow definition of what it
means to use speech act theory as the basis for an analysis of sentence mood. In
this section, we are going to discuss approaches to sentence mood which endorse
the following claims:

1. The logical form of a root sentence used to perform a speech act is F(c), with
F the illocutionary force and c the propositional content.
2. Except possibly in the case of explicit performatives, the identity of F is
constrained by the sentence’s clause type.

 As this book was being completed, I became aware of some new papers advocating the com-

positional approach (Farkas and Roelofsen ; Roberts, to appear). Farkas and Roelofsen’s work
is developed in an inquisitive semantics framework. Roberts’ adopts the property-based analysis of
imperatives originating with Hausser and Portner. Their ideas about the relation between clause type
and sentential force appear to be quite similar to those of the authors discussed above.

i i

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

i i

158 sentence mood

3. The nature of illocutionary force is to be explained in terms of conventions

which allow speakers to perform social acts when the felicity conditions for
those acts are met.

Notice the first point concerns the logical form; it says that the fully specified
meaning of a root clause should be decomposed into a force and a content. It does
not necessarily imply that these two components are independently represented in
the syntax, even at an abstract level like deep structure or Logical Form. The second
point is meant to exclude work on speech act theory which does not hold enough
concern for its relation to grammar to be considered a theory of sentence mood.
The third point serves to distinguish analyses based on speech act theory from
certain closely related dynamic theories of sentence mood. Specifically, a dynamic
semantics or dynamic pragmatics theory which analyses sentential force as an
update function defined in terms of a formal model of discourse will not count as
being based on speech act theory (though it might owe much to speech act theory),
because the update functions are treated primarily as formally defined meanings,
rather than as social acts. Many analyses are close to the line between speech act
theories and dynamic theories, for example Charlow (2011), Starr (2013), and Krifka
(2014). We should not see this line as a border, but rather as an organizational
principle.

3.2.1 The performative hypothesis

The first attempt to use speech act theory as the basis for a linguistically-oriented
theory of sentence mood was based on the idea that the underlying syntactic
form of root sentences contains a marker of illocutionary force. This underlying
marker would both encode the illocutionary force of the sentence and drive the
transformational processes which lead to declaratives, interrogatives, and imper-
atives having distinct surface forms. An early statement of this idea, from which
much generative work on the topic was derived, is due to Katz and Postal (1964).
Katz and Postal propose that interrogative and imperative sentences contain the
abstract markers Q and I in deep structure. In meaning, these markers are meant
to correspond closely to the highest clause of explicit performatives, I ask and
I request. We can thus say that, according to this theory, the sentential forces
of interrogatives and imperatives are represented in logical form and encoded
in syntax (in this framework, deep structure). In contrast to interrogatives and
imperatives, declaratives do not contain an “assertion” prefix, and it is not clear
how their sentential force is derived.
Katz and Postal’s ideas developed into an important approach to sentence mood.
The performative hypothesis states that the deep structures of root sentences
contain, as their highest clause, material very close in syntax and semantics to the

 Based on their professed allegiances, I treat Charlow () and Krifka () as followers of

speech act theory, but Starr () as an adherent of the dynamic approach.

i i

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

i i

sentence mood in speech act theory 159

highest clauses of explicit performatives. Thus, (37) illustrates the deep structures of
declarative, interrogative, and imperative sentences according to Sadock (1974).

(37) a. You will sing.

[S I [declare you [you will sing]]]
b. Will you sing?
[S I [ask you [you will sing]]]
c. Sing!
[S I [request you [you will sing]]]

Compared to Katz and Postal’s analysis, what’s crucially new in (37) is that the force
markers are built from a performative verb and its arguments, rather than each
being an atomic mood marker, operator, or syntactic feature.
The performative hypothesis has been criticized on various grounds, but for our
purposes it is most important to recognize two. First, the hypothesis accepts the
claim of classical speech act theory that an explicit performative has the force
indicated by its highest clause and the content denoted by the complement of
the performative verb. As a result, it must either treat the performative verb and
its arguments as contributing to regular semantics (and then derive the division
between the force and content of the speech act from additional principles) or
treat that material as having a special, force-indicating meaning. Neither alternative
turns out to be easy to defend. And second, it identifies the meanings (at both
the semantic and speech act levels) of simple declaratives, interrogatives, and im-
peratives with the meanings of their performative paraphrases (perhaps adjusting
slightly for the difference in meaning between declare and declare). But the two
do show differences in meaning. For example, (38a) is true if and only if it is raining,
but its performative paraphrase (38b) is true iff the speaker declares that it rains.
(Sadock 2004 discusses this example.)

(38) a. It is raining.
b. I declare that it is raining.

The performative hypothesis does not make it easier to solve these problems.
See Boër and Lycan (1980), Levinson (1980, 1983), and Sadock (1985, 2004) and
references therein for detailed discussion of these and many other important issues
regarding the performative hypothesis.
The performative hypothesis is certainly no longer pursued as actively as it once
was. As a syntactic hypothesis, it is associated with the now-abandoned Generative
Semantics model. However, some of the main ideas of the hypothesis continue
to be pursued. For example, Condoravdi and Lauer (2011) aim to develop an
analysis of explicit performative sentences like (38b) which treats them as normal

 These trees are Sadock’s initial presentation of his ideas; the analysis becomes more sophisticated

later on. Sadock’s proposal is used here to represent a range of Generative Semantics theories which
endorse the performative hypothesis.

i i

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

i i

160 sentence mood

declaratives. Based on this, Condoravdi and Lauer (2012) propose a semantics

for imperatives which they present in terms of an operator imp with the same
basic features as performative verbs like order and request. This analysis can
be seen as an updated performative hypothesis. (It is important to be aware,
though, that at the end of their paper Condoravdi and Lauer (2012) suggest a
different account which would be better classified as a version of the dynamic
force hypothesis (see Section 3.2.3 below). It seems that Condoravdi and Lauer
are not committed to a particular approach to sentence mood, and consider the
performative operator imp to merely provide a convenient way to state some other
ideas about imperatives.)
Whether honestly defeated or simply superseded, scholars no longer treat the
performative hypothesis as an important position on the nature of sentence mood.
We do often see, however, remarks that a particular theory is reminiscent of the
performative hypothesis. Scholars today look back on the performative hypothesis
as an admirably explicit presentation, within the syntactic theory of its day, of
three key claims which are still very much with us today. Specifically, we should
keep in mind that the following ideas were introduced or developed to new
levels of explicitness as parts of the overarching performative hypothesis: (i) the
claim that the logical forms of speech act theory are in fact explicit at some
level of syntax, (ii) the assumption that the major clause types have characteristic
underlying syntactic structures and logical forms, and (iii) the hypothesis that the
force-indicating component of the grammatical representation may be articulated
into a predicate–argument structure with distinct parts corresponding to the
speaker, the addressee, and the force itself. The fact that these ideas are still being
pursued within linguistics thirty years after the last days of the performative
hypothesis shows how important a project it was in the history of syntactic and
semantic theory.

3.2.2 Adjustments to classical speech act theory

These days, it is a common assumption among linguists that speech act theory
offers support for a view of sentence mood according to which the grammatical
forms of root sentences contain a marker of illocutionary force which combines
with a phrase denoting the propositional content to which it applies, but in fact
during the period when speech acts were a topic of active collaboration among
linguists and philosophers, it was not taken for granted that there was such a simple
relation between the force-content logical form of an illocutionary act and the
syntactic structure. Rather, alongside the analysis of Katz and Postal (1964) and the
performative hypothesis, which did roughly put the F(c) logical form into syntax,
we see many scholars whose thinking is more closely connected to the traditional
view of sentence types. These scholars, prominently including Searle, understand
the logical form’s force-marker to be represented in the grammatical structure in
diverse ways, sometimes explicitly (as with explicit performatives) and sometimes
inexplicitly through a pattern of grammatical properties. This understanding of the
relation between form and force is captured quite well by Searle’s term “function
indicating device”:

i i

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

i i

sentence mood in speech act theory 161

Function indicating devices in English include word order, stress, intonation contour,
punctuation, the mood of the verb, and finally a set of so-called performative verbs: I may
indicate the kind of illocutionary act I am performing by beginning the sentence with
‘I apologize’, ‘I warn’, ‘I state’, etc. Often in actual speech situations the context will make
it clear what the illocutionary force of the utterance is, without its being necessary to invoke
the appropriate function indicating device. (Searle 1965, p.6)

Searle’s understanding of the grammar of function indicating devices has much

in common with the traditional conception of sentence types. A wide range of
morphological, syntactic, phonological, and orthographic properties can serve as
function indicating devices, and no theory is given for why particular properties
mark the particular functions they do.
The central problem which arises as one tries to develop a theory of sentence
mood in speech act theory is to resolve the tension between two facts: (i) There are
very many illocutionary forces; and (ii) there is a small set of grammatical patterns,
namely the clause types, which should surely qualify as function indicating devices.
It is impossible to say that each clause type indicates just one illocutionary force, so
how can we define their force-indicating function? There are two ways to answer
this question.

1. Illocutionary forces are decomposed into more primitive components of

meaning, and the clause type indicates one of these more primitive com-
ponents. This would imply, in our terminology, that sentential forces are
not to be identified with illocutionary forces, but rather with components of
illocutionary forces.
2. Illocutionary forces are grouped into natural classes, and the clause types
indicate a class rather than a specific force. On this view, a sentential force
is a partial specification of an illocutionary force.

In the remainder of this section, we will look at some of the important work which
follows one or the other of these strategies.

Searle and Vanderveken’s compositional analysis of illocutionary acts. We have

already discussed the theory of Searle and Vanderveken (Searle 1975b; Searle
and Vanderveken 1985; Vanderveken 1990, 1991) in Section 1.4.2, but it may be
helpful to review the main points relevant to sentence mood here. They develop
a compositional analysis of illocutionary force according to which the overall force
is built from six properties: illocutionary point, mode of achievement, degree of
strength, propositional content conditions, preparatory conditions, and sincer-
ity conditions. Each of these can be determined by different properties of the
sentence, or may even be unspecified, in which case they will receive a default
or inferred value. For example, (39b) is associated with the illocutionary force
of answering-and-lamenting, with alas introducing the sincerity conditions of a
lamentation and yes the preparatory conditions of an answer (see Vanderveken
1990, pp. 16–17):

i i

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

i i

162 sentence mood

(39) a. Are you quitting?

b. Alas, yes, I’m quitting.

Thus, this analysis assumes that the relation between grammatical form and
sentential force is complex and often inexplicit.
Searle and Vanderveken have more to say about sentence mood. The composi-
tional analysis of force allows them to identify simpler components of force which
can be associated with grammatical form. From the perspective of this framework,
the clause types look like this:

1. Basic types
(a) Declaratives have the assertive point, except for explicit performatives,
which have the declarative point.
(b) Imperatives have the directive point.
(c) Interrogatives have the directive point plus special propositional content
conditions.
2. Minor types
(a) Exclamatives have the expressive point.
(b) Promissives have the commissive point.

What exactly are the principles which give rise to these associations? Vanderveken
(1990, p. 108) states that the declarative, imperative, and exclamative sentence
moods “express” the assertive, directive, and expressive illocutionary points, re-
spectively. He argues that in explicit performatives, the combination of declarative
sentence type and additional markers (such as hereby) expresses the declarative
point. While it might seem strange that the grammatical pattern which expresses
one point is contained within the pattern which expresses another, he justifies this
unusual situation by claiming that declarations “contain” assertions with the same
propositional content. Interrogatives are an even more difficult case. As in classical
speech act theory, asking a question is assumed to be a type of directive act, and
one might expect that the two clause types which express directive acts, imperatives
and interrogatives, would have some morphosyntactic similarity. Yet interrogatives
and imperatives each have more in common with declaratives than they do with
each other. I do not know of any specific statement within this framework about
which grammatical features express the directive point of interrogatives or the
other components of interrogative force.
Zaefferer (1983, 2001) argues that the classification of asking acts as direct-
ive poses a serious problem for the theory of speech acts developed by Searle
and Vanderveken. In addition, he points out several other similar discrep-
ancies between the typology of speech acts assumed by this framework and

 A similar perspective is developed by Allan (). Allan proposes that each clause type is

associated with a particular primary illocution as part of its semantics, and that the primary illocution
serves as a force-indicating device. It seems that Allan in effect accepts the literal force hypothesis, but
it remains somewhat unclear what the relation is between primary illocution and illocutionary force.
Allan’s theory also has some commonalities with that of Bach and Harnish ().

i i

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

i i

sentence mood in speech act theory 163

grammatically defined classes of sentences. An important one concerns the status

of commissives. Even though commissives form a basic category in Searle and
Vanderveken’s framework (as they do in classical speech act theory), very few
languages have promissives, the clause type specialized for the commissive point. If
the illocutionary points are really the most fundamental classification of conversa-
tional functions relevant to grammar, we might expect that each such point would
normally (or at least often) have its own, equally fundamental grammatical form.
Zaefferer argues that commissive declaratives should be classified fundamentally
as assertions, while true commissive promissives are distinct.
Vanderveken (1990, p. 109) addresses the issue of why promissives are rare by
boldly stating that the commissive point “is less important than the other points
for the purpose of linguistic communication.” He suggests that our inexplicit way
of normally marking commissives may actually be an advantage for speakers, since
they can then more easily manipulate their interlocutors by pretending not to
have actually made a promise when convenient. What this diagnosis leaves out, of
course, is that for every promiser, there is a promisee, who is also a participant in the
linguistic community; it can be very important for one who needs a commitment
from another to know that the commitment is without a doubt on the record.
Similarly, Searle (2001, p. 287) states without qualification that “commissives are
rather uncommon in ordinary speech.” In any case, Searle’s and Vanderveken’s
counterarguments here are really no more than speculation. We are not given any
social theory or empirical data supporting their claims about the importance or
frequency of commissives compared to other illocutionary acts.
I think we can evaluate this exchange between Zaefferer and Searle and
Vanderveken as follows. The grammatical evidence from interrogatives and com-
missives suggests that, if speech act theory is to be the basis of a theory of sentence
mood, we will need a different classification of act types from the one provided
by Searle and Vanderveken. Zaefferer thinks that speech act theory should serve
as the basis for the theory of sentence mood, and so he sets out to change it; we
will look briefly at his proposals below. Searle and Vanderveken think that speech
act theory provides the background for the theory of sentence mood, but that
the typology of sentence moods is not derived from the typology of illocutionary
acts in a very direct way. For them, illocutionary point is sometimes central to
explaining sentence mood and sometimes not. A third response, of course, is to
look for a theory of information flow in discourse which allows us to better explain
the properties of sentence mood and clause types. We will see how this perspective
plays out in Section 3.3.

 He also makes some philosophical objections to criteria by which Searle and Vanderveken classify

illocutionary acts, but Searle counters these (Searle ), and in any case, Zaefferer’s comments are too
brief to be evaluated effectively.
 He also says that “it is a rather big deal to make a promise, and consequently you do not have

common separate devices for making promises.” I do not understand why more important acts would
be less likely to be explicitly marked.

i i

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

i i

164 sentence mood

Rett on exclamative force. Rett (2011) presents an analysis of exclamations

based on assumptions about illocutionary force which could easily be integrated
with Vanderveken’s views. It is useful to study her proposal with the goal of
understanding how recent versions of speech act theory can be used as part of the
analysis of sentence mood. Rett discusses the similarities and differences between
sentence exclamations and exclamatives (her (40)):

(40) a. (Wow), John arrived early!

b. How (very) early John arrived!

Exclamations like (40a), in Rett’s terminology, are sentences with the morphosyn-
tax of a declarative but special intonation associated with the illocutionary force
of exclamation. Exclamatives like (40b), in contrast, form a separate clause type
from declaratives and interrogatives and have an illocutionary force very similar to
exclamations. The essence of Rett’s analysis is that exclamations and exclamatives
share the same illocutionary force, represented by the illocutionary operator
e-force, but that they differ in the details of the propositional content which the
operator takes as its argument. She argues that exclamatives introduce a predication
involving degrees, while exclamations do not. In the case of (40), the difference is
that (40a) expresses that John wasn’t expected to arrive early (but asserts that he
did), while (40b) expresses that the degree of earliness of John’s arrival exceeds the
speaker’s expectation (and asserts nothing).
For the purposes of understanding Rett’s views about sentence mood, what’s
important are her comments on E-force. The first thing to note is that it is not quite
clear whether exclamations and exclamatives constitute a single sentence mood.
We have defined sentence mood in terms of the association between grammatical
form and sentential force, and (40a,b) share the illocutionary force introduced
by E-force. So, in this respect, they have the same sentential force and might be
regarded as one sentence mood. But Rett proposes that the exclamation (40a) is
also associated with the function of asserting that John arrived early, and so it can
be seen as sharing its sentential force with ordinary declaratives. As we will see
below, Rett proposes that the assertive force of (40a) follows as a kind of entailment,
and so from our perspective, the right conclusion is probably that the examples in
(40) belong to the same sentence mood.
A second important point to observe about Rett’s theory is that she does
not associate the illocutionary force operator E-force with any specific piece of
grammatical structure. She only states that this operator “models” the force of
exclamation. From the context of her discussion, we can infer that she thinks
that the meaning of (40a) involves E-force because of its characteristic intonation
(indicated in writing by the exclamation point). With regard to (40b), it is not
clear whether Rett associates E-force with its intonation or its grammatical form.
Overall, it seems that Rett thinks that illocutionary force is sometimes inexplicitly
represented by grammatical form, but her discussion leaves open the possibility

 Gutiérrez-Rexach () sketches an analysis with many similarities to Rett’s. See also Castroviejo

(b) for discussion of Gutiérrez-Rexach’s view.

i i

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

i i

sentence mood in speech act theory 165

that syntactically defined clause types play a very important role in the determin-
ation of illocutionary force. Such a perspective on the relation between form and
force is completely compatible with Searle and Vanderveken’s views.
Now we can look at the meaning which Rett assigns to E-force (Rett 2011, p. 429):

(41) E-force(p), uttered by sC , is appropriate in a context C if p is salient and true

in wC . When appropriate, E-force(p) counts as an expression that sC had not
expected that p.

The definition (41) begins with an appropriateness condition which could as

well have been called a preparatory or sincerity condition. Then, in saying that
E-force(p) counts as an expression that sC had not expected that p, Rett identifies
its illocutionary point as Vanderveken’s expressive point. Therefore, as far as I can
tell, Rett’s analysis should be seen as operating within Vanderveken’s (1990) theory
of speech acts.
Although E-force represents the illocutionary force of both exclamations and
exclamatives, it has a somewhat different effect in the two cases, and she traces
this difference to a contrast between the semantic values of the two types. With
an exclamation, the semantic value of the clause is an ordinary proposition; for
example, the propositional content of (40a) is that John arrived early. When E-force
takes this proposition as argument, we have a speech act which is only appropriate if
John actually arrived early, and which expresses that the speaker did not expect that
he would arrive early. The hearer can normally infer that the speaker expected that
John would not arrive early (although the theory suggests that a weaker inference
would be good enough, that is, that the speaker didn’t have any expectation).
Rett also states that the appropriateness condition implies that the speech act of
exclamation entails a speech act of assertion. In the case of (40a), the point is that
E-force(p) illocutionarily entails (in the sense of Vanderveken 1990) an assertion
that John arrived early. Therefore, the fact that (40a) asserts that John was early
is a byproduct of the felicity condition on exclamations, and not due to its having
declarative sentence mood.
Turning to the exclamative (40b), Rett proposes that its content is not funda-
mentally propositional. Rather, its semantic denotation is a property of degrees:
λd[early(j, d)]. As a result of its having this kind of denotation, E-force produces
a speech act in which the speaker expresses an attitude towards a degree; in the
case of (40b), the speaker’s use of the sentence counts as expressing that, for
some degree of earliness d, he did not expect that John would be d-early. (See
also Katz 2005; Castroviejo 2006b for discussion of attitudes towards a degree.)
In my opinion, the steps which lead to this expressive act are not motivated in a
clear way: she says that context provides a degree argument for an exclamative’s
denotation, and that the resulting proposition (e.g. early(j, d )) is the argument of
E-force and determines the speech act’s appropriateness condition; in turn, because
of the unbound degree variable in the argument of E-force (here d ), the precise

 The inference that Rett uses “expresses” to point to Vanderveken’s definition of the expressive point

is supported by her discussion of Vanderveken’s () treatment of lamentations.

i i

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

i i

166 sentence mood

attitude it expresses is adjusted to the one noted above. Rett also suggests that
the unbound variable is responsible for the fact that exclamatives, in contrast to
declarative-like exclamations, do not assert anything—although she admits that
she does not have a clear idea for why this would be ((40b) should illocutionarily
entail the assertion that John arrived d -early).
Summing up, the most important thing to observe about Rett’s paper is that
she develops a sophisticated analysis of exclamatives and exclamations within the
general framework outlined by Searle and Vanderveken. In comparison to Searle
and Vanderveken’s own discussion, her proposals are grounded more rigorously
in linguistic facts and are embedded in a more detailed theory of syntax and
semantics. Therefore, Rett’s work provides a good illustration of the use to which
the notion of illocutionary point could be put within the theory of sentence mood.
Conversely, some of the problems we identify with Rett’s analysis seem to be
derived from the version of speech act theory she adopts. Most significantly from
our perspective, Rett does not give principles for deriving logical forms containing
E-force from the syntax of exclamatives and exclamations. We do not know, in the
end, why the exclamation (40a) is not assigned the force of assertion, normally
associated with its declarative clause type, and gets E-force instead. Nor do we see
a role for the special syntax of the exclamative (40b) in deriving its logical form
with E-force. Of course, one can imagine ways of addressing these issues which
are compatible with Rett’s analysis, but it is an open question whether any of them
could work as a general theory of sentence mood.

Zaefferer’s alternative classification of illocutionary acts. Alongside the critique

of the versions of speech act theory developed by Searle and Vanderveken men-
tioned above, Zaefferer (1981, 1983, 1986, 1998, 2001, 2006b,a, 2007) attempts to
radically revise the classification of illocutionary acts to meet the concerns he
raises. His ideas are worth examining because they give a sense of how a version
of speech act theory based more closely on linguistic data would go, but the
new classification he proposes is quite complex and not motivated in sufficient
detail to be considered a theory of sentence mood. In one paper (Zaefferer 2007),
for instance, he classifies acts according to the following criteria, among others:
structuredness (whether they have grammatical structure), telicity, epistemicness
(whether they have to do with an agent’s knowledge of the propositional content),
holophoricity (whether the semantics of the sentence refers to the illocutionary act
itself), and opacity (whether it aims at a partially specified propositional content).
Some of the more important sentence moods are treated as follows:

1. He has a separate category for question acts, the “structured opaque epistemic
telic acts.” Question acts are distinguished from representatives, commissives,
and declarations in virtue of the property of opacity. The idea here seems
to be that a question word in an interrogative introduces a variable or

 One would think that the sentence expresses that the speaker did not expect John to arrive d -early,

rather than what Rett actually proposes, that there is some degree such that the speaker didn’t expect
John to be that early.

i i

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

i i

sentence mood in speech act theory 167

placeholder into the semantics, so that the sentence’s propositional content

is only partially specified.
2. Directives, promissives, and exhortatives form a class of “structured direct
telic” acts. The members of this class are distinguished from the other major
types (declaratives and interrogatives) by virtue of the property of non-
epistemicness (what Zaefferer calls “direct”). The idea is that in an imperative,
what is important is that the propositional content of the sentence become
true, whereas with a declarative, what is important is that someone believe it
to be true.
3. Despite the fact that they seem to perform the same type of illocutionary
act, promissives and declarative-like commissives are separated into two
categories. The former are grouped with directives (as above), while the latter
are grouped with representatives.
4. Expressive acts appear to be split among three categories (Holistic,
i.e. non-structured (Wow!), Structured-Epistemic-Telic-Holophoric-Social-
Intralinguistic-Institutive (Thanks!), and Structured-Representative-Atelic-
Hybrid (I’m not sure what type of case Zaefferer has in mind)).

There are several serious problems with Zaefferer’s ideas, though to some extent
these may be due to the fact that they were presented in only a few short papers. As
can be observed, speech acts are classified in terms of a great variety of properties,
which combine or fail to combine in mysterious ways. (For example, there are
no opaque atelic acts, opaque holophoric acts, or atelic epistemic acts.) More
importantly, the crucial properties are not clearly defined and are not integrated
into a broader theory of communication and meaning.
Zaefferer’s alternative classification of speech acts has not had much influence on
later research, but it is worth thinking about because it exemplifies one reasonable
approach towards developing an explanatory theory of sentence mood. Zaefferer
attempts to use the typology of sentence moods and clause types to provide
guidance for a new classification of illocutionary acts. A novel classification of
illocutionary acts will require, of course, a whole new theory of speech acts. In
other words, the advocate of a new typology of illocutionary acts will need to go
back to the foundations of speech act theory to see to it that the whole apparatus
still functions as well as it does in the classical system systematized by Searle. The
essential point of Searle’s (2001) response to Zaefferer is that Zaefferer fails to do
this and so fails to appreciate the full motivation for the classical typology.

Bach and Harnish. Searle and Vanderveken’s ideas about the relation between
grammatical form and sentence mood are naturally linked with either the operator-
based or construction-based approaches to the interface between syntax and
sentential force. This is so because they seem committed to the idea that the
illocutionary force of an utterance is normally represented (even if inexplicitly)
in its grammatical form. However, speech act theory is also compatible with the
compositional view of the syntax–force interface. Though they may not have
thought of their work in quite this way, the theory of Bach and Harnish (e.g.
Bach and Harnish 1979; Harnish 1994) is better suited to the compositional view

i i

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

i i

168 sentence mood

of the mood–force interface. Here I would like to make a few remarks on what
their theory—which is quite deep and comprehensive in its re-evaluation of central
features of speech act theory, but not at all focused on developing a linguistic theory
that can explain how particular sentences are linked with their sentential forces—
has to say about the nature of sentence mood.
According to Bach and Harnish, the mapping between sentence moods and
illocutionary force happens in two steps. The first concerns a relation between the
grammatical forms and propositional attitudes. They say that each basic clause type
expresses a characteristic kind of propositional attitude. The second step gives a
relation between expressed attitudes and illocutionary forces; they propose that
each type of attitude constrains the range of illocutionary forces that the sentence
can be assigned. For example, according to Harnish (1994, pp. 433–4), the interfaces
work like this:

Step (1): Form–Expressed Attitude

(a) Declaratives
• Form: NP+AUX+VP
• Expressed Attitude: S believes that p.
(b) Imperatives
• Form: (NP)+VP[+imp]
• Expressed Attitude: S desires/intends that H make it the case
that p.
(c) Yes/No Interrogatives
• Form: AUX+NP+VP or NP+AUX+VPrising
• Expressed Attitude: S desires that H tell S whether or not it is the
case that p.
(d) Wh-Interrogatives
• Form: Wh-x+AUX+(NP)+VP or
NP+AUX+[ . . . Wh-/x/ . . . ]VP
• Expressed Attitude: S desires that H tell S Wh-x such that p(x).
Step (2): Expressed Attitude–Illocutionary Force Potential
(a) Declaratives
• Expressed Attitude: S believes that p.
• Illocutionary Force Potential: Any act with the belief that p as a
necessary condition.
(b) Imperatives
• Expressed Attitude: S desires/intends that H make it the case
that p.
• Illocutionary Force Potential: Any act with the desire/intention
that H make it the case that p as a necessary condition.

 The earlier but more well-known joint book (Bach and Harnish ) is quite similar in its overall

perspective on these issues, but not as explicit about its understanding of the nature of clause types.
Note that I have adjusted the notation for grammatical patterns to make it more transparent.

i i

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

i i

sentence mood in speech act theory 169

(c) Yes/No Interrogatives

• Expressed Attitude: S desires that H tell S whether or not it is the
case that p.
• Illocutionary Force Potential: Any act with the desire that H tell
S whether or not it is the case that p as a necessary condition.
(d) Wh-Interrogatives
• Expressed Attitude: S desires that H tell S Wh-x such that p(x).
• Illocutionary Force Potential: Any act with the desire that H tell
S Wh-x such that p(x) as a necessary condition.

Unlike Searle and Vanderveken’s, this analysis does not seek to decompose illocu-
tionary force to find some components which can be shared by all members of a
sentence mood. Rather, it sees the assignment of illocutionary force as the result
of inference, and seeks to identify the constraints which sentence mood places on
the assignment of force as part of that inferential process. As Bach and Harnish see
it, the sentence moods constrain the force which their sentences can be assigned
through the types of attitudes that they express.
Seen this way, Bach and Harnish’s approach to sentence mood seems to exem-
plify the compositional approach to sentence mood. For them, force is not encoded
(either through an operator or a grammatical pattern), but rather inferred on the
basis of an aspect of meaning which is closer to grammar than speech acts are.
However, it is not entirely clear how the expressed attitudes are derived. Looking
at the Form–Attitude relation for imperatives, presumably (NP) + VP[+imp] com-
positionally denotes p, but where does attitude of desire/intention come from? It
could arise from an interpretive rule triggered by the structure (NP) + VP[+imp]
(a sentence type or construction) or inferred from the precise content p. This
second way of thinking would represent a doubling-down on the compositional
approach to sentence mood.

3.2.3 The dynamic force hypothesis in speech act theory

Since the 1980s, the dynamic approach to meaning has increasingly become the
mainstream way of thinking about information flow in discourse within formal
semantics and pragmatics. As a result, it is not surprising that scholars have sought
to combine the insights of that approach with the framework of speech act theory.
As discussed in Section 1.4.3, Gazdar (1981) was a pioneer in this way of thinking;
more recently, we see such authors as Charlow (2011), Kaufmann (2012), and Krifka
(2014) pursuing the idea in different ways. In this section, we will review their
proposals about the nature of sentence mood.

Kaufmann: Illocutionary force potential under the dynamic force hypothesis.

Kaufmann’s discussion builds on the most basic concepts of the dynamic approach
already introduced in Chapter 1, and we begin with it. She suggests we define
illocutionary acts as sequences of contexts. Intuitively, for example, we can think
of a successful act of assertion as a move from a context in which the propositional
content of the act is unsettled, through one in which a speaker says something

i i

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

i i

170 sentence mood

appropriate, to an ending context in which the speaker and addressee agree that
the propositional content is true. More formally, we have the following:

1. Contexts
(a) A context c has four components cS , cA , cT , cW , where cS is the speaker,
cA is the addressee, cT is the time, and cW is the world of c.
(b) From c we can specify various additional parameters, including crucially:
• cE is the “disambiguated LF of the linguistic object uttered” in c (if there
is one).
• CS(c) is the context set of c.
• (c) is the salient decision problem in c, if there is one.
2. Speech acts
(a) A speech act is a description of a sequence of contexts c , c, c , where:
• Felicity conditions of the act are conditions on c .
• The speaker cS utters some linguistic object cE in c. In effect, the
locutionary act happens in c.
• The essential effect of the speech act obtains in c .
(b) Example: assert(φ)
• Both φ and ¬φ are compatible with CS(c ).
• The assumption that cS has reasons to believe φ is compatible with
CS(c ).
• cS says cE in c.
• CS(c ) entails that cs believes φ.

Notice that, in the definition of assertion, no link is drawn between the asserted
proposition φ and the linguistic expression cE which the speaker utters in c. Clearly,
and as Kaufmann notes, we need to incorporate a semantic or pragmatic relation
between cE and φ and to explain the role cE plays in deriving an output context
where the speaker is entailed to believe φ.
Later on in her discussion Kaufmann makes the connections among cE , φ, and
c more explicit in a way very similar to Gazdar’s (1981) proposal. She assumes
a basic force function J which applies to propositions or questions to yield an
update function, that is a function from contexts to contexts. This J is sensitive to
the semantic type of its content argument m, that is, whether it is a proposition or
a question.

1. If m is a proposition, J (m) is an assertive update, mapping c to a c where m

is entailed by the context set of c .
2. If m is a question, J (m) is a partitioning update, mapping c to a c where m
partitions the context set of c .

 Kaufmann gives these definitions in terms of the common ground, rather than the context set, but

I state them in terms of the context set in order to be consistent with portions of her theory presented
above. In her terminology, J is called the “update function,” but in our terminology, it is better referred
to as a force.

i i

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

i i

sentence mood in speech act theory 171

Let us leave aside the nature of the partitioning update. It is derived from the
dynamic approach to questions discussed below (Section 3.3.2). What’s crucial here
is that J is defined in overtly dynamic terms and does its job on the basis of the
semantic type of its argument.
Finally, Kaufmann uses this apparatus to explain what a “theory of speech
acts” consists in. A theory of speech acts is essentially a specification, for every
triple of contexts c , c, c , of one of the speech acts like assert (defined above),
command, permit, request, marry, and so forth. A proper speech act as-
signment must respect J in the sense of relating c and c in the way that J
says: J ([[ cE ]] c )(c) = c . This description of speech act assignment implies that
a sentence which denotes a proposition can only receive an illocutionary force
which involves an assertion-like update, while a sentence which denotes a question
can only receive an illocutionary force which involves an asking-like update. (For
example, though the act performed by I now pronounce you husband and wife is
not an act of assertion, it should be assertion-like in that it results in a common
ground which entails that the speaker did pronounce them husband and wife.)
I think it also implies (whether intentionally or not) that phrases which do not
denote propositions or questions cannot receive any speech act assignment at all.
Kaufmann’s ideas here suggest quite a few consequences for the theory of sen-
tence mood. As noted, the dynamic update determined by J is fully determined by
the semantic type of cE , and hence strongly influenced by (maybe even determined
by) the grammar of cE . In contrast to this specificity, Kaufmann does not present
any comprehensive theory of speech acts. We do not know how diverse she takes
the set of illocutionary acts to be—is it very fine-grained as in classical speech act
theory, or a smaller set? However, she does make heavy use of the acts like assertion,
commanding, and permitting in the analysis of imperatives, and they allow us to
understand further her view of sentence mood.
Kaufmann treats imperatives as semantically equivalent to declarative sentences
with should, for example:

(42) a. Sit down!

b. You should sit down.

While there is a close relation between these forms, they are obviously not se-
mantically identical, and so the puzzle for Kaufmann is to explain how they differ.
Her approach to this puzzle begins with the distinction between the performative
and descriptive uses of (42b). The performative use can be exemplified by a parent
speaking to a child, uttering (42b) with the intention of telling her to sit down.
On this use, it is not naturally described as true or false and is much closer to the
imperative (42a). On the descriptive use, in contrast, (42b) can readily be judged
true or false and can be used to make an assertion about what is in the best interests
of the addressee. The descriptive use is obvious when the subject is third person
(That guy over there should sit down because it looks like he might pass out).
According to Kaufmann, the performative and descriptive uses of the modal
sentence differ in speech act assignment: the performative use might involve an
assignment of command, while the descriptive use might involve assert. What

i i

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

i i

172 sentence mood

makes for the choice between these two are the preconditions of the respective act
types. An act of the command type requires that the speaker cS is an authority on
the truth of the propositional content [[ cE ]] and that [[ cE ]] provides an answer
to some salient open issue about what to do, the “decision problem” (c). For
example, (42a) would answer the child’s decision problem “Do I stand or sit now?”
And, according to Kaufmann, imperatives presuppose that these same conditions
hold. Since an imperative presupposes that the speaker is an authority and that the
imperative answers a decision problem, it is compatible with being assigned
the speech act command. But in contrast, she assumes that the presuppositions of
the imperative make it incompatible with assert as its speech act type. As a result,
imperatives have the pragmatic properties of performative deontic sentences, such
as the property of not being naturally described as true or false.
A worrisome gap in Kaufmann’s analysis is that she does not explain precisely
what condition on assertion is incompatible with the presupposition of imperat-
ives. The framework states that an act of assertion assert(φ) places a condition on
c or c which cannot be met when φ is an imperative. It appears that Kaufmann takes
as crucial the authority condition and the condition that φ must answer (c). But I
do not know why these conditions would rule out assertion. In the case of the child
who needs to decide whether to sit or stand, I see no reason why the existence of
this decision problem, together with the assumption that I will definitely be correct
in my assessment of what she should do, would make it impossible to assert that
she should sit down. This would amount to pointing out the right decision without
imposing it. More generally, the crucial assumption for Kaufmann is that there
is some property P of contexts which allows only a performative use of deontic
modals; she proposes that imperatives presuppose P. But we are not given any
positive argument that there really is any such property.
The issue just raised for Kaufmann’s theory has been presented in a somewhat
abstract way, but we can see its concrete consequences in the following contrast,
from Portner (to appear):

(43) You should not park in the dry cleaner’s lot, because you’ll get a ticket if you
do. So, . . .
a. do not park in the dry cleaner’s lot!
b. ??you should not park in the dry cleaner’s lot!

What is crucial here is that the imperative is felicitous in a context in which a

declarative with should is not. This fact contradicts the proposal that imperatives
essentially are should sentences, just restricted to contexts in which the should is
performative. In (43a), the imperative has a normal imperative function, namely
urging the addressee not to park in the lot, which the modal sentence cannot have
in the same context.
Similarly, von Fintel and Iatridou (2015) point out a related problem for
Kaufmann’s theory. In certain contexts, imperatives can readily be used to give

 Jary and Kissine () critique Kaufmann’s analysis for similar reasons.

i i

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

i i

sentence mood in speech act theory 173

permission or express acquiescence, but the deontic modals lack these uses (von
Fintel and Iatridou 2015, p. 6):

(44) A: May I open the door?

B: Sure, go ahead, open it!
B : Sure, go ahead, #you must open it.
B : Yes, in fact: you must open it!
C: Sure, go ahead, you should open it.

They point out that B can be understood as giving permission without endorsing
the choice to open the door at all, whereas B and C imply that the speaker requires
or endorses the action. If imperatives are equivalent to sentences with must or
should in any context in which they are felicitous, the difference is unexpected.
What does all of this tell us about sentence mood? Within Kaufmann’s theory,
there are three candidates for the status of sentential force: (i) the type of update
potential assigned by J , (ii) a specific force which is most prototypically associated
with J , or (iii) the set of forces compatible with a given sentence mood. If we
think of sentential force as (i), there are only two sentential forces correlating with
the two semantic types of root sentences: the intersective update of propositions
and the partitioning update of questions. From this perspective, imperatives and
declaratives would have the same sentential force, and so belong to the same
sentence mood. The second view (ii) implies that there is some distinguished subset
of triples c , c, c  which are related by a particular force; much of Kaufmann’s
discussion suggests that the sentential force of imperatives is order because
ordering is compatible with all “unmarked” contexts. If sentential force is (iii), we
are treating sentential force as an illocutionary force potential in a sense similar to
Bach and Harnish (1979). It seems to me that (iii) fits best with her overall approach.
I believe that Kaufmann accepts the standard moods of declarative, interrogat-
ive, and imperative, and we can infer the following illocutionary force potentials
from her discussion:

1. The sentential force of declaratives: the potential to be assigned any speech

act which involves adding propositional content to the context set.
2. The sentential force of interrogatives: the potential to be assigned any speech
act which involves partitioning the context set.
3. The sentential force of imperatives: the potential to be assigned any speech act
which involves adding propositional content to the context set and answering
a salient decision problem.

Kaufmann’s approach seems to fit readily with the compositional approach to the
assignment of sentential force. The update potential J (S) is assigned based on
S’s semantic type, while its illocutionary force potential is constrained by both
J (S) and other presuppositions of S. Specifically in the case of greatest interest to

 See especially the discussion in her Sections .–. and ...

i i

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

i i

174 sentence mood

Kaufmann, she hypothesizes that the illocutionary force potential of imperatives is

limited to appropriate options like ordering, permitting, advising, and the like by
its semantic type (proposition) and its presuppositions.

Krifka: The operator-based view under the dynamic force hypothesis. Whereas
Kaufmann develops a theory of sentence mood which relies on an understanding
of sentential force as the class of illocutionary acts which a sentence type can
perform, Krifka (2014) takes up the view that sentential force is explicitly expressed
within syntax. Building on the syntactic analysis of Rizzi (1997), he proposes that
operators such as ASSERT and PERFORM occupy the head of a ForceP projection.
(PERFORM is for explicit performatives. He suggests that there would be addi-
tional operators for questions and commands.) It thus seems that Krifka assumes a
taxonomy of force operators which would identify the three basic sentence moods,
plus one more to account for performatives. Krifka does not provide any new
syntactic arguments which would support the idea that all sentences of a given
sentence mood share a particular syntactic operator, and so would have difficulty
explaining the diverse range of syntactic structures which can realize a single
sentence mood (Section 3.1.3).
Although it takes a familiar approach to sentence mood, Krifka’s analysis is
innovative when it comes to its implementation of the dynamic force hypothesis.
He describes the standard dynamic approach of Stalnaker as giving a picture of
“a dynamic conversation and static world: a sentence changes the common ground
and the set of available discourse referents, but the world and time of the utterance
stays the same” (Krifka 2014, p. 5). In contrast, he wants to develop a theory in
which speech acts change worlds and times: they map indices (world–time pairs)
onto other indices. Krifka believes that such a picture better captures the true
nature of speech acts.
To give an example of how Krifka sees speech acts, we can consider his treatment
of the ASSERT operator. The idea is that ASSERT maps indices in which the
speaker lacks a commitment to the asserted content onto ones in which she has
such a commitment. We begin with a predicate committed which describes an
individual as liable towards another individual for the truth of a proposition.

(45) committed(i, p, x, y) = x is liable toward y for the truth of p at i.

Given (45), the operator ASSERT takes a propositional content p and a context
c and maps indices onto indices. (The following definition simplifies at several
points. As in the discussion of Kaufmann’s theory, cS is the speaker of context c
and cA the addressee; cI is the world–time index.)

(46) [[ ASSERT ]] (p) =

λcλi ιi[i = cI and i differs from cI in that committed(i, p, cS , cA )]

 Krifka calls this predicate assert and the speech act operator ASSERT.

i i

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

i i

sentence mood in speech act theory 175

With regards to an example like It’s raining, the operator (46) assigns a function
which maps a context c onto a function which maps cI onto the index which results
from it when the speaker cS becomes liable towards the addressee cA for the truth
of p. It “changes the index” from the existing one to one in which the speaker is
committed to p in conversation with the addressee.
Krifka calls assertion defined in this way “an index changing device,” but by
itself, a function from indices to indices does not change any indices. We need
some additional principle to account for what happens when an assertion is made
in a particular discourse context. Krifka discusses several options for how to
integrate (46) into a model of conversational update, each cast in terms of an update
operator +. The most interesting attempt looks like this:

(47) For any context C = CS , CA , Ccs : C + A =

CS , CA , {A(C)(i) : i ∈ Ccs }

This definition is a bit subtle, because we have two notions of context in play. In
the previous definition of ASSERT, a context is a triple of a speaker, an addressee,
and an index. We use lower-case c for this type of context. The update function
(47) treats a context as a triple of a speaker CS , an addressee CA , and a context set
Ccs . (Recall that a context set is a set of indices.) I use upper-case C for this new
notion of context. Upper-case contexts differ from lower-case ones in that their
third component is a set of indices, rather than an index.
What definition (47) says is that, when you update a context with A, you adjust
the context set of the context by replacing each index in it with one in which
the commitments specified by A are in place. Notice that the definition does not
say that the propositional content of A is added to the context set. That is, in the
case of ASSERT, the proposition asserted does not automatically become common
ground—only the proposition that the speaker is liable for its truth. In this way, it
is different from most dynamic theories derived from Stalnaker’s work.
As mentioned above, Krifka views his proposal as capturing the idea that speech
acts, as “index-changing devices,” do more than change the state of the conversation
in the way which is standard in Stalnakerian dynamic pragmatics. Krifka’s assertion
operator derives a function (46) which modifies each world–time pair in the
context set Ccs , thereby creating the new context set, rather than by eliminating
indices through an intersective update, as happens in theories more closely based
on Stalnaker’s treatment of assertion. It is useful to ask, though, whether Krifka’s
theory really gives us more than a different implementation of the same procedure
of update. One difference between Krifka’s theory and Stalnaker’s is that Krifka
assumes that the primary effect of assertion is to adjust the speaker’s commitments.
The same update can certainly be expressed in an intersective theory. Suppose
that the context set cs is a set of world–time pairs t, w where the speaker is not
committed to p (at t in w), and that, at time t + 1, the speaker may or may not
be committed to p (that is, there is at least one t, w  ∈ cs where the speaker is

 The contextual update recorded by ASSERT is essentially the secondary effect of assertion in a

theory such as Stalnaker’s. Thanks to Malte Willer for pointing out this way of understanding ASSERT.

i i

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

i i

176 sentence mood

committed to p at t + 1 in w and at least one t, w  ∈ cs where the speaker is not
committed to p at t+1 in w ). The speaker then takes one second to utter a sentence
asserting p. The time index is now t + 1 and the context set only contains pairs
t + 1, w where the speaker is committed to p. The world w has been eliminated
from cs.
It seems that Krifka assumes that the context set does not contain indices
which differ in the way we have described t, w  and t, w  in this discussion.
The index-to-index mapping (46) would not be different from intersective update
if the worlds where the speaker comes to have a commitment to p are already
present in cs. Instead, Krifka implies a view of the context set on which indices
do not represent the presence or absence of discourse commitments unless the
commitment is explictly introduced by a speech act. In this respect, Krifka’s ideas
require a philosophical understanding of possible worlds different from Stalnaker’s.
Indeed, I am not confident that it would even be possible to develop a coherent
concept of possible worlds which does the job Krifka’s analysis needs it to do. It
certainly seems that the context set can contain worlds in which a commitment
may or may not be taken on in the future, since a sentence like (48) can be true:

(48) You may be about to commit yourself to the claim that it’s snowing.

The central idea of Krifka’s analysis could be maintained, however, by enriching the
notion of context or index to contain more than just individuals, worlds, and times.
It should contain some component in addition to the context set which ASSERT
changes to record a discourse commitment. If modified in this way, Krifka’s analysis
would be quite similar to dynamic theories which propose individual commitment
slates in the context (see Section 3.3).

Charlow: Update potential as sentential force under the dynamic force

hypothesis. Charlow (2011) shows another way of using the tools of the dynamic
approach to analyze sentential force within speech act theory. In its broad outlines,
his theory is extremely similar to the pioneering ideas of Gazdar (1981) discussed
in Section 1.4.3. Charlow proposes (2011, p. 80) that:

1. “A force—the semantic denotation of a force operator—is represented as a

function from propositions into functions from states into states.”
2. “A speech act is represented as the result of applying a force to a proposition.”

These definitions imply that a speech act is a function from states into states,
and except for its use of (cognitive) states in the definition, rather than contexts,
these definitions match Gazdar’s. Moreover, Charlow understands states in a way
derived, ultimately, from Hamblin’s (1971) commitment slates, one inspiration
for Gazdar’s conception of context. Therefore, we can see Charlow’s work as
representing the approach to speech acts originally sketched by Gazdar.

 We also see Gazdar’s ideas reflected in much work which we will classify under the dynamic

approach, in particular Roberts (, ); Portner (, ); Starr (). However, Charlow
presents his ideas as a form of speech act theory.

i i

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

i i

sentence mood in speech act theory 177

Charlow has a traditional, operator-based view of sentence mood which iden-

tifies sentence moods closely with clause types. He assumes that a root clause’s
grammatical form can be factored into a type-indicator and a clausal radical, giving
its “clause type representation” (2011, p. 71):

(49) a. dec(the window is shut)

b. int(the window is shut)
c. imp(the window is shut)

The clause types are conventionally associated with logical forms representing
the illocutionary act they will perform, but this relation is based on the speaker’s
goals in the the context and hence defeasible. The basic clause types of (49) are
conventionally but defeasibly associated with familiar-looking logical forms:

(50) a. (the window is shut)

b. ?(the window is shut)
c. !(the window is shut)

Each speech act operator determines a characteristic update on the cognitive state.
Charlow assumes that the cognitive state is structured to allow an explanatory
theory of speech acts of assertion, asking, and “necessitation” (the act convention-
ally performed by imperatives). Each speech act operator updates a characteristic
component of the discourse context based on its propositional content: (φ)
modifies the informational component so that φ is included in the information
in the state, and !(φ) modifies an action-guiding component so that φ is necessary
in view of the state. (He is uncertain whether interrogatives should be thought of
as updating the informational component or a third component of their own.)
To summarize, in Charlow’s theory sentence moods are identified with clause
types, and sentential forces are the functions designated by the speech act
operators , ?, !. The relation between moods and forces is given by an inferential
component of the theory which incorporates defeasible rules linking each clause
type with its sentential force. The overall structure of the theory implies that the
range of sentential forces is itself explained by the range of natural updates which
can be performed on a cognitive state, and so the model of the cognitive state is
itself quite crucial. Overall, Charlow offers a coherent theory addressing many
of the issues which are central to the analysis of sentence mood. As has been
mentioned, the picture he offers is similar to several theories within the dynamic
approach. What’s unique, apart from details we have set aside here concerning the
specific update performed by imperatives, is the fact that Charlow integrates his
ideas into speech act theory—albeit a “Gazdarian,” not a “Searlean,” version of it.

 Compare Charlow’s ideas on this relation to earlier work by Bierwisch (); see Section ...
 In its use of the structure of the context to provide the basis for an explanatory theory of sentence
mood, Charlow follows closely previous and contemporary work in the dynamic approach. We will
return to this idea below (see especially Section .). Note that the actual updates which Charlow
proposes are fairly complex, but the details are not important to our goals here.

i i

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

i i

178 sentence mood

3.3 Sentence mood in the dynamic approach

Since the 1980s, dynamic semantics and pragmatics have come to be the basis for
much important research on the nature of sentence mood. As we see from its early
incarnations, such as Hamblin’s (1971) commitment slates and Stalnaker’s (1978)
common ground, the dynamic approach seeks to model a body of information
and how that information is affected through communication. Sentence moods
can then be understood as indicating the linguistically basic ways in which that
body of information is affected by utterances of full sentences.
In order to understand how different versions of the dynamic approach work, we
have to focus on the vague phrase “body of information.” In its narrowest sense, the
body of information is specifically a model of what the world is like. In simple cases,
we often think of the common ground as a repository of factual presuppositions: as
the participants in a conversation share knowledge with one another, the common
ground grows to include more and more facts. You say “It’s cold” and I say “It’s
raining,” so now we both know it’s cold and raining. This pooling of knowledge
can obviously be of great benefit to those who share access to the pool.
This view of information flow in discourse as only the sharing of factual nuggets
is too simple, and so we can map out how various versions of the dynamic approach
work by tracking how their conception of discourse information differs from it.
There are three main issues on which approaches differ.

1. How do we understand the “body of information”? It could represent:

(a) A state of a conversation, for example Stalnaker’s view of the common
ground as mutual presuppositions.
(b) A cognitive state, for example an individual’s beliefs.
(c) Any type of hypothetical scenario, such as a story, a dream, or the
consequences of a hypothesis.
2. What goes into the “body of information”? It could contain information
about:
(a) How things are, for example the common ground.
(b) What further information is relevant, needed, or expected.
(c) What priorities are operative for judging situations as more or less
preferred.
(d) What entities are available to be referred to.
3. How the theory models changes to the “body of information” over time. It
could represent:
(a) Information at a particular point in time. A theory which takes this
approach would then associate each sentence with a context change
potential or update potential, a model of how the sentence affects the
body of information.
(b) The history of information states through an entire conversation, with
rules governing which sequences constitute a proper conversation.

Because dynamic theories differ on a variety of fundamental philosophical para-

meters, the list above can only be expressed accurately in a fairly abstract way. For

i i

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

i i

sentence mood in the dynamic approach 179

example, the exact interpretation of “how things are” in point 2 depends on the
answer to question 1. When a dynamic theory models how things are, this might
concern how things are according to the presuppositions of a conversation, or
someone’s beliefs, or some hypothetical scenario. But in actual practice, the various
dynamic approaches aren’t so difficult to understand, and they are not as different
from one another as their philosophical differences might lead one to expect. Let
us look briefly at a few early dynamic approaches, to see how they stand on the
issues above:

1. Stalnaker’s common ground model understands the body of information it

represents to be the shared presuppositions of the participants in a conversa-
tion at a given point in time (Stalnaker 1974, 1978). It is only concerned with
how things are (according to those presuppositions). Therefore, the model is
simply a set of propositions. It contains a pragmatic mechanism (assertion)
for changing the body of information.
2. Hamblin’s commitment slate model understands the body of information
to represent the commitments of the participants in a conversation as they
change through the course of a conversation (Hamblin 1971). It represents
each participant’s commitments pertaining to how things are (factual com-
mitments) and to a limited extent expectations about what information will
be added next (inquiries). In order to model the ways in which commitment
slates change through a conversation, it defines a dialogue as a legal sequence
of locutions and commitment slates. (We will discuss Hamblin’s paper in more
detail in Section 3.3.1 below.)
3. Kamp’s Discourse Representation Theory understands the body of informa-
tion to be the semantic content and pragmatic context of what has been said
in a conversation at a given point in time (Kamp 1981). It represents both
how things are and the discourse referents which are available for subsequent
reference. It understands the discourse representation to be a mental model
and precisely defines mechanisms for changing the mental model as new
utterances are interpreted.

The differences among these approaches are significant, and we have only sum-
marized three early papers. But at a practical level, we can describe them as all
sharing a model of factual information (the common ground, factual component
of the commitment slate, and discourse representation). To this core, Hamblin
adds ideas to deal with questions and answers, while Kamp adds mechanisms to
deal with discourse referents. As we consider various theories within the dynamic
approach, I will aim to emphasize their commonalities. I will treat as less important
those differences among theories which do not seem likely to play a role in
explaining the nature of sentence mood. In particular, I’ll de-emphasize differences
which have to do with whether the factual information they represent should be
thought of as the common ground, interlocutors’ beliefs, or some other sense of
communicated meaning. Instead, I’ll focus on the formal structure of the model
and the use to which it can be put as we aim to develop a better understanding of
sentence mood.

i i

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

i i

180 sentence mood

The structured context. As discussed just above, different dynamic theories

think of the body of information which is changed through communication in
different ways. Among those which attempt to address the properties and nature
of sentence mood, however, they virtually all think of that body of information as
the discourse context in the following specific sense:

(51) The discourse context is constituted by all of the features of a conversation

which are essential to linguistic communication.

Stalnaker’s common ground is understood as a partial model of the discourse

context because the theory seeks to explain the important linguistic concepts of
assertion and presupposition in terms of it. For the remainder of this section, we
will use this notion of the conversational context to explain, compare, and further
articulate what the various dynamic semantics and pragmatics theories have to say
about sentence mood. We will do this even though, admittedly, some of the authors
might not describe their own ideas using the term “discourse context.”
The central idea of the dynamic approach is that each sentence is associated
with a context change potential (see Section 1.4.1). This context change potential
is meant to explain how the sentence contributes to the flow of information in
discourse. Within this tradition, it is natural to think of different sentence moods
as contributing differently to the flow of information, and hence many scholars
have pursued the idea that each sentence mood has its own characteristic kind of
context change potential. From this perspective, Stalnaker’s definition of assertion
is just the beginning. Assertion, so the thought goes, is the force of declaratives
(or maybe of just some declaratives); we need an analysis like that of assertion for
the forces of interrogatives, imperatives, and sentences exemplifying other sentence
moods. So the question becomes: what other kinds of updates can there be, beyond
updates of assertion?
In the remainder of this section, we will see various answers to this question, but
as we set out, it is important to note a theme common to many of them. This is the
idea that, if we are to define more sentential forces, there must be more structure
in the discourse context. We will call a model of the conversational context which
is structured so as to allow an explanation of the properties and nature of sentence
mood the “structured discourse context”:

(52) A structured discourse context is a model of the discourse context

which has internal structure allowing an explanatory theory of sentence
mood.

A simple example of a theory like this is due to Portner (2004). Portner defines
a context as a triple consisting of the common ground, the question set, and a
function assigning a to-do list to each participant in the conversation.

(53) A Portner context is a triple cg, qs, tdl:

a. The common ground cg, a set of propositions.

i i

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

i i

sentence mood in the dynamic approach 181

b. The question set qs, a set of question-denotations.

c. The to-do list function tdl, assigning a set of properties to each
participant p.

Portner uses the structure of (53) to define the sentential forces as in Table 3.2. We
will see how Portner’s model works in detail in Section 3.3.3 (and by that point we
will have discussed the previous work on which it builds). The key point for now
is to see it as an example of the structured discourse context.

Table . Structured discourse context of Portner (2004)

Sentence mood Force Component of context

declaratives assertion updates cg

interrogatives asking updates qs
imperatives requiring updates tdl

Dynamic theories which employ a version of the structured discourse context

do not necessarily think of that structure in terms of a list of simple components.
For example, Inquisitive Semantics proposes that a single component can have
enough structure to model the functions of both declaratives and interrogatives
(see Section 3.3.2). In evaluating a theory of sentence mood within the dynamic
approach, the main question often becomes whether the particular version of
the structured discourse context which it proposes allows an analysis of all of the
sentential forces of the sentence moods it attends to, and more generally whether
it serves as the basis for an explanation of important properties of sentence mood
systems in natural language.

Dynamic semantics vs. dynamic pragmatics. Before we go on to discuss what

theories have been developed of the sentence moods in the dynamic approach,
I want to briefly highlight several ways in which these theories differ that may
not be so easy to see as the discussion proceeds. The first of these has to do
with the difference between dynamic pragmatics and dynamic semantics. As we’ve
defined it in Section 1.4.1, dynamic pragmatics is a dynamic theory in which the
semantic content of a sentence is not inherently dynamic; it is a traditional, static
semantics value like a proposition or a property. Within this approach, sentences or
utterances are assigned an update potential based on their static semantic values
according to some pragmatic principles. For example, within Stalnaker’s theory,
sentences denote propositions which can be asserted. The assertion of a proposition
updates the common ground, and so we can think of assertion as a pragmatic tool
associating a proposition with an update potential. Heim’s File Change Semantics
(Heim 1982, 1983) is even more explicit about the relation between the content of
a sentence (a satisfaction set Sat(S)) and its update potential (the context change
potential defined implicitly as F + Sat(S)).
In dynamic semantics, the meaning of a sentence is identified with its update
potential. In fact, as mentioned in Section 1.4.1, Heim ultimately endorses dynamic

i i

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

i i

182 sentence mood

semantics as she says that the meaning of S is its context change potential. Dynamic
semantics is often associated with the Amsterdam school of logic (e.g. Groenendijk
and Stokhof 1990, 1991; Groenendijk et al. 1997; Aloni and van Rooy 2002;
Groenendijk and Roelofsen 2009). It is not clear how important the difference
between dynamic pragmatics and dynamic semantics is when it comes to
evaluating theories of sentence mood. We see, for example, very similar theories
about sentence mood in the dynamic pragmatics analysis of Portner (2004), the
speech act analysis of Charlow (2011), and the dynamic semantics analysis of
Starr (2013). All of these treatments pursue the hypothesis that we have three
basic sentence moods—declaratives, interrogatives, and imperatives—because
there are three natural ways to update the structured discourse context; these
three ways of updating the context serve as the sentential forces of the basic
sentence moods. Nevertheless, there are important differences of detail among the
proposals. All three authors focus on imperatives and yet they treat the update
potential of imperatives in different ways and have different proposals about how
the syntax of imperatives determines their update potential. For the most part,
these differences simply represent different ideas about the semantic/pragmatic
system, and cannot be attributed to the deeper differences among speech act
theory, dynamic pragmatics, and dynamic semantics.
As we go forward, we will also see differences among dynamic theories in how
they view the interface between grammar and update potential. In Section 3.1.3,
we identified three approaches to this interface: the operator-based approach,
the construction-based approach, and the compositional approach. In dynamic
semantics and pragmatics, scholars often assume an operator-based explanation
of the relation between sentence mood and sentential force. We also find examples
of the compositional approach (Portner 2004; Farkas and Roelofsen 2017; Roberts
to appear). I do not know of any explicitly construction-based dynamic theories of
sentence mood, but there is no reason one could not exist.
A third difference among the theories discussed in this section concerns whether
they formulate their ideas in terms of the discourse context which is changed by the
update potentials of sentences, or whether they aim to given an overall “grammar”
of discourse. Most dynamic theories take the first shape, and I will typically discuss
all of them in that way. But we see another way in the work of Hamblin (1971)
and Roberts (1996/2012, 2004). These authors aim to define a well-formed dialogue
as a sequence of sentences which conforms to a set of rules; these rules can be
seen as defining a generative grammar for dialogues. However, as we will see, in
practice the latter type of theory can always be recast in the more typical context-
and-update format, and I will do so in order to ease explanations and comparison
with other theories.

 As noted in Section .., it would be quite feasible to marry Ginzburg and Sag’s ()

construction-based view of syntax and semantics with a dynamic analysis of sentential force. It is also
reasonable to view Lauer’s () theory as implicitly assuming a construction-based view.
 See also Portner (to appear) for additional discussion of the variants of the dynamic approach.

i i

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

i i

sentence mood in the dynamic approach 183

3.3.1 Declaratives in the dynamic approach

For most work in dynamic semantics and pragmatics, the analysis of the sentential
force of declaratives is obvious. Following Stalnaker, the discourse context is repres-
ented by (or contains as a part) the common ground or a closely related construct.
Propositions can be asserted, and the essential effect of a successful assertion is to
add the proposition to the common ground (Section 1.4.1). If the discourse context
is modeled as a context set rather than a common ground, the essential effect
of assertion is to intersect the proposition with the context set. In File Change
Semantics (Heim 1982, 1983) and classical dynamic semantics (Groenendijk and
Stokhof 1990, 1991; Groenendijk et al. 1996), the discourse context is a context
set augmented to represent discourse referents, and the dynamic meaning of a
sentence is a function updating this augmented context set.
Hamblin (1971) presented a framework for analyzing dialogue which preceded
Stalnaker’s papers by a few years and which contains very similar ideas. His
approach is to define a grammar of dialogues—that is, a set of rules which define
as well-formed certain sequences of “locutions.” He defines a number of systems
within this general framework, several of which concern dialogues involving
assertions and retractions. Intuitively, these would be dialogues in which all of the
sentences uttered are declarative. These systems can be summarized as follows:

1. There are two types of locutions: assertions and retractions.

2. A dialogue is a sequence of locutions l0 , l1 , . . . , li . Each locution is associ-
ated with a speaker.
3. A commitment slate is a set of assertions.
4. The relation between locutions and commitment slates: For each locution
in a dialogue D = l0 , l1 , . . . , li , CSj (p) is a commitment slate representing
participant p’s commitments in D after locution j.
5. Rules define possible sequences of locutions.

The rules mentioned in point 5 say what locution i can be, given the preceding com-
mitment slates CSi−1 (p), and what CSi (p) can be, given locution i and CSi−1 (p). For
example, Hamblin proposes the following rules:

(54) a. Following an assertion, everyone’s commitment slate includes that

assertion.
b. Following a retraction by participant p, p’s commitment slate does not
include the retracted assertion, but the other participants’ commitment
slates remain unaltered.

Hamblin’s various systems differ in interesting ways, and overall the paper is
an exploration of various models of dialogue. For example, systems differ in
whether commitment slates are sets of locutions (as in the presentation above)

 Hamblin’s theory, including the rules, is given in a dense set-theoretic notation. I present the rules

in non-formal language so that we can focus on their role in his theory of sentence mood.

i i

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

i i

184 sentence mood

or sets of propositions (i.e. the denotations of locutions). Another difference

concerns whether a commitment brings with it commitment to other propos-
itions, in particular logical consequences of it. The relation between assertions
and their consequences becomes difficult when the dialogue contains retractions
(as well as when it contains questions and answers, discussed in Section 3.3.2),
and his discussion of these problems is interesting, but not very relevant to
sentence mood.
Because the rules which define possible sequences of locutions involve the
relation between locution li and the commitment slates of the participants at i − 1,
Hamblin’s systems can be translated into rules for contextual update along the
lines familiar from Stalnaker’s work. In fact, as discussed in Section 1.4.3, Gazdar
presciently pointed out how well Hamblin’s model fits with the dynamic force
hypothesis in which “an assertion that φ is a function that changes a context in
which the speaker is not committed to justifiable true belief in φ into a context
in which he is so committed” and moreover “the familiar ‘common ground’ view
of the context can be defined on the basis of the set-of-commitment-slates view”
(Gazdar 1981, p. 69). From this perspective, Hamblin’s notion of assertion is quite
close to Stalnaker’s.
Although there is widespread agreement within the dynamic approach that an
assertion results in either the addition of a proposition to the common ground
(or the speaker’s commitment slate) or the intersection of a proposition with
the context set (or file, or information state), scholars are for the most part not
explicit about how assertion relates to the declarative sentence mood. Scholars in
this tradition use the word “assertion” in slightly different ways. Stalnaker talks
of assertions as acts: “assertions are made in a context” (Stalnaker 1974, p. 315).
Hamblin treats them as a variety of “locution.” Roberts describes them as “moves,”
writing: “Note that moves, on the interpretation I will give them, are not speech
acts, but the semantic objects which are used in speech acts: A speech act is the
act of proffering a move” (Roberts 2012, p. 5). In contrast, Portner uses “assertion”
primarily to name a sentential force, specifically the force of declaratives. I do
not think it’s possible to paper over these differences. Stalnaker’s conception of
assertions as acts seems quite different from Roberts’ idea that they are semantic
objects used in speech acts.
Despite these differences, it probably is reasonable to assume a common per-
spective on the sentential force of declaratives among all followers of the dynamic
approach. Using the terminology favored in this book, the canonical view includes
the following tenets:

(55) The sentential force of declaratives is assertion, understood as either (a)

or (b):
a. The addition of the proposition denoted by the sentence to a component
of the structured discourse context representing either
(i) the mutually presupposed information, i.e. the common ground, or
(ii) the speaker’s commitments, i.e. the speaker’s commitment slate.
b. The operation of intersecting the proposition denoted by the sentence
with a proposition representing (i) or (ii), i.e. the context set.

i i

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

i i

sentence mood in the dynamic approach 185

(Perhaps Hamblin would disagree with (55), since retractions presumably take the
form of declaratives in most cases; if so, this fact would leave Hamblin without a
good candidate for the sentential force of declaratives.)
Although (55) is widely accepted within the dynamic approach, only a few
scholars present explicit ideas about how declaratives are associated with the
force of assertion. Portner (2004) follows the compositional approach; he assumes
that declaratives, but not interrogatives or imperatives, have propositions as their
denotations, and proposes that a pragmatic principle assigns assertion as the force
of declaratives on the basis of their propositional type. In contrast, Lauer (2013)
argues against a compositional approach based on semantic type and proposes a
normative “declarative convention.” While he does not address the grammatical
status of clause types in detail, it seems that he treats declarative as an unanalyzed
construction to which sentential force is associated holistically by the declarative
convention. The operator-based view is represented by Starr (2013) and Charlow
(2011), both of whom assume an assertion operator.

The relation between assertion and asking. In a number of dynamic theories,

the analysis of assertion cannot be understood simply by thinking about the
function of declaratives. Rather, they develop, in one way or another, ideas about
assertion which are closely related to the analysis of questions. A clear example
of this perspective comes from dynamic semantics and early forms of inquisitive
semantics. In these frameworks, several scholars have proposed that the discourse
context is a relation over a set of possible worlds, that is, a set of pairs of worlds
(Jäger 1996; Hulstijn 1997; Aloni and van Rooy 2002; Aloni et al. 2007; Mascarenhas
2009; Groenendijk 2009). Assertion affects the domain of the relation, that is,
the set of worlds which participate in the pairs. This view of assertion is not
fundamentally different from the standard one, since it amounts to intersection.
For example, if we assume that S denotes proposition p, to assert S in context C
amounts to the following:

(56) Assertion in dynamic semantics for questions: The result of asserting S in

context c is c ∩ (p × p).
• In the notation of dynamic semantics: c[S] = c ∩ (p × p).

The motivation for treating the context as a relation (rather than a simple set of
worlds) comes from the dynamic approach to questions. A question affects the set
of pairs in a different way from assertion. We will see how dynamic theories of
questions work in the next subsection.
Other theories which closely relate assertions and questions are more radical. As
discussed above (Section 3.1.1), Gunlogson (2001) does not clearly distinguish the
sentential force of declaratives from that of interrogatives. In her view, declaratives
always involve postulating that some participant in the conversation is committed

 Note that Starr’s theory draws on Inquisitive Semantics, and so involves a more complex view of

the discourse context than assumed so far. We will discuss it below. Charlow’s theory is a version of
speech act theory which accepts the dynamic force hypothesis.

i i

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

i i

186 sentence mood

to their propositional content, but when this participant is the addressee (the case
of rising declaratives), Gunlogson describes the result as a question. An example
of a declarative question is (57a) (=(12a) above). In contrast, when the declarative
postulates that the speaker is committed to the propositional content (as with the
falling declarative (57b)), the sentence functions as an assertion. (An interrogative
(57c) always functions as a question.)

(57) a. It’s raining? (rising intonation, declarative)

b. It’s raining. (falling intonation, declarative)
c. Is it raining? (rising intonation, interrogative)

It is not precisely clear in Gunlogson’s dissertation exactly how the meaning asso-
ciated with the declarative clause type works. With rising intonation, a declarative
seeks confirmation that the proposition is in the addressee’s commitment set, but
with falling intonation, it adds the proposition to the speaker’s commitment set.
This difference between seeking confirmation and adding a proposition is never
fully explained, but the theory is nevertheless interesting because of the way it
attempts to treat assertion and some cases of asking a question as subvarieties of a
single discourse function, differing only in which participant’s commitment slate
is targeted for the postulation of a commitment., 
It is difficult to say what Gunlogson’s theory implies for the broader theory
of sentence mood. She does not give an explicit analysis of interrogatives, so we
cannot determine whether they have a sentential force which is significantly dif-
ferent from rising declaratives. Nevertheless, she describes both rising declaratives
and interrogatives as questions, suggesting that rising declaratives have more in
common with interrogatives in terms of their pragmatic function. If this is right,
Gunlogson’s ideas seemingly would imply that interrogatives and some declaratives
constitute one sentence mood, and the other declaratives another. This conclusion
would imply that declarative is not a clause type in the sense we use the term here.
Some of the unclarities that we find in Gunlogson’s theory are addressed in later
work by Farkas and Bruce (2010). They define the structured discourse context as
follows:

(58) A Farkas/Bruce context is a tuple CS, cg, qs, ps, where:

a. CS is a function from discourse participants p to p’s discourse
commitments.
b. cg is a common ground.
c. qs, or the ‘table’, is a stack of questions under discussion.
d. ps is a set of projected extensions of cg.

 For other important work on rising declaratives, see Pierrehumbert and Hirschberg (),

Šafářová and Swerts (), Šafářová (, ), Truckenbrodt (a), and Trinh and Crnič ().
 In recent work, I have also argued that the difference between strong imperatives like orders and

weak ones like those which offer advice or permission can be explained in a version of Gunlogson’s
system (Portner, to appear).

i i

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

i i

sentence mood in the dynamic approach 187

The component CS associates each participant with a commitment slate, as in

Hamblin’s and Gunlogson’s work; cg is the common ground, and cg is defined to
be the commitments shared by all participants. We will set aside discussion of qs
for now; Farkas and Bruce call it the “table,” and it is equivalent to Roberts’ (2012)
question under discussion stack. The final component ps is essential to clarifying
some of the issues raised about Gunlogson’s system above, since it links a current
discourse move like making an assertion or asking a question with conventional
expectations about what effects that move will have.
Suppose the speaker asserts (59):

(59) I am hungry.

According to Farkas and Bruce, an assertion operator A applies to the sentence

and context, yielding a new context in which the proposition that the speaker is
hungry has been added to the speaker’s commitment slate. So far, this is similar to
Gunlogson’s and Hamblin’s systems (though Hamblin would immediately add it to
all participants’ commitment slates). But crucially, in addition the proposition is
simultaneously added to the projected extensions of cg in ps. This can be seen as
representing the idea that the speaker expects his commitment to become a shared
commitment. At the next stage of the discourse, the addressee can either agree (or
fail to object), in which case the projected extension of the common ground will
take effect (the proposition that the speaker is hungry is added to the common
ground cg), or dissent.
Farkas and Bruce (2010) do not discuss rising declaratives, and so their analysis
does not lead to any cloudiness concerning the difference between declaratives
and interrogatives. In fact, they assume that the two types are distinguished by
“sentential features” [D] and [I] and that different force operators apply to sentences
marked by [D] and [I]. Their ideas about clause types and sentence mood thus seem
similar to the views of Stenius (1967) and Krifka (2001, 2014). Some recent work
has built on Farkas and Bruce’s model to incorporate analyses of the pragmatic
effects of rising and falling intonation (Malamud and Stephenson 2015; Roelofsen
and Farkas 2015; Farkas and Roelofsen 2017).
Another theory which diminishes the difference between assertion and asking
a question is the later form of inquisitive semantics (Ciardelli et al. 2013). As this
is primarily an analysis of questions, we will discuss it just below in Section 3.3.2.
For our purposes here, I only want to point out that in inquisitive semantics, the
updates associated with statements and questions are both intersective, and there
are hybrid sentences as well (sentences with both assertion-like and question-like
effects).

Declaratives in a probability-based discourse semantics. As discussed in

Section 2.1, a number of probability-based theories of sentential and subsentential

 A similar sequence occurs when a question is added to qs, except that, because a question projects

multiple extensions of the common ground in ps, an answer is required to determine which proposition
is to be added to cg.

i i

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

i i

188 sentence mood

modality have been developed in recent years, and in some of them the composi-
tional meaning of a sentence is not a possible worlds proposition but rather some-
thing which encodes information about probabilities. In the particular version of
the idea which was sketched above, the denotation of a sentence is a set of credal
information states as defined in (38) in Chapter 2. In this section, we will consider
the consequences of such theories for the analysis of sentence mood.
Yalcin (2012) and Swanson (2015) give the most linguistically sophisticated
versions of the view according to which sentence meanings are credal information
states. Swanson defines assertion as follows:

(60) Assertion of credal information: In asserting that φ, a speaker advises her

addressees to conform their credences to the semantic value of ‘φ’. Holding
all other factors fixed, the advice associated with ‘φ’ is stronger than the
advice associated with ‘ψ’ iff ‘φ’ is more informative than ‘ψ’. (Swanson
2015)

Swanson’s definition here captures the essential intuition of the credal information
semantics: People have credences, sentences express credal information, and the
function of assertion is to guide one’s interlocutors in adjusting their credences.
Swanson’s proposal is not explicit about the status of “advice” in the theory of
sentential force.
Yalcin (2012, p. 6) proposes a dynamic semantics in which the common ground
is reanalyzed in terms of probabilities.

(61) A credal context is a set of credal information states.

(Yalcin uses the term “blunt information state” for what is labeled “credal con-
text” here.) Given this definition of the conversational context, together with the
proposal that sentence meanings are also sets of credal information states, the
sentential force of declaratives can be represented as intersection:

(62) Assertion in the probability-based semantics: The result of asserting S in

context c = c ∩ [[ S ]].

Statement (62) conforms to the standard analysis of assertion within the dynamic
approach, and for this reason it does not have any unique implications for the
theory of sentence mood. As far as can be determined on the basis of existing
proposals, the probability-based approach is compatible with the same theories of
sentence mood as other dynamic approaches are. Of course, since the probability-
based approach has not been extended to other sentence moods, we cannot say
that it is as well-supported as other, non-probabilistic dynamic approaches. It may

 Yalcin discusses the choice between a dynamic semantics treatment, on which sentence meanings

are update potentials, and a dynamic pragmatics version, on which they are sets of credal information
states. Swanson follows the latter approach.

i i

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

i i

sentence mood in the dynamic approach 189

or may not be able to incorporate adequate theories of interrogatives, imperatives,

and minor clause types.

3.3.2 Interrogatives in the dynamic approach

Background on the semantics of questions. In order to understand what dy-
namic theories have to say about the sentence mood of interrogatives, we must
build on some basic knowledge concerning the semantics of interrogatives. We
refer to the semantic values of interrogatives as questions. (As with ‘assertion’,
the term ‘question’ is used in several ways: for a type of utterance, sentence, act,
or semantic value. Here we use it only for denotations of the kind assigned to
interrogatives.) In some of the earliest work on the topic in a modern semantic
framework, Hamblin (1958, 1971, 1973) develops both an analysis of questions as
semantic objects and a dynamic model of how questions and answers function in
dialogue, and as might be expected, the analysis of dialogue draws naturally on his
ideas about questions. Subsequent work in the dynamic tradition has followed
the same pattern, incorporating specific ideas about what a question is into analyses
of what asking a question does. As we will see, to some extent advances and
disagreements in the literature on the semantics of interrogatives can be ignored
as we look at their function in discourse, but it will be necessary to have a basic
understanding of semantic developments, if only to allow a useful statement of
each scholar’s views.
We can begin with Hamblin’s semantics for questions. He treats questions as sets
of propositions, where each proposition is understood to be a possible answer to the
question. This idea is developed into a compositional analysis of interrogatives in
Hamblin (1973); there he treats (63a) as denoting the answer-set (63b), and likewise
for (64):

(63) a. Is it raining?
b. {{w: it is raining in w}, W − {w : it is raining in w}}
(64) a. Who cried?
b. {{w : a cried in w} : a is an individual} =
{p : for some individual a, p = {w : a cried in w}}

The hypothesis that the semantics of interrogatives is something like a set of

answers is motivated by an intuition about the discourse function of questions,
namely that to ask a question involves changing the context in such a way that
some party to the conversation will be motivated to answer it. (Obviously this
intuition is not unique to the dynamic approach; speech act theory develops it in
a different way.) Since we observe that people do in fact often provide answers to
our questions, it seems that knowing what constitutes an answer is at least part of
knowing the meaning of a question.

 Another well-known early theory of questions is due to Karttunen ().

i i

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

i i

190 sentence mood

The idea that a question denotes a set of possible answers will be mostly sufficient
for our purposes, but as discussed by Cross and Roelofsen (2014), it is not derived
from a detailed study of which kinds of responses to a question should count as
an answer. As they point out, Hamblin’s formal theory treats (65a) as an answer to
(64a), but not (65b) or (65c):

(65) a. Beatrice cried.

b. Beatrice alone cried.
c. A girl cried.

Groenendijk and Stokhof (1982, 1984, 1997) develop a theory which conforms to
Hamblin’s and Karttunen’s general approach but which is more specific about the
notion of answerhood which underlies it. They argue that certain semantic and
logical issues can be solved by building the semantics of questions specifically on
one distinguished type of answer, the true, exhaustive answer. More precisely, they
take the meaning of an interrogative to be a function from worlds to propositions,
where the proposition represents the true, exhaustive answer to the question in
that world. As an example, assume that in the actual world, Beatrice was the
only individual who cried. Then, the question denoted by (64a) would assign the
proposition expressed by (65b) to the actual world.
Because it isolates the true, exhaustive answer to a question as the basis for the
semantics, Groenendijk and Stokhof’s theory can be neatly expressed in another
way. Notice that, if a question Q maps a given world w to proposition p (thus, p is
the true exhaustive answer to Q in w), Q will also map any world in p to p. This
is so because the true, exhaustive answer to Q is the same for every world in p.
In terms of our example, [[ (64a) ]] (wactual ) = the set of worlds in which Beatrice
alone cried, and in any of those worlds, the true exhaustive answer to (64a) is also
that Beatrice alone cried. In symbols:

(66) For every world w ∈ p, Q(w) = p.

Generalizing this reasoning, the semantics of a question divides up the set of

possible worlds into subsets corresponding to the true, exhaustive answer, and
assigns each of these subsets to every world in it. The result can be visualized as
in Figure 3.1. Another way to represent the same fundamental information would
be to treat the question as a relation holding between any two worlds in the same
answer, as in Figure 3.2. (Note that each world is also paired with itself, but this
is left out of the figure for simplicity.) This structure is a partition of the set of
possible worlds, and Groenendijk and Stokhof ’s analysis is often referred to as the
partition theory.
There are many places for readers interested in learning more about the se-
mantics of questions to look. The handbook articles Groenendijk and Stokhof
(1997) and Krifka (2011) are good places to start, covering a range of issues and
theories. Dayal (2016) is more comprehensive, especially on issues related to the
syntax/semantics interface.

i i

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

i i

sentence mood in the dynamic approach 191

w w
w w Possible world: w
w
w True, exhaustive answer:
w
w
World-proposition pairs in Q:
w w

Figure . The question as function from worlds to answers

w w
w w Possible world: w
w
w True, exhaustive answer:
w
w
World-world pairs in Q:
w w

Figure . The question as a partition

Dynamic theories of asking a question. Above in Section 3.3.1, I presented a

summary of one of Hamblin’s (1971) systems of dialogue. That system was designed
to account for assertions by proposing that the discourse effect of an assertion was
to add it to the commitment slates of all participants in the conversation. In the
same paper, he presents more elaborate systems which also cover questions. The
following is a partial description which aims to give the basic idea of how his rather
complex system works:

1. There are five types of locutions: assertions, retractions, inquiries,

retraction-demands, and I don’t know.
In symbols: The set L consists of five non-overlapping subsets λ, μ, ξ , η, L0 .
2. A dialogue is a sequence of locutions l0 , l1 , . . . , li .
3. A commitment slate is a set of assertions.
4. The relation between locutions and commitment slates: For each locution
in a dialogue D = l0 , l1 , . . . , li , CSj (p) is a commitment slate representing
participant p’s commitments in D after locution j.

The preceding is the same as the system described earlier, except that there are
more kinds of locutions. The rules which define the set of well-formed dialogues
are different so as to account for inquiries, retraction-demands, and I don’t know.
Some of the relevant rules are paraphrased below:

 I am describing Hamblin’s (, pp. –) System  here.

i i

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

i i

192 sentence mood

(67) a. Following an assertion, everyone’s commitment slate includes that asser-

tion.
b. Following a retraction by participant p, p’s commitment slate does not
include that retraction, but every other participant’s commitment slate
remains unaltered.
c. An inquiry is not allowed if an answer to it is in anyone’s commitment
slate.
d. Following an inquiry, the next locution must be an assertion by someone
other than the speaker of the inquiry and must answer it.

As before, a well-formed dialogue is one in which each locution lj follows the rules
given the preceding locution lj−1 and the participants’ commitment slates CSj−1 (p).
In turn, each participant’s commitment slate CSj (p) associated with lj is determined
by that participant’s preceding commitment slate CSj−1 (p) and the latest locution
lj . This means we can think of locution lj−1 and all of the participants’ associated
commitment slates as the discourse context for locution lj :

(68) A Hamblin context is a locution and an assignment of commitment slates

to participants.
Specifically, the context for locution lj is:
lj−1 , CSj−1 , where CSj−1 (p) is defined for each participant p.

This notion of context lets us recast Hamblin’s theory of inquiries in the more
familiar context-and-update model of dynamic approaches. Assertions affect the
context in two ways, by becoming the locution component lj−1 of the new context
and by being added to each commitment slate. This, as we said above, amounts
to a version of the standard dynamic treatment of assertion. Inquiries affect the
context in only one way, by becoming the locution component of the new context.
Inquiries do not affect anyone’s commitment slate, but when the last locution
was an inquiry, discourse rules constrain the choice of the next locution: after
an inquiry must come an assertion which answers it. Hamblin’s dynamic theory
of inquiries therefore amounts to saying that the context change potential of an
inquiry is to limit the range of locutions which can come next to a specific set of
assertions. As we noted above, Hamblin does not say enough to allow us to attribute
a specific theory of sentence mood; crucially, it is not clear that the inquiries
correspond exactly to interrogatives. (In fact, Hamblin (1973, p. 52) suggests that
the interrogative What entity is an entity? “is really an indicative,” because its
denotation contains a single proposition.)
Roberts (1996, 2004, 2012) formulates a theory of assertions and questions
which is similar to Hamblin’s in several ways.,  Like Hamblin’s, the theory is

 This is so because, for each entity x, the proposition that x is an entity is the same proposition, the

tautology W.
 Ginzburg (a,b, ) develops ideas very similar to Hamblin’s and Roberts’ within a Situation

Semantics framework.
 Roberts () is essentially the same paper as Roberts (). There are some corrections in the

newer version.

i i

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

i i

sentence mood in the dynamic approach 193

formulated in terms of a grammar of dialogue, though Roberts uses the term

information structure for what Hamblin called the dialogue. The information
structure employs a structured discourse context as a key part and the portion of
the context which is devoted to questions is more sophisticated than Hamblin’s.
I will give a partial description of Roberts’ notion of information structure here.

(69) An information structure has seven components:

1. A set of moves M.
2. A set of questions Q, a subset of M.
3. A set of assertions A, a subset of M.
4. The precedence relation < over M, reflecting the order in which the moves
are made in the discourse.
5. The set of accepted moves acc, a subset of M.
6. The common ground function cg which assigns a common ground cg
to each move in M.
7. The question under discussion function qud which assigns a question
set qs to each move in M.

The first four parts of the information structure correspond directly to features
of Hamblin’s model. The component acc is an innovation and allows for the
possibility that certain moves do not affect the other components because the
participants other than the speaker reject them. The common ground is of course
familiar. Note that cg(m) is the common ground “just prior to the utterance of m.”
It will include the effects of all moves preceding m as well as any other information
which has become presupposed through other means. The real innovation is in the
final component, which assigns to each move a question set. Unlike Hamblin’s
system, where a question only has an effect on the next locution, in Roberts’
theory multiple questions can influence the discourse at a single point in time,
and they stand in both semantic and pragmatic relations to assertions and other
questions. These relations allow a better explanation of the role which questions
play in dialogue.
For a given move m, the appropriateness and effect of m are defined entirely
in terms of the common ground and question set associated with m. Some of the
important rules relevant to assertions and questions may be paraphrased as follows:

(70) a. An accepted assertion ai is in cg(mj ) for all subsequent moves mj .

b. An accepted question qi is in qs for all subsequent moves mj , unless an
answer to it is entailed by cg(mj ) or “qi is determined to be practically
unanswerable.”
c. As long as a question is in qs, any new question must be more specific
than it.

 In reality, moves are probably not rejected so completely and abruptly as the model suggests. In

Hamblin’s system, cases of rejection would be handled through retraction-demands and retractions.
Merin () creates a system in which some of the complexity of negotiation can be represented.
 Roberts calls the question set a “question under discussion stack” but since the order associated

with the stack is just <, we can treat qs as a simple set.

i i

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

i i

194 sentence mood

More precisely: For any questions q and q both in qud(m), if q < q ,

then a complete answer to q entails a partial answer to q.

The fact that the appropriateness and effect of each accepted move may be de-
termined on the basis of its common ground and question set makes it possible to
extract a more standard notion of context from Roberts’ definition of information
structure.

(71) A Roberts context is a common ground and a question set.

Definition (71) allows us to restate Roberts’ theory in terms of a more standard

context-and-update-potential format. Assertion, as usual, is the addition of a
proposition to the common ground (assuming the move is accepted). Asking a
question is the addition of a question to the question set, subject to the condition
that it be more specific than any other question in the set (see Roberts 2012, (17)).
There are other interesting aspects of conversational dynamics which can be
explored in this framework. For example, Rule (70b) implies that a question is
removed from the question set as soon as the common ground entails an answer to
it; this could happen because an assertion provides the answer. Roberts also defines
a notion of Relevance for moves:

(72) The question under discussion for move m is the most recent member of
qud(m).
(73) A move is Relevant to the question under discussion Q if and only if:
a. m is an assertion, and it provides at least a partial answer to Q, or
b. m is a question and a complete answer to m entails at least a partial answer
to Q.

After presenting her basic theory of information structure, Roberts goes on to

investigate the role of focus in discourse.
Roberts’ theory represents a milestone in the development of dynamic pragmat-
ics theories of questions in discourse, and has been the basis for much continuing
work (e.g. Büring 2003; Beaver and Clark 2008; Biezma and Rawlins 2012). But she
does not quite set things up in a way which provides a clear notion of sentential
force or a theory of sentence mood. Her discussion is not explicit about the
relation among interrogative sentences, utterances of them, question denotations,
the “proffer” of a question denotation, and speech acts. As we are looking for ideas
about the nature of sentence mood, however, I think it’s reasonable to associate
with Roberts’ theory the following claim, even if it may not be a hypothesis she has
explicitly endorsed:

 Roberts’ version of (b) employs a notion of a “strategy” to answer a question, but from what I

can tell, rule (b) guarantees that it is part of a strategy in the relevant sense.

i i

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

i i

sentence mood in the dynamic approach 195

(74) The sentential force of interrogatives is asking, understood as the addition

of the question denoted by the interrogative to a component of the struc-
tured discourse context representing the mutual commitment to add to the
common ground information which answers the question.

In Roberts’ approach, the question set encodes even more than this, in particular
a commitment to answer the questions in a certain last-to-first order, but I think
she intends for the issue of how the interlocutors should address the questions in
the question set to be a matter of Gricean cooperativeness rather than an aspect of
their sentential force.

The relation between asking and assertion. In Section 3.3.1, we began an over-
view of one important approach to questions in dynamic semantics. This work
treats the context as a relation over a set of worlds, with assertion equivalent to
the operation of restricting the domain of that relation (Jäger 1996; Hulstijn 1997;
Aloni and van Rooy 2002; Aloni et al. 2007; Mascarenhas 2009; Groenendijk 2009).
Now we are ready to see how the update potential of interrogatives is explained in
this approach, and in order to do this, we have to understand precisely what type
of relation the context is and what it represents. The dynamic semantics analysis of
questions is built on two principles:

1. Questions are analyzed following the partition theory.

2. The context is a big question.

Let us begin with the second point. Since questions are understood as partitions, to
say that the context is a question is to analyze it as a partition. Specifically, each cell
in the partition represents a complete answer to all open questions. See Figure 3.3.
(Note that the worlds in each cell of the partition are also related to themselves—
that is, the relation is reflexive—but the reflexive pairs are left out of the diagram
for simplicity.) Intuitively, the partition represents a conversation which will reach
its resolution as soon as the domain of the context is a subset of one of the cells in
the partition. An assertion of proposition p in the context illustrated would answer
all of the open questions, leading to a new context with a single cell consisting of
those two worlds.

w w
w w A complete answer to
w all open questions:
w
w Worlds related by C:
w
w w Asserted proposition p:

Figure . The context as a partition

i i

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

i i

196 sentence mood

w w w
w
w w w w
w w
w w
w w
w w

w w w w

question q new context after q

asserted proposition p:

Figure . Asking a question in the partition context

If the context is a partition and a question is likewise a partition, the effect of

asking a new question can be modeled as dividing the cells in the context along the
lines between the cells of the question. Figure 3.4 is an example. The new context
represents the result of incorporating question q into the context of Figure 3.3. In
order to answer all of the questions in the new context, p is no longer informative
enough, because it overlaps two cells in the partition. The reason for this is that p
provides only a partial answer to q (it overlaps two cells in q). The result of asserting
p in the context illustrated will be a new context with two disconnected worlds, each
constituting its own cell in the partition.
As we have described the dynamic semantics analysis of questions so far,
assertion and asking a question are closely related operations on the context.

(75) Assertion and asking in dynamic semantics for questions:

a. if S denotes a proposition p, C + S = C ∩ (p × p)
b. if S denotes a partition q, C + S = C ∩ q

Though it has not previously been identified in the literature on dynamic semantics,
so far as I am aware, an interesting intuition about sentence mood can be found
in (75). This intuition is that sentences which denote propositions and those
which denote questions have almost the same sentential force. Unlike Roberts’
theory, where assertion and asking a question update different components of the
discourse context in the same way, in the version of dynamic semantics we are
considering, they update the same component of the discourse context in different
(though similar) ways.
It is possible to extend the dynamic framework so that there is no force-
like difference between assertion and asking. Suppose that we no longer assume
that any sentences denote sets of possible worlds. Rather, we assume that even
declaratives denote partitions, but that their partitions are so rudimentary that they
do not introduce any new questions into the context. It is easy to do this within the

 The conceptualization of assertion and asking in () can readily be represented as proper dynamic

semantic values. The meaning of sentence S would simply be the function λc[c + S].

i i

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

i i

sentence mood in the dynamic approach 197

framework of dynamic semantics. Suppose that we would normally assume that S

denotes some proposition p. Now we take it to denote p × p. Then the same update
can be applied using both rudimentary and articulated partitions:

(76) Unification of assertion and asking in dynamic semantics:

For any sentence S denoting partition p, C + S = C ∩ p

This type of unification of assertion and asking is found in inquisitive semantics

(see especially Ciardelli et al. 2013).
Now we have three intuitions about the relation between assertion and asking
in the dynamic approach:

1. Assertion and asking update different components of the structured discourse

context in the same way (Roberts).
2. Assertion and asking update the same component of the structured discourse
context in similar ways (first version of dynamic semantics for questions).
3. Assertion and asking update the same component of the structured discourse
context in the same way (second version of dynamic semantics, inquisitive
semantics).

Though, as far as I know, no scholars working in the dynamic tradition have

done so, these three intuitions can be seen as representing three ideas about
sentence mood. The first says roughly that declaratives and interrogatives have
different sentential forces. This view coincides with the traditional conception
of sentence mood, according to which declaratives and interrogatives are basic
and not related to one another in an especially close way. The second says that
declaratives and interrogatives have different sentential forces, but that these forces
are closely related. This view fits well with the idea that the force of declaratives and
interrogatives share something important, as proposed in very different ways by
such authors as Truckenbrodt (2006a) and Lohnstein (2007). The third intuition
says that the difference between declaratives and interrogatives is entirely a matter
of their compositional semantic values, and that at the most fundamental level,
they have the same conversational function. In a sense, the third approach implies
that declaratives and interrogatives constitute a single sentence mood.
Now we turn to the most recent incarnation of dynamic semantics for questions,
inquisitive semantics. Inquisitive semantics differs from earlier dynamic the-
ories mainly in how it represents the meanings of questions (including the “big
question,” the structured context). In order to see this point in detail, we want to
focus on how inquisitive semantics diverges from the partition theory of questions

 Some but not all proposals within the inquisitive semantics tradition technically fall within

dynamic semantics, as we are defining that category here. I think it’s correct to understand all work
labeled “inquisitive semantics” as following the dynamic approach, but some of it probably assumes a
dynamic pragmatics conception of the relationship between semantic content and discourse meaning.
I will present it using the techniques of dynamic semantics, however, because that allows us to see more
clearly the links to earlier theories. Thanks to Malte Willer for drawing my attention to this issue.

i i

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

i i

198 sentence mood

w w
w w A possible assertion which
w would completely settle all
w open questions:
w
w
w w

Figure . The inquisitive context

assumed by earlier versions of dynamic semantics. For technical reasons that

need not concern us here, inquisitive semantics replaces the view of questions
as partitions with the closely related hypothesis that a question is a downward-
closed set of propositions based on an equivalence relation. (A context is still a
big question, and so a context is also a downward-closed set of propositions based
on an equivalence relation.) The context previously represented as in Figure 3.3
corresponds to the one in Figure 3.5; the context is the set of all of the sets of
worlds represented by ellipses in the figure. Each of these sets can be thought of
as a proposition which, if asserted, would completely settle all open questions.
As with the version of dynamic semantics which unifies assertion and asking (i.e.
(76)), in the inquisitive approach all sentences denote semantic objects of the same
sort, in this case downward-closed sets of propositions, and when a sentence
updates the context, the context is reduced through intersection. The contents
of some sentences are so rudimentary that they only reduce the domain, and do
not split any cells of the context; these have “informative content.” Others split
cells in essentially the same way as illustrated in Figure 3.4; they have “inquisitive
content.” (The theory also allows “hybrids,” contents which both reduce the domain
and split cells.) Given this nice distinction between informative and inquisitive
content, as we seek to apply the methods of inquisitive semantics to natural
language, it is natural to ask: which sentences have informative content, and which
have inquisitive content? One might think that the answer is that declaratives
have informative content and interrogatives inquisitive content, but as it has been
developed, inquisitive semantics does not postulate such a neat correspondence
with clause types.
Disjunction provides the noteworthy exception to the attractive idea that we
can link declaratives with informative content and interrogatives with inquisitive
content. Within inquisitive semantics, disjunction creates a semantic value with
inquisitive content in just the same way that an interrogative operator does. And
this is not just a matter of mismatch between the logical formalism in which
inquisitive semantics is normally pursued and natural language. Roelofsen and
Farkas (2015) propose rules for interpreting Logical Forms of English according to
which (77a) and (77b), both uttered with rising intonation on the disjuncts, have
the same meaning:

 Or, more accurately, an update potential based on such a set, but we are simplifying here.

i i

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

i i

sentence mood in the dynamic approach 199

(77) a. Igor speaks English or French.

b. Does Igor speak English or French?

In fact, Roelofsen and Farkas (2015) treat rising declaratives as equivalent to rising
interrogatives quite generally, in this respect both agreeing with and clarifying
the perspective of Gunlogson (2001). As mentioned in Section 3.3.1, recent work
by Farkas and Roelofsen (2017) develops the line of research on declaratives and
interrogatives which began with Gunlogson’s work within an inquisitive semantics
framework.

3.3.3 Imperatives in the dynamic approach

Background on the semantics of imperatives. As with interrogatives, theories of
the sentential force of imperatives are based on a variety of different assumptions
about their semantics. We find scholars who assume that imperatives denote
propositions, properties, and actions (as well as the dynamic semantics versions
of some of these things). These differences are not primarily motivated by the
need to analyze sentence mood, but rather by a number of syntactic and semantic
properties of imperative clauses. In this section, I will summarize these properties
and highlight the ones which are most significant as we evaluate theories of
sentence mood.
The following properties have been important in the literature on imperatives:

1. Imperatives cannot be naturally described as true or false.

2. In some languages, imperatives cannot be embedded; in others, they can be
embedded only in a limited range of contexts (Han 1998; Pak 2008; Crnič and
Trinh 2009; Kaufmann 2012; Kaufmann and Poschmann 2013).
3. In many languages, the imperative subject must refer to the addressee, refer to
a set containing the addressee, or quantify over a set of addressees (Downing
1969; Schmerling 1982; Platzack and Rosengren 1994; Potsdam 1996; Jensen
2003; Rupp 2003; Zanuttini 2008; Kaufmann 2012; Zanuttini et al. 2012).
4. Even when the imperative subject is not associated with the addressee as in
point 3, the imperative is normally used to tell the addressee what to do. Some
languages have minor clause types (promissives and exhortatives) which are
similar to imperatives in discourse function, but which target an individual
other than the addressee (Pak et al. 2008a,b; Zanuttini et al. 2012).
5. Imperatives are normally future-oriented, but past-oriented types may exist
in some languages (Palmer 1986; Mastop 2005, 2011).

 Some important work on the semantics of imperatives will not be discussed because it does not

address to a sufficient extent the nature of sentential force and sentence mood. This includes much of
the tradition of imperative logic (for example, Ross ; Geach ; Lemmon ; Åqvist ;
Chellas , ; Kanger ; Segerberg ; Vranas , ). Some linguistically-oriented
work discusses imperatives in semantic terms without introducing any distinctive component of force
(for example, the modal analysis of Aloni ).

i i

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

i i

200 sentence mood

6. In many languages, imperatives take on a different morphosyntactic form

when negated or when they express pragmatic features of speech style and
politeness (Rivero and Terzi 1995; Zanuttini 1997; Han 1998; Zanuttini and
Portner 2003; Han and Lee 2007; Kaufmann 2012).
7. Imperatives can be conjoined and disjoined with one another and with
declaratives and interrogatives. When they are conjoined or disjoined with
other clause types, interesting semantic effects arise; for example, Touch me
again and you’re in big trouble! implies that the addressee should not touch the
speaker again. Imperatives can also serve as the main clause of a conditional
sentence (Bolinger 1967; Davies 1986; Clark 1993; Han 1998; Franke 2005;
Russell 2007; Kaufmann 2012; Starr 2013; von Fintel and Iatridou 2015).
8. Imperatives have a different logic than declaratives. As a result, they show
the free choice inference and give rise to Ross’s paradox (some significant
references, among many others, are Ross 1944; Kamp 1973; Han 1998; Van
Rooij 2000; Asher and Bonevac 2005; Aloni 2007; Barker 2010; Mastop 2011;
Portner 2011b; Aloni and Ciardelli 2013; Fine 2016; Kaufmann 2016).
9. Imperatives are functionally diverse. They can be used to give orders, make
requests, offer suggestions, give permission, record instructions, and express
wishes, among other functions (Wilson and Sperber 1988; Portner 2004, 2007;
Charlow 2011; Condoravdi and Lauer 2012; Kaufmann 2012).

The references listed above are only representative. There is extensive literature on
several of these properties, and I also note the handbook articles by Han (2011) and
Portner (2016).
For our purposes, the only properties which require further explanation are
points 8–9. Free choice refers to the observation that (78a) implies the truth of
(78b). Not only imperatives, but many modal sentences as well, show the free choice
inference.

(78) a. Have an apple or a pear!

b. You may have an apple, and you may have a pear.
(79) a. You must/may talk to John or Mary.
b. You may talk to John and you may talk to Mary.

Free choice is not explained by standard theories of modal semantics derived from
modal logic. For example, (p ∨ q), interpreted as “every accessible world is in the
union of the p worlds with the q worlds,” does not entail ♦p ∧ ♦q, “there is at least
one accessible p world and at least one accessible q world,” because (for example)
the former would be true if all accessible worlds were p worlds.
Ross’s paradox (Ross 1944) is the closely related pattern whereby (80a) does not
imply (80b), in that you can give the former as an order without agreeing that the
latter would be a good order to give.

(80) a. Have an apple!

b. Have an apple or a pear!

i i

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

i i

sentence mood in the dynamic approach 201

(81) a. Noah ate an apple.

b. Noah ate an apple or a pear.

The lack of inference in (80) is different from the pattern seen with declaratives,
where (81a) entails (81b).
The other important property of imperatives is their functional diversity, some
of which can be observed in the following:

(82) a. March! (order)

b. Please don’t tell mom about this. (request)
c. Take the 3A bus. (suggestion)
d. Have some wine! (invitation)
e. Add the wet ingredients to the dry. (instruction)
f. Stop raining! (wish)

Though the differences among some or all of these types of uses may be semantic in
nature, as on Kaufmann’s (2012) modal analysis, within the dynamic approach, they
have not been seen that way. Instead, these differences are understood as reflecting
either the social basis justifying the use of the imperative or the speaker’s reasons
for issuing it. For example, an imperative cannot be called an order unless it is
backed by the right kind of social authority, and suggestions differ from orders
and requests in that the speaker thinks the action mentioned would serve the
addressee’s interests or goals, and leaves it to the addressee whether to commit to
the action.
All of the properties mentioned in this section are important for theories of
imperative meaning and discourse function, but we will for the most part not
explore what the various dynamic analyses discussed in this section have to say
about them. Rather, our goal is to bring out the consequences of each version of
the dynamic approach for our understanding of sentence mood, and I will present
the various scholars’ work in ways which make it easier to do so. It should be kept in
mind that different theories may be more or less adequate to account for these other
properties; readers who wish for a more complete understanding of imperative
semantics should study the literature cited. I would also note that readers interested
in the force of imperatives generally should compare the theories discussed in
this section with Kaufmann’s and Charlow’s analyses embedded in the speech act
theory framework (Section 3.2.3).

Dynamic theories of imperatives and directive meaning: early work. David

Lewis (1979b) initiated research on imperatives within the dynamic approach as
he described a language game involving three participants: the master, the slave,
and the kibbitzer. He analyzes imperatives as having a nearly standard modal
semantics of the kind familiar from modal logic (see Section 1.3.2). Specifically,

 We could alternatively see Lewis’s paper as falling within speech act theory and so as being an

early representative of the dynamic force hypothesis. However, in his only reference to speech acts he
calls himself an “Old Harry” with respect to the performative/constative distinction and does not make

i i

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

i i

202 sentence mood

he proposes that imperatives are interpreted with respect to two accessibility

functions: One of these, fsp , defines the sphere of permissibility, the set of worlds
representing the requirements governing the slave’s actions at w and t; the other,
fsa , is a historical accessibility relation defining the sphere of accessibility, the
set of worlds compatible with the actual history of w up to time t. An order-type
imperative has the logical form !φ, with ! understood as a modal logic  interpreted
with respect to the intersection of these two relations.

(83) a. !φ is true with respect to w, t iff φ is true at every world both permissible
and accessible from w and t.
b. Formally: [[ !φ ]] w,t = 1 iff
 
fsp (w, t) ∩ fsa (w, t) ⊆ {w , t   : [[ φ ]] w ,t  = 1}.

To this standard modal semantics, Lewis adds a pragmatic component to his

analysis which ties it closely to dynamic pragmatics. In a context where the speaker
has authority over the addressee (in Lewis’s language game, the master is speaking
to the slave), the contextual parameters which influence the truth conditions of
the imperative must be understood in such a way that what the speaker says is
true. Given that the two relevant parameters are fsp and fsa , and that the latter is
an objective circumstance of the world, this means that fsp must be understood
(if possible) in such a way that the imperative is true. If !φ is used in a context in
which it is not true, fsp must be adjusted so as to make it true. Lewis (1979c) sees
this adjustment as an example of a general rule of accommodation, of a kind with
other cases of accommodation involving presuppositions, indexicality, vagueness,
and other pragmatic features of meaning.
Lewis points out that with command-type imperatives, it is easy to describe how
the sphere of permissibility adjusts to make an imperative sentence true. In general,
fsp (w, t) adjusts to fsp (w, t)∩[[ φ ]] . This intersective effect of an imperative on
the sphere of permissibilty is analogous to assertion (relative to the context set) on
Stalnaker’s analysis, but as mentioned above, it differs in that Lewis sees it as due to
a general principle of accommodation. That is, the dynamic pragmatic effect of an
imperative is not, for Lewis, a basic matter of updating the sphere of permissibility
by the imperative’s content, but rather derived indirectly from its modal semantics,
and the principle of accommodation.
Lewis’s theory is able to partially unify the analyses of deontic modal sentences
and imperatives. When used in a situation in which accommodation is not ap-
plicable, sentences of the logical form !φ are simply understood as deontic modal
assertions; this would be the case when the speaker is the slave or kibbitzer, or when
the speaker is the master but the requirement is already understood in the context.
For example, (84), spoken by the master to the slave (in a context in which it is not
understood that the slave has to carry rocks), means “Carry rocks!,” but spoken by
the kibbitzer to the master, it means “The slave has to carry rocks.”

any reference to illocutionary acts or force. Hence, I think it’s more reasonable to see his ideas as falling
within the dynamic approach.

i i

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

i i

sentence mood in the dynamic approach 203

(84) !The slave carries rocks

a. Master to slave: Carry rocks!
b. Kibbitzer to master: The slave has to carry rocks.

But this unification of modal and imperative meanings is also a major weakness of
Lewis’s analysis, if it is seen as a linguistic theory of imperatives. It is not, in fact,
possible for the slave to use an imperative sentence to assert that he has to carry
rocks, or for the master or kibbitzer to use one to assert that the slave has to carry
rocks, but they can certainly use a deontic modal sentence this way. The master
can well say (85a) to gloat about the awful day the slave has ahead of him, but (85b)
cannot have this use.

(85) a. You have to carry rocks.

b. Carry rocks!

We see, then, that imperatives (at least, order-type imperatives) have a key property
distinguishing them from modal necessity statements: there are contexts in which
a deontic modal statement can be used to make an assertion about what is required,
while the imperative cannot. This points to a problem in Lewis’s use of accommod-
ation to derive the context change effect of imperatives. As is understood elsewhere
in Lewis’s writings and in the subsequent literature, accommodation occurs to
bring a context into alignment with what it needs to be for an utterance to be
“correct.” (In the case of imperatives, the idea would be that they are not correct
unless they are true.) But when the context is one in which the utterance would
already be correct, accommodation is unnecessary. But, it seems, Lewis would
need to say that imperatives can only be used in contexts where they would trigger
accommodation. This amounts to saying they can only be correctly used in contexts
in which they are not correct. The preceding discussion suggests that Lewis’s
analysis would be more accurately applied to explicit modal sentences, which can
be used both descriptively (to make an assertion about what’s permissible) and
performatively (to create a change in what’s permissible).
It is worth noting at this point the relevance of Kaufmann’s (2012) analysis of
imperatives as obligatorily performative deontic modal sentences. As discussed
in Section 3.2.3, Kaufmann argues that imperatives are associated with a set of
presuppositions which ensure that they are only properly used in contexts in which
a modal sentence would be performative. Suppose we incorporate this idea into
Lewis’s theory; that is, we add to the meaning of imperatives a presupposition
that they only occur in contexts in which they are not true. Then the rule of
accommodation would have to be defined in such a way that it adjusts the sphere
of permissibility to make the imperative true while ignoring the fact that this
will cause its presupposition to be no longer satisfied. It seems quite difficult to
understand accommodation in a way that would give this result, and this fact points
out the problem which afflicts Lewis’s use of accommodation.
We can now see the way in which Lewis’s theory contributes to the dy-
namic approach. Setting aside the precise mechanisms involved, Lewis proposes
that order-type imperatives normally have the contextual effect of shrinking a

i i

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

i i

204 sentence mood

component of the discourse context, namely the sphere of permissibility. The

sphere of permissibility can therefore be seen as analogous to the common
ground/context set and the question set. The sphere of permissibility is the “target”
for update by imperative sentences, just as on the dynamic approach the common
ground is the target for update by declaratives and the question set is the target
for update by interrogatives. Though Lewis did not make this move himself, we
can build on his idea and elaborate our theory of the structured discourse context
to include a component or feature (like the sphere of permissibility) which is
characteristically targeted by imperatives. Subsequent work on imperatives in the
dynamic approach has in fact built on Lewis’s ideas in this way.
The preceding discussion has focused on Lewis’s analysis of order-type im-
peratives, but the main focus of Lewis (1979b) concerns a problem which arises
when he tries to extend it to permission sentences. He assumes that permission
sentences have the logical form ¡φ and that ¡ has the semantics of a possibility
modal. Just as with sentences of the form !φ, when a sentence ¡φ is uttered by the
master to the slave in a context in which ¡φ is not true given the current sphere of
permissibility, the sphere of permissibility must be adjusted so as to make it true.
Given the possibility semantics of ¡φ, this requires that the sphere of permissibility
be expanded. However—and this is the key point—unlike in the case of order-type
imperatives, there is no simple recipe for expanding the sphere of permissibility.
Crucially, adjusting fsp (w, t) to fsp (w, t) ∪ p will not work, because it would
permit every way of bringing about φ. For example, if the master permits the slave
to take the day off, he would thereby permit the slave to take the day off and kill
the master. But this is clearly not what the master would intend.
Lewis considers various responses to this problem of permission, and one of
them foreshadows subsequent developments in the dynamic approach to imper-
atives. He observes that the expansion of the sphere of permissibility dealt by a
permission sentence should be “minimal,” in that it should bring into permissibility
those impermissible worlds which are, prior to the permission sentence being
uttered, the “least impermissible” (Lewis 1979b, p. 171). In our example, taking the
day off and killing the master is highly impermissible, while taking the day off and
resting is as minimally impermissible as a world can be in which the slave takes
the day off. Thus, when the master utters Take the day off!, only worlds of the latter
type should be added to the sphere of permissibility. Though he thinks this idea
is a step in the right direction, he notes that we critically would need a principle
governing how the comparative permissibility relation itself evolves over time. As
we will see, later work in the dynamic approach solves the problem by replacing the
sphere of permissibility with something that models the relative permissibility of
worlds (see in particular the discussion of Van der Torre and Tan 1998; Van Rooij
2000; Mastop 2005; Portner 2004, 2011b).
Two important works published prior to Lewis’s (1979b) foreshadow later devel-
opments among dynamic theories of imperatives, and I will mention each briefly.
First, van Fraassen (1973) develops an analysis of ought statements based on the
premise that there is, at each point in time, a set of imperatives “in force.” These

 Lemmon () also uses this notion.

i i

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

i i

sentence mood in the dynamic approach 205

imperatives can be seen as very similar to a conversational background in the

ordering semantics theory of modality. He defines the score of a state of affairs
as the set of imperatives in force which it satisfies, and then gives a semantics to
modal sentences very similar to Kratzer’s analysis of weak necessity. Treating states
of affairs as possible worlds and imperatives as sets of possible worlds, his semantics
amounts to:

(86) [[ ought φ ]] = 1 iff for some possible world w ∈ [[ φ ]] , there is no w ∈

[[ ¬φ ]] such that score(w)⊆score(w ).

What is interesting for us about the paper is not the precise semantics for ought,
but the function of imperatives. Concerning the question of how we know what
imperatives are in force, he writes:

. . . First, there are many conceivable imperatives, but only some are in force. The problem
of the ontogenesis of moral imperatives is the problem of what brings imperatives in force.
I assume their sources are legion: conscience, ideals, values, duties, commitments, and so on.
The process must be complicated, because one imperative may be overridden by another
(under given circumstances); this means presumably that one imperative’s being in force
may prevent another’s being in force. In this way, direct orders may cancel standing orders,
and circumstances may take away the authority under which a standing order is the reason
that a certain imperative is in force. (van Fraassen 1973, p. 15)

It seems that van Fraassen uses the term “imperative” to refer to a fundamental
object of the moral universe, and not a type of sentence. He does, however, imply
that members of the imperative clause type can express imperatives. Specifically, he
assumes that Honor thy father or thy mother! can express an imperative (p. 18), sug-
gesting that the imperatives in force can sometimes be identified with the semantic
values of sentences of the imperative clause type. From van Fraassen’s paper, we
can draw the idea that imperative sentences can contribute to a component of the
discourse context (for van Fraassen, the set of imperatives in force) which affects
the semantics of modal sentences with ought. This idea will be developed further
within the dynamic approach.
Kamp (1973) develops an idea very similar to van Fraassen’s as part of a discus-
sion of free choice permission. Kamp’s theory begins with a “set of social, legal and
moral prohibitions to which B is subject” and then analyzes an act of permitting as
the lifting by an agent A of some prohibition (or prohibitions) to which B is subject
under the authority of A. For example, if P0 is a set of prohibitions to which B is
subject under the authority of A, and A uses (88), B’s new prohibition-set will be P1 :

(87) a. P0 = {‘You don’t go to the beach’, ‘You don’t go to the cinema’}

b. P1 = {‘You don’t go to the cinema’}
(88) You may go to the beach!

This change in prohibition-set gives rise to an expansion in the set of worlds

permissible for B. With respect to P0 , no worlds in which B goes to the beach are

i i

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

i i

206 sentence mood

permissible, while with respect to P1 , worlds in which he goes to the beach but not
the cinema are. We see in this way that Kamp gives an analysis of permission which
fits the essential characteristics of the dynamic approach.
From this point, Kamp goes on to discuss sentences with free choice disjunction.
He proposes that (89) lifts both prohibitions in P0 , thus explaining why it leads to
a situation where both going to the beach and going to the cinema are permitted.

(89) You may go to the beach or the cinema.

On the assumption that a disjunctive permission sentence like (89) lifts two
prohibitions, Kamp is able to explain certain cases of the free choice inference. He
offers a definition of entailment under which (89) entails (88) (Kamp 1973, p. 66):

(90) For any sentences s and s , s p-entails s iff in any situation in which both
s and s can be used to grant permissions, the class of actions which would
be moved by a permission statement involving s would include the class of
actions moved by a permission statement involving s .

Kamp’s notion of p-entailment is novel and influential because it explicitly relies

on a dynamic conception of meaning.
Although Kamp’s idea that permission amounts to the lifting of a prohibition
is important, he does not develop a complete analysis of permission sentences.
He seems to assume that a standard analysis of disjunction as set-theoretic union
by itself predicts that (89) removes both prohibitions from P0 , but I do not see
how this is the case, since we can render P0 consistent with (89) by lifting either
one (or both) of the prohibitions. Thus, Kamp cannot be seen as having already
provided a solution to the problem of permission subsequently discussed by Lewis.
For our purposes, the importance of Kamp’s theory is in the way it builds an
analysis of permission sentences based on the effect they have on a feature of the
discourse context. The set of prohibitions in place at a given point in time can
be compared to the set of imperatives in force within van Fraassen’s theory (each
prohibition is an imperative which requires someone not to do something). From
this perspective, van Fraassen’s and Kamp’s ideas foreshadow modern versions of
dynamic semantics and pragmatics in which imperative sentences contribute to
the set of requirements and prohibitions in force at a given point in time.

Dynamic theories of imperatives: the development of contemporary theories.

Ideas about the semantics of imperatives and other directive sentences within the
dynamic approach began to take their contemporary shape at the very end of the
1990s. Building on the earlier work just discussed, we see two main strands of
scholarship. One of these has its origins in dynamic logic. This research has largely
been focused on the logical problems associated with imperatives, such as free
choice permission and Ross’s paradox. The other is more linguistically oriented
and gains much of its motivation from a concern for understanding clause types
and sentence mood. As we will see, these two strands have slowly converged to the
point that they should no longer be seen as separate traditions of research.

i i

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

i i

sentence mood in the dynamic approach 207

We begin with two representative and important works published in the same
year, Han (1998) and Van der Torre and Tan (1998). On the one hand, we have Han’s
dissertation, a study of the syntax and semantics of imperatives focusing on such
issues as the relation between imperatives and negation, the nature of suppletive
imperatives (imperative sentences with the morphological form of infinitives or
subjunctives), embeddability, and the history of the English imperative. On the
other, Van der Torre and Tan’s paper applies Veltman’s (1996) dynamic logic
framework to an analysis of prescriptive sentences. Seen from our perspective of
seeking an understanding of sentence mood, Han’s and Van der Torre and Tan’s
ideas are complementary: Although she touches on some insights concerning
imperatives in dynamic semantics, Han’s major contributions concern the nature
of sentence mood at the syntax/semantics interface. In contrast, Van der Torre and
Tan have an insightful discussion of the type of contextual update which could
represent the imposition of new obligations in discourse, but they do not have
anything to say about the kinds of natural language sentences which might be
associated with those effects. In what follows, we will briefly review both of these
works, using each as a basis for discussing other important research on imperatives
within the dynamic approach.
Han (1998) begins with the hypothesis that the syntax of imperatives contains
a force-indicating operator and a sentence radical expressing a propositional
content. The operator consists of two features, [directive] and [irrealis], where
the former is unique to the imperative clause type and encodes sentential force.
In contrast, an infinitive or subjunctive only has the [irrealis] feature, with no
representation of sentential force. As discussed in Section 3.1.3, she proposes that
the combination of sentential negation and the [directive] feature is ruled out
in certain languages (such as Spanish (28)) because negation would illicitly take
scope over the force-encoding feature. (In other languages, such as English, the
syntax of negation would give it scope unproblematically below [directive].) When
sentences with [directive] cannot be combined with negation, languages employ
another form with [irrealis], with the intended directive force derived by pragmatic
reasoning.
Han explains the directive force encoded by [directive] as follows:

. . . the directive force of imperatives turns the sentence into a directive action, which we
in turn define as an instruction to the hearer to update his/her plan set. A plan set is a set
of propositions that specify the hearer’s intentions, and it represents the state of affairs that
the hearer intends to bring about. (Han 1998, pp. 149–50)

In an obvious way, Han’s perspective on the force of imperatives is similar to

Stalnaker’s analysis of assertion relative to the common ground. She does not,
however, give any more precise semantic/pragmatic content to the notion of plan
set. For example, she does not say what exactly is implied about an individual’s
linguistic or non-linguistic behavior given the fact that some proposition p is
in her plan set. Thus, we are left uncertain of basic matters like what goes
wrong when someone accepts an imperative and then shows no inclination to do
what it says.

i i

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

i i

208 sentence mood

We turn now to the contribution of Van der Torre and Tan (1998). Their
analysis builds on Veltman’s (1996) discussion of defaults in dynamic semantics,
and so, in what follows, I will briefly review Veltman’s logic, following closely the
presentation of Portner (2009, sect. 3.2). As discussed in Chapter 1, the simplest
model of context in dynamic semantics is that of an information state s, which
may be understood as simply a set of possible worlds. In order to account for
expectations, Veltman adds to the information state a second component ε, the
expectation pattern, an ordering of worlds. We have an information state σ =
ε, s, where s represents the factual information in σ and w ≤ε v represents the
judgment in σ that w is at least as expected as v.
The following three crucial concepts can be defined in terms of this new version
of the information state:

(91) (a) Absolutely normal worlds:

nε = {w ∈ W : for all v ∈ W, w ≤ε v}
(b) Maximally normal worlds in s:
mε,s = {w ∈ s : there is no v ∈ s such that v <ε w}
(c) Refinement of an expectation pattern ε with p:
ε ◦ p = {w, v ∈ ε : if v ∈ p, then w ∈ p}

Veltman proposes meanings for two operators within his logic: presumably(φ) uses
(91b) to test whether φ is expected (given the information in s), while normally(φ)
uses (91c) to refine the expectation pattern so that φ becomes a new expectation.
More precisely:

(92) For any information state σ = ε, s,

1. σ [presumably(φ)]M =
(a) σ , if mσ ∩ {w : [[ φ ]] w,M = 1} = mσ ;
(b) †, otherwise.
2. σ [normally(φ)]M =
(a) ε ◦ {w : [[ φ ]] w,M = 1}, s, if nε ∩ {w : [[ φ ]] w,M = 1} = ∅;
(b) †, otherwise.

Notice that presumably is a test: it has no effect on an information state which

passes the test, and reduces to absurdity (†) an information state which does not
pass. Normally refines the expectation pattern, subject to the presupposition that
all existing expectations and the new expectation being added could be satisfied
together.
To account for obligation sentences, Van der Torre and Tan (1998) propose a
deontic state with a structure very similar to Veltman’s information state. A
deontic state δ is a triple δ = W, s, ≤δ , with W the set of possible worlds, s
a subset of W representing factual information, and ≤δ indicating the deontic
relation of comparative “goodness.” The reason for including W in the deontic state

 In Portner (, sect. ..), I sketched an analysis similar to Van der Torre and Tan’s of how

Veltman’s () logic could be applied to imperatives. I was unaware of their work at the time.

i i

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

i i

sentence mood in the dynamic approach 209

arises from the interesting observation that, as part of deontic reasoning, we may
need to talk about both the set of possible worlds which are deemed realizable at a
point in time, and a wider set of options which might go beyond those ever deemed
realizable. In δ, the set s represents the realizable worlds; it is called the context of
deliberation and is akin to Lewis’s sphere of accessibility. The set W, or context
of justification, represents all of the possible worlds which are relevant more
broadly to deontic judgments. The dynamic meaning of a factual sentence affects s
and the dynamic effect of a deontic sentence affects ≤. The context of justification
W is not changed by any updates.
Van der Torre and Tan define four deontic operators, two of which target the
context of justification and two of which target the context of deliberation. Among
each pair, one is similar to presumably in performing a test on the ordering ≤δ and
the other similar to normally in refining the ordering. , 

(93) For any deontic state δ = W, s, ≤δ ,

1. Operators of the context of justification
(a) δ[ideal(p)] =
(i) δ, if nδ ⊆ p;
(ii) †, otherwise.
(b) δ[oblige(p)] =
(i) δ  p, if nδp ⊆ p;
(ii) †, otherwise.
2. Operators of the context of deliberation
(a) δ[ideal∗ (p)] =
(i) δ, if mδ ⊆ p;
(ii) †, otherwise.
(b) δ[oblige∗ (p)] =
(i) δ  p, if mδp ⊆ p;
(ii) †, otherwise.

The operators ideal and ideal∗ act as tests on the deontic state (targeting W and s,
respectively), while oblige and oblige∗ refine the ordering (on the worlds in W and
s, respectively). This differentiation of the operators of the context of justification

 In what follows, I simplify by ignoring conditional obligations. In fact, all four operators take

conditional form, e.g. ideal(α|β). I also note that they give the presupposition of the operators in a more
sophisticated way which takes account of the fact that the ordering may have multiple distinct maxima
and may include infinite chains of ever-better worlds. In this presentation, I simplify the definitions to
the same level as those in ()–().
 We need new definitions of n , m , δ  p, and δ  p:
δ δ
(i) For any deontic state δ = W, s, ≤ δ ,
n δ = {w ∈ W : for all v ∈ W, w ≤ δ v}
m δ = {w ∈ s : there is no v ∈ s such that v ≤ δ w}
δ  p = W, s, {w, v : w ≤ δ v and if v ∈ p, then w ∈ p}
δ  p = W, s, {w, v : w, v ∈ (s × s)∩ ≤ δ and if v ∈ p ∩ s, then w ∈ p ∩ s}
These formulations are simplified and adjusted from Van der Torre and Tan’s to make them more
parallel to (), and as a result are not equivalent to the originals in all settings.

i i

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

i i

210 sentence mood

from the operators of the context of deliberation is related to Thomason’s important

work (Thomason 1981a,b), and it is significant because it allows one to model both
deontic language which pertains to what is obliged independently of whether it can
obtain and deontic language which is relevant only to what is understood to be a
realizable outcome.
Van der Torre and Tan do not make any clear claim that their logic should be
used as a direct analysis of the meanings of any linguistic forms, but it is tempting
to use oblige∗ as a model of the sentential force of imperatives. This amounts to the
idea that imperatives function to refine the deontic ordering within the space of
actually realizable possibilities, subject to the condition that by doing so they create
an ordering in which the ideal worlds (relative to what is considered realizable)
are ones in which the imperative’s propositional content is true. We could also
consider using oblige to give a theory of imperatives and imperative-like sentences
which seem to affect our assumptions about which worlds are preferable setting
aside whether they are realizable. Good examples of this latter type might be absent
wishes like (94a) and counterfactual deontic sentences like (94b) (from Kaufmann
2012; Mastop 2011, respectively):

(94) a. Please, be rich! (on the way to a blind date)

b. Was toch lekker thuisgebleven. (Dutch)
was prt prt at.home.stay-pp
‘You should just have stayed at home.’

Similarly, the ideal operators might be understood as models of the meanings of

descriptive deontic modals. However, Van der Torre and Tan do not describe how
these operators are interpreted in complex sentences, and an analysis based on
them would be at best highly incomplete. More recent work in dynamic deontic
logic, such as Willer (2014), comes closer to providing a plausible theory of deontic
modals.
Following Van der Torre and Tan (1998), there has been other interesting work in
dynamic deontic logic, including Žarnić (2003a,b), van Benthem and Liu (2004),
Yamada (2007a,b,c), and Willer (2014). However, as our goal is to better understand
the significance of this tradition for understanding the nature of sentence mood,
we should focus on dynamic semantics research which aims to address linguistic
issues. Important works which meet this criterion are Van Rooij (2000), Mastop
(2005), and Starr (2013); we will review these authors’ ideas briefly before returning
to research which builds on Han’s more linguistically-oriented ideas.
We begin with Van Rooij (2000). He gives an analysis of imperatives based on
the assumption that their meanings can be represented in terms of command and
permission logical forms Must(p) and May(p). He accepts a dynamic semantics of
command imperatives very close to the oblige operator of Van der Torre and Tan
(1998), but then turns his attention to incorporating permission imperatives.
Recall that the problem of permission, as outlined by Lewis, is that it is not
clear how to express exactly which worlds enter the sphere of permissibility when
a permission is granted. As we saw, an appealing intuition is that May(p) adds the

i i

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

i i

sentence mood in the dynamic approach 211

p worlds which are, prior to permission being granted, the least impermissible.
The framework developed by Van der Torre and Tan (1998) seems promising for
addressing this problem, because the ordering relation ≤δ which is used to determ-
ine the ideal worlds also expresses which non-ideal worlds are better than others. If
we identify the ideal worlds with the sphere of permissibility, permission can then
be formulated as adding the highest-ranked p worlds into the permissibility set.

(95) After permission May(p) is granted in deontic state δ, the permissible worlds
are mδ ∪ {w ∈ p : for no v ∈ p, v ≤δ w}.

While (95) identifies the expansion of the sphere of permissibility in the way
suggested by Lewis, it does not define the deontic state which results from updating
δ with May(p). Crucially, it does not say how ≤δ itself changes. In order to have
a full dynamic analysis, the theory must identify the result of updating δ for a
permission sentence, δ[May(p)].
To make progress on this problem, Van Rooij proposes to make use of an
additional set S of propositions “that potentially determine reprehensibility.” His
idea is that, when permission May(p) is granted, we add p worlds into the ideal set
and then order them with respect to one another on the basis of S (using a technique
of Harper 1976). The problem, however, is that it is not made clear what it means
for a proposition to be in S, and Van Rooij is inconsistent in describing it. At one
point, he says that “the simplest way is to let this set of designated propositions be
the set of atomic propositions”; at another that it “is contextually determined, and
stays constant during the ‘discourse’ ” (Van Rooij 2000, p. 132). However, neither
of these choices seems appropriate, since the set of propositions which determine
reprehensibility could itself change as a discourse progresses—indeed, imperatives
affect the determination of reprehensibility. The analysis of imperatives under the
dynamic approach should be expected to explain how they affect the relation of
comparative reprehensibility.
Mastop’s (2005, 2011) dissertation on imperatives differs from other dynamic
semantics work on the topic in several ways. First, he rejects the claim that
sentence moods should be understood in terms of logical forms consisting of a
force indicator and a propositional content radical. Rather, he proposes that the
semantic values of imperatives are “instructions,” a basic type of semantic object on
a par with propositions. And second, he constructs the states which represent the
evolving context from propositions and instructions, rather than in the usual way
from possible worlds. Inspired by Hamblin (1971), he labels his notion of a context
the “commitment slate” defined as a fact sheet and a commitment function.
The component relevant for imperatives, the commitment function, is defined as
a function from worlds to sets of pairs of the form i, do or i, don’t, where i is

 I carry out this discussion in the formalism developed during our discussion of Van der Torre and

Tan (). Van Rooij presents the ideas in a rather different format.
 Commitment slates have a third component, the consequence function, which we can ig-

nore here.

i i

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

i i

212 sentence mood

an instruction. Intuitively, a commitment function gives, for each world, a set of

collections of do’s and don’t’s, and an agent acts in w in accord with his commitment
function if he does all the do’s and avoids all the don’t’s of any one of the collections
assigned to w.
Mastop’s system is rather complex, and interested readers should study the works
cited for details. For our purposes, it’s important to point out several positive
developments. First, Mastop aims to explain a wider range of linguistic phenomena
than previous work in dynamic semantics. Second, he aims to give an overall
theory of sentence mood based on the idea that imperatives have a unique type of
semantic value, what he calls “instructions,” and not propositions. This proposal
is in accord with the compositional view of sentence mood. Third, he suggests
that the definition of commitment slates as assigning sets of collections of do’s
and don’t’s to each world provides the basis for an understanding of free choice.
And fourth, because within the commitment slate we retain a record of previously
accepted instructions, it is possible to model the retraction of a commitment
as the removal of pairs i, do or i, don’t. This type of retraction might allow
an analysis of permission along the lines of Kamp’s (1973). Mastop’s theory has
important weaknesses as well, however. Most notably, by treating instructions as
sui generis semantic objects, he cannot explain their logical connections to facts
and to one another. More generally, since they are not given a deeper foundation
in a cognitive or discourse model, I find it difficult to understand exactly what
instructions are.
Starr (2010, 2013) develops a dynamic semantics theory of imperatives based on
the operator theory of sentence mood, with a force marker attached to a sentence
radical !φ. Unlike Mastop, he defines his version of the structured discourse context
from possible worlds, but in a slightly different way than Van der Torre and Tan
(1998). Whereas for Van der Torre and Tan, the deontic state crucially involved an
ordering relation over worlds, Starr uses an ordering over propositions. Specifically,
a preference state R is a set of pairs of propositions, where we interpret p, q ∈ R
to mean that p is preferred to q.

(96) a. A preference state R is a set of pairs of propositions.

 set of R, cR , is the union of all of the propositions related by
b. The context
R: cR = {p : ∃a, b ∈ R[p = a ∨ p = b]}

An imperative !φ introduces a preference for the proposition p denoted by φ over

its negation. In a manner similar to the way assertion works in the dynamic se-
mantics of questions, Starr proposes that a declarative φ updates cR by restricting
every proposition in one of the pairs of R and introducing a preference for its
propositional content over ∅.

 Vranas () and Starr () note other serious issues with Mastop’s analysis. Jary and Kissine

() criticize analyses like Mastop’s and Portner’s (discussed in what follows) which assign non-
propositional denotations to imperatives.
 Technically, the sentence radical is given a dynamic semantics, but its intuitive propositional

content can be recovered from its dynamic value.

i i

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

i i

sentence mood in the dynamic approach 213

(97) For any preference state R and sentence radical φ with meaning p,
a. R[!φ] = R ∪ {cR [p]cR [¬p]} ∪ {q[p], q[¬p] : ∃a, b ∈ R[q = a ∨ q =
b] ∧ q ∩ p  = ∅}
b. R[ φ] = {a[p], b[p] : a, b ∈ R ∧ a[p] = ∅} ∪ {cR [p], ∅}

Starr is not clear about the motivation for analyzing imperatives in terms of an
ordering of propositions, as opposed to an ordering of worlds (like Van der Torre
and Tan) or a set of propositions (like Han and others). He suggests that one reason
is the fact that an ordering of propositions allows him to describe a situation where
p is preferred to ¬p and q to ¬q, even though these two preferences don’t add up
to an overall preference for p ∧ q over p ∧ ¬q and ¬p ∧ q. However, this result is
not a matter of including pairs of propositions in the context; we could interpret
a Han-type plan set {p, q} as not implying a preference for p ∧ q. Admittedly,
Portner (2004) uses a construct similar to the plan set in such a way as to prefer
worlds in which all of the individual preferences are satisfied, but as Starr notes,
this could be accounted for by making p ∧ ¬q a preference, not simply p. In any
case, Starr’s actual proposal for imperatives (97) does imply that updating with
!φ and !ψ does normally result in a context preferring φ ∧ ψ-worlds. So it’s not
clear what empirical consequences follow from Starr’s decision to define R as a set
of pairs of propositions. There is a significant technical consequence, however; by
treating the preference relation as a set of pairs, the context set can be recovered
from the preference relation (as in (96b)). In contrast, with a Han-style plan set {p},
we cannot tell which not-p worlds are considered relevant. This is why in Portner’s
treatment which builds on Han’s, the common ground is included explicitly as a
separate part of the context. Perhaps Starr’s main motivation for the way he defines
the preference state is that he wishes to integrate the model of preferences and
the representation of factual information as seamlessly as possible in a single set-
theoretic object.
Finally we turn to Portner’s (2004, 2007, 2011b) dynamic pragmatics theory of
imperatives. At the most essential level, Portner’s analysis is most similar to Han’s.
Where Han has the plan set, understood in psychological terms, Portner proposes a
new component of the discourse context, the to-do list function tdl. For reasons
we will discuss below, Portner treats imperatives as denoting properties, and so, for
each participant a in a conversation, a’s to-do list tdl(a) is a set of properties. The
to-do list defines an ordering of worlds, as follows:

(98) For any worlds w and v,

w <tdl(a) v iff {P ∈ tdl(a) : P(a)(v) = 1} ⊂ {P ∈ tdl(a) : P(a)(w) = 1}

 If the context set were itself included in the plan set, and all of the other propositions in the plan

set were required to be a subset of the context set, we could recover the context set as the union of the
plan set in a way very similar to Starr’s (b).

i i

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

i i

214 sentence mood

A world w is ranked above another v if the agent does a better job of carrying out
his to-do list in w, in the standard sense that in w he has every one of the to-do list
properties he has in v, plus at least one more.
While we can see Portner’s analysis as similar to Han’s (as well as Mastop’s and
Starr’s) in that it keeps track of the contribution of each imperative, this use of
the ordering relation also makes it similar to Van der Torre and Tan’s. Unlike
earlier authors (specifically Han and Van der Torre and Tan), he attempts to give a
pragmatic interpretation to the to-do list. As discussed at the outset of this section,
Portner extends Stalnaker’s theory of assertion and Roberts’ theory of questions in
discourse in the following way (see Portner 2004, 2007):

(99) A Portner context is a triple cg, qs, tdl:

a. The common ground cg, a set of propositions.
b. The question set qs, a set of question-denotations.
c. The to-do list function tdl, assigning a set of properties to each parti-
cipant a.
(100) Agent’s commitment: For any participant i, the participants in the conver-
sation mutually agree to deem i’s actions rational and cooperative to the
extent that those actions in any world w ∈ ∩cg tend to make it more likely
that there is no v ∈ ∩cg such that v <tdl(a) w.

The basic idea here is that the agent’s actions will be judged according to how
well they can be expected to make the actual world turn out as highly-ranked
as possible, given the factual assumptions in the common ground. For example,
if I agree to give you a piece of bread, and then eat it, my actions would not be
expected (given the common ground) to maximize the satisfaction of my to-do
list, and you are justified in judging me negatively—as one who is either confused
or dissimulating.
As discussed in Section 3.1.3, a major goal of Portner’s work is to argue in favor of
the compositional approach to sentence mood. He assumes that each major clause
type has a characteristic semantics. Declaratives denote propositions, interrogat-
ives sets of propositions, and imperatives properties restricted to the addressee.
For example, Leave! denotes the property of leaving, with the presupposition that
its subject argument is the addressee:

(101) [[ Leave! ]] c = [λxλw : x = addressee(c) . x leaves in w]

Given his assumptions about the discourse context and the semantics of clause
types, Portner hypothesizes that the sentential force of each sentence mood is
determined by a pragmatic principle which uses the denotation of sentences with
that mood in the simplest, most natural way to update the discourse context. The
sentential force of declaratives is assertion (adding to the common ground) because

 Portner () uses the < relation in the opposite direction, but I reverse it here to make the

presentation more similar to other theories discussed in the section.

i i

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

i i

sentence mood in the dynamic approach 215

Table . Main contributions of dynamic theories of imperatives

Han imperatives add to a set of imperative-meanings;

feature-based analysis of imperative operator
Van der Torre and Tan dynamic semantics framework;
imperatives update an ordering relation;
context of deliberation vs. context of justification
Portner dynamic pragmatics framework;
pragmatic interpretation of ordering relation;
compositional approach to sentence mood
Mastop multiple action plans represent choice;
operation of retraction
Starr encoding of context set within ordering relation

the denotation of a declarative is a proposition and the most simple and natural
way to incorporate a proposition into a context of the form (99) is to add it to cg.
Similarly, the sentential force of imperatives is “requiring,” defined as updating the
addressee’s to-do list, and the sentential force of interrogatives is asking.
This overview of Portner’s theory concludes our discussion of recent theories
of imperatives within the dynamic approach. Before moving on to consider some
other relevant work, I summarize in Table 3.3 what I take to be the main ideas
or contributions which each one first brought into the literature. It should be
noted that several of these authors have important ideas about other topics beyond
our scope here, such as conditional imperatives, Ross’s paradox, and imperatives
conjoined or disjoined with declaratives.

Actions. Building on Segerberg (1990), a few papers in the dynamic tradition

propose that imperatives denote actions, defined as functions from worlds to
worlds (e.g. Lascarides and Asher 2003; Barker 2012). The idea is that an imper-
ative tells you ways of making a transition from worlds at which the action has not
been performed to other worlds at which it has. Taking an example from Barker,
assume that in world w1 , Mary can eat one apple and thereby arrive at w3 , or eat
a different apple and arrive at w4 . In world w2 there is just one apple, and if she
eats it she arrives at world w5 . In this context, the action of Mary eating an apple is
represented by the following function:

(102) the action of Mary eating an apple = {w1 , w3 , w1 , w4 , w2 , w5 }

 If the same basic idea was discovered more than once, I attribute it to the author who was, to the

best of my knowledge, the first to introduce it.

 The notion of action here is similar to speech acts as conceived by Krifka (), discussed in

Section ...
 Note that the notion of “possible world” here cannot be the standard one of a complete way things

could be. In a standard possible world, Mary either eats the apple or she doesn’t, and nothing she does
will move her to a different world. Barker must assume a different notion of possible world, for example
sets of possible worlds in the standard sense.

i i

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

i i

216 sentence mood

The semantic value of the imperative Eat an apple!, addressed to Mary, is this action.
As Barker points out, the idea that imperatives denote actions has the immediate
benefit that it explains why imperatives are not naturally described as true or false.
(Of course, all other dynamic theories, with the possible exception of Han’s, have
the same benefit.)
Action-based analyses of imperatives are not entirely explicit about the nature of
directive force. Working within a DRS framework, Lascarides and Asher say “the
defining characteristic of a discourse which includes a commanded imperative is
that its ccp changes the input world into an output one where the action has been
performed” (Lascarides and Asher 2003, p. 6). This makes it sound as if adding
the imperative into the DRS amounts to performing the action which it specifies.
Barker suggests a version of the compositional approach to sentence mood:

There is no direct update effect. If uttering an imperative causes the addressee to believe that
the speaker desires for an action to be performed, and if the addressee is inclined to fulfill
the desires of the speaker, the imperative may influence the behavior of the addressee, not
through grammatical regulation, but through simple pragmatical reasoning, very much in
the way that thrusting a broom into the hands of an idle person and pointing at some dirt
can cause the thrustee to behave as if they were newly under an obligation to begin sweeping.
(Barker 2012, p. 60)

This analogy is not strong enough to explain sentential force, however. I could
imply many different things by the act of handing you the broom and pointing
at the dirt; maybe I just wish to show you that I have a broom and imply that
I will sweep up later. The link between imperative form and the directive function is
tighter than this, but one could imagine integrating his ideas into a more complete
pragmatic explanation.
The main goal of Barker’s analysis is to solve Ross’s paradox and the problem of
free choice disjunction. He points out that the action expressed by Eat an apple! is
a subset of that expressed by Eat an apple or a pear!, but by itself this does not
solve the problem. Of course this subset relation also exists when propositions
are disjoined, and Ross’s paradox is the fact that a subset does not amount to an
entailment.
Barker advocates an analysis of permission which makes (103a) mean (103b):

(103) a. You may eat an apple.

b. All worlds which can be reached by your eating an apple are ok.
(104) You may eat an apple or a pear.

On such a semantics, (103a) does not entail (104), but from what I understand this
result does not depend on the action-based semantics. Example (103b) could be
stated in terms of propositions as “all accessible worlds in which you eat an apple
are ok.”
Overall, these works which treat imperatives as denoting actions offer an in-
teresting version of the idea that imperatives differ from declaratives semantically,

i i

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

i i

sentence mood in the dynamic approach 217

and not just in terms of their force. This idea fits naturally within the compositional
approach to sentence mood. However, they have not yet given a detailed analysis
of what that force is, and we have not yet seen an argument that actions will lead
to improvements over existing dynamic analysis.

Notes on free choice, permission, and Ross’s paradox. Much research on im-
peratives takes as a primary goal to explain the free choice permission and Ross’s
paradox. While modal theories of imperatives, like those of Aloni (2007) and
Kaufmann (2012, 2016), aim to reduce free choice in imperatives to a general
explanation of free choice in modal sentences, scholars who favor dynamic theories
often argue that the specific analysis they propose explains these phenomena.
Unfortunately, by and large such arguments are not made in the context of
comparing one proposal about free choice to another—that is, we sometimes see
such an argument presented without any detailed discussion of how the given
explanation differs from and is better than other dynamic theories’ explanations.
In this section, I will consider how a number of dynamic theories approach the
problems of free choice and Ross’s paradox, with the goal of understanding what
some of the options for developing a theory are in this area.
The earliest works we have discussed in this section, Han (1998) and Van der
Torre and Tan (1998), do not address free choice or Ross’s paradox. They are
major preoccupations of Van Rooij (2000), however. As we saw above, Van Rooij
develops Lewis’s intuition that permission works by expanding the sphere of
permissibility to include the least-reprehensible worlds in which the proposition
being permitted is true. He points out that, with a disjunctive permission sentence
with logical form May(p ∨ q), if p and q are equally reprehensible (more precisely,
if the least reprehensible p worlds and the least reprehensible q worlds are equally
reprehensible), then expanding the sphere of permissibility to include the least
reprehensible p-or-q worlds will lead to both p worlds and q worlds being included.
Thus, (105), used performatively, leads to a context in which both eating an apple
and eating a pear are permitted:

(105) You may eat an apple or a pear.

(106) a. You may eat an apple.
b. You may eat a pear.

Van Rooij also implicitly addresses Ross’s paradox when he cites Kamp’s (1973)
definition of p-entailment.

(107) For all permission sentences p and q and for every context c,
p p-entails q iff c + p = (c + p) + q.

Given this definition, (105) p-entails (106a), but not vice versa.

 He also notes the similarity of the definition to a more general definition of validity given by

Veltman ().

i i

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

i i

218 sentence mood

Although Van Rooij gives insightful explanations of free choice and entailment
patterns with permission sentences, he does not address the problems as they arise
more generally with imperatives. It is also important to note that Van Rooij uses a
classical boolean semantics for disjunction; disjunction of sentences is interpreted
as a union of propositions, and free choice is derived from an additional pragmatic
stipulation which requires that the disjuncts be of equal reprehensibility. In contrast
to this, later work on free choice and Ross’s paradox in the dynamic approach
usually assumes a non-standard “choice-offering” semantics for disjunction.
On such approaches, disjunction of some category X yields a set of alternative
meanings of the type normally assigned to X. For example, the phrase Noah or
Ben might have the denotation {n, b}. Choice-offering theories of disjunction are
often incorporated into a more general Hamblin-semantics framework on which
all phrases are taken to denote sets of meanings of the type they would normally be
associated with. For background on these ideas, which intersect with much broader
issues in the theory of scalar implicature and free choice, see (among many other
important works) Hamblin (1973), Zimmermann (2000), Kratzer and Shimoyama
(2002), Aloni and van Rooij (2004), Chierchia (2004), Simons (2005), Menéndez-
Benito (2005), Geurts (2005), Alonso-Ovalle (2006), Aloni (2007), Fox (2007), and
Aloni and Ciardelli (2013).
Mastop (2011) is one scholar whose ideas about free choice with imperat-
ives depend on a choice-offering analysis of disjunction. According to Mastop,
a disjunctive imperative creates a commitment function with multiple sets of
instructions. For example, starting with an empty commitment function s and
accepting (108), we get the commitment function in (109):

(108) Post this letter or burn it!

(109) For every world w, s[p ∨ b](w) = {{p, do}, {b, do}}

Though he does not discuss the point clearly, I believe Mastop’s intention is that this
commitment function represents a state in which the agent has the choice between
posting the letter and burning it. More generally, in a world w, the agent has the
choice to follow any of the sets of do’s and don’t’s given by the commitment function
in w. This framework can also avoid Ross’s paradox. Mastop demonstrates this
point by showing that accepting ¬b! in the context of (109) yields a contradiction
(since b, don’t is added to both sets in (109)).
Portner’s (2011b) analysis of free choice and permission is in some ways similar to
Mastop’s. He suggests that contexts in which an agent is permitted but not required
to do something should be represented via a to-do list which includes each option
for what he may do. For example, if both Turn left! and Turn right! are incorporated
into a’s to-do list (and assuming that a can’t both turn left and turn right), the

 I also point out Aloni and Ciardelli’s () work on free choice imperatives. It builds on a

choice-offering semantics for disjunction, and is designed to fit into a dynamic semantics approach to
imperative meaning (specifically, within inquisitive semantics); however, it does not develop a dynamic
treatment or explain the properties of imperatives in terms of their sentential force.

i i

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

i i

sentence mood in the dynamic approach 219

ordering of worlds derived on Portner’s analysis has two disjoint sets of “best”
worlds, one in which a turns left and one in which a turns right. Since the agent’s
commitment principle requires that a take actions which tend to make the actual
world one which is not worse than any other world, a is free to either turn left or
turn right.
Portner’s understanding of permission fits naturally with a choice-offering ana-
lysis of disjunctions. Assuming (108) denotes a set of exclusive properties, as in
(110), and that each alternative is added to the to-do list, that to-do list will give an
ordering with two maxima.

(110) {{P ∩ ¬B}, {B ∩ ¬P}}

Thus (108) normally leads to a context in which the addressee is permitted to post
the letter, is permitted to burn it, and is required to do one or the other. Adopting
a notion of consequence very similar to p-entailment (which he labels “warrant”),
Portner can then solve Ross’s paradox in essentially the same way as Mastop: (108)
p-entails (and warrants) Post this letter!, but not vice versa.
Starr (2013) criticizes Portner’s analysis by stating that “The fact that a reasonable
speaker can place inconsistent properties on the To-Do List suggest[s] that it really
isn’t a To-Do List at all, that is, it is not the thing which is guiding agents’ actions
and practical reasoning.” On the one hand, he is making the good point here that
while the ordering relation is essential to the analysis as given above, the to-do list
itself is not; on the other, he subsequently suggests that the possibility of having an
inconsistent to-do list makes Portner’s whole construct of to-do list incoherent, and
I do not see why that should be. Whether empirically correct or not, the definitions
of ordering and agent’s commitment determine exactly what an inconsistent to-do
list represents. (Note that as this book was going into production, Starr released a
revised version of his paper.)
Starr has a more significant objection based on the observation that, on Portner’s
analysis, a disjunction warrants each of its disjuncts. This might be problematical,
because it suggests that (111) should be equivalent to (108). Yet one can read the se-
quence of imperatives (111) as in the end requiring that the addressee burn the letter.

(111) Post the letter or burn it! Burn it!

The fact that the final sentence of (111) has a discourse effect shows that Burn
it! cannot be interpreted in Portner’s theory by the ordinary update operation.
Portner (like Mastop) would assume that the final sentence causes a retraction of
Post the letter! (={P ∩ ¬B}) from the to-do list, not an addition of Burn it! (An
addressee would infer that Burn it! must be interpreted as proposing a retraction,
since it would have no effect on the context otherwise.) In contrast, Starr proposes
a system in which the sequence can be interpreted by the normal update function
and still lead to a context different from that derived from (108) alone. The details
of his derivation are complex, however, so I will not attempt to present them here.
For our purposes, the key point is that disjunction is not interpreted as a union
of propositions, but rather as the union of the dynamic updates performed by the

i i

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

i i

220 sentence mood

disjuncts individually. That is, (108) says to accumulate all of the effects we would
get by updating the state by Post this letter! and by Burn it! This move has a clear
similarity to the way in which the choice-offering analysis of disjunction is used in
the other dynamic theories of imperatives.

3.3.4 Minor types: optatives and exclamatives

There has been a limited amount of research on minor clause types within the
dynamic approach. In this section we will look briefly at some proposals about
optatives and exclamatives.

Optatives. Several scholars have studied optatives from the perspective of the
dynamic approach (Biezma 2011; Grosz 2011, 2014). Even though they do not
identify optatives as a distinct sentence mood, both Biezma and Grosz have
interesting ideas concerning the derivation of optative meaning. Biezma builds on
Roberts’ version of the structured discourse context, proposing that if optatives like
(112) receive an optative reading because they answer a certain kind of question
under discussion.

(112) If only John had driven more carefully!

Biezma argues that the compositional semantics of (112), specifically the contribu-
tions of if and only, give it a meaning which is appropriate to answering the question
under discussion “How could John have gotten to the meeting on time?” (or, more
precisely, “What X is such that, if X, John got to the meeting on time?”). Given that
it answers this question, the implication that it is desirable that John drive more
carefully follows from the desirability of the consequence for which it is a sufficient
condition (that John got to the meeting on time). On Biezma’s view, optatives do not
constitute a true clause type, but it is an interesting question whether her approach
implies that they are a sentence mood. If the job of answering a question of the
kind which leads to the optative meaning is a fundamental type of update within
semantic/pragmatic theory, we could see the if (only) pattern as a grammatically
complex sentence mood.
Grosz (2014) analyzes a broader range of optative patterns and presents an
intriguing game-theoretic model of how the optative meaning is marked. Unlike
Biezma, who explains the presence of if and only in the optative (112) on the
basis of their semantic contributions, Grosz thinks they are examples of truth-
conditionally vacuous optative particles. Other such particles include just and but
in English and nur, bloß, and doch in German. He proposes that their presence is
explained not by a standard semantics, but rather by the role they play in context to

 As this manuscript was being finalized, a new article on permission by Starr () appeared.
 As mentioned above, there is also relevant research on promissives and exhortatives; see Section
.. and Portner (), Mastop (), and Zanuttini et al. (). Within the dynamic approach,
these minor types are treated as closely related to imperatives (or they are joined with imperatives into
a broader jussive clause type), and so I will not review analyses of them here.

i i

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

i i

sentence mood in the dynamic approach 221

signal the speaker’s intention that his or her utterance receive an optative meaning,
as opposed to other meanings it might receive in context. In other words, the
particles serve as “cues” for the optative reading. Grosz argues in favor of this
perspective by showing that optative particles can only be left out in contexts in
which an optative meaning is highly expected. Grosz formalizes his analysis as a
signaling game using the techniques explored by Franke (2009).
Grosz’s theory relies on language users being aware that the optative meaning
is one alternative interpretation available to sentences like (112), and in order to
assess the significance of Grosz’s approach to our understanding of sentence mood
generally, we would need to know exactly how the optative meaning fits into a
broader paradigm of possible conversational functions. One possibility is to take
as a fundamental fact about language that there is a specific inventory of such
meanings, for each of which there are cues which language users may employ
to signal their communicative intentions. Such a perspective can be seen as a
game-theoretic analogue of the traditional view of sentence types (see Section
3.1.1), and would fit well with the construction-based approach to the marking of
sentential force. Alternatively, it could be that optative meanings are cued in this
way exactly because of their peripheral status within clause type systems. On this
view, the basic clause types would be associated with their sentential forces in a
principled way (for example, with an operator or through a pragmatic principle
based on their compositional meaning), while minor types would be learned in
a language-specific way on the basis of probabilistic associations. We are clearly
in no position to decide if either of these possibilities is correct, but it does seem
likely that game-theoretic ideas will be explored more extensively in future work
on sentence mood.

Exclamatives. A number of works have analyzed exclamatives using ideas drawn

from dynamic semantics and pragmatics. In this section, we will look at the
proposals of Castroviejo (2006a,b) and Zanuttini and Portner (Zanuttini and
Portner 2000, 2003; Portner and Zanuttini 2000, 2006).
Both Castroviejo and Zanuttini and Portner follow the compositional approach
to sentence mood. Focusing on wh-exclamatives in Catalan like (113a), Castroviejo
proposes that exclamatives are degree constructions containing a special high-
degree morpheme tan. Example (113a) has the truth conditions (113b) (Castroviejo
2006a, p. 141):

(113) a. Quina pellícula tan entretinguda que vaig veure!

which movie so entertaining that go.1sg see
‘What an entertaining movie I saw!’
b. ∃x[movie(x) ∧ TAN(entertaining(x))(di ) ∧ see(s)(x)]

Castroviejo cites the ontology of Ginzburg and Sag (2001), taking up their proposal
that exclamatives denote facts (see Section 3.1.3). However, unlike Ginzburg and

 Readers interested in exclamatives will want to connect this section to the discussion of Rett’s

theory of exclamatives in Section ...

i i

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

i i

222 sentence mood

Sag, she does not adopt a performative analysis. Instead, she derives the discourse
function of the exclamative from its semantic type. On the assumption that facts
cannot be asserted, asked, or ordered, it is not possible to assign exclamatives to
any of the functions of the basic sentence moods. Instead, the assigned meaning
is an attitude expressed by the speaker. Castroviejo (2006a, p. 142) describes this
attitude as follows:

(114) Contribution: the speaker experiences an attitude towards

∧ TAN(entretinguda(x))(d )
i

Castroviejo does not explain exactly how this particular contribution is derived
from the factive semantics, but she does say that it is “not verbally encoded, but
implicated” (Castroviejo 2006a, p. 144). This perspective identifies Castroviejo’s
analysis as an example of the compositional approach to sentence mood.
Zanuttini and Portner (2003) also give an analysis of exclamatives on which their
discourse function is derived from their ordinary semantic meaning, and they are
more explicit than Castroviejo in the consequences of their analysis for the general
theory of sentence mood. Through a detailed discussion of wh-exclamatives in
Italian, Paduan, and English, they argue that all such clauses contain a factive
operator in CP. Their proposal is that the combination of factivity, encoded by
this operator, and the standard semantics of wh-clauses is sufficient to identify the
sentential force of exclamatives. I will sketch how their system works below.
The force of exclamatives, labeled widening, can be given informally as follows
(see Zanuttini and Portner 2003, (32) for their official definition):

(115) Widening: Widen the initial domain of quantification, D1 , of the exclam-

ative wh-operator to a new domain, D2 , in such a way that D2 − D1 only
contains objects higher on the salient scale than the members of D1 .

Applied to Castroviejo’s example (113a), widening affects the set of movies over
which Quina pellícula ranges: initially, it contains some set of movies which the
speaker is presupposed to have seen; after widening, it contains some more movies,
which the speaker is also presupposed to have seen, and which exceed the initial
set of movies in terms of how entertaining they are. This shift is meant to capture
the idea that the most entertaining movies which the speaker saw were in fact
so entertaining that they were not even under consideration in the initial, pre-
widening context. From widening, Zanuttini and Portner derive more specific
exclamative meanings like “surprise” as pragmatic inferences.
Several authors have criticized Zanuttini and Portner’s theory because it does not
give a central role to degrees in the semantics of wh-exclamatives (e.g. Castroviejo
2006a,b; Rett 2008a,b), and in fact the discussion of (113) above does make an
unwarranted leap from the idea that certain movies were not initially in the domain
of quantification to the statement that they were not because they were so highly
entertaining. The adjective ‘entertaining’ does not make any explicit contribution
to the direction of widening in Zanuttini and Portner’s analysis. Assuming the
criticism is valid, they might want to reformulate their theory in such a way that

i i

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

i i

theories of clause type systems and sentence mood 223

widening can take account of an adjective or other scalar term within the wh-
phrase whose domain is being widened.
Setting aside these concerns about the definition of widening, we can see
how Zanuttini and Portner’s analysis exemplifies the compositional approach to
sentence mood. The sentential force of exclamatives, namely widening, is not
syntactically encoded in all exclamatives. (They leave open the possibility that
some exclamatives represent widening—for example, the very of English examples
like How very cute she is!—but are clear that not all do.) Rather, they think that
factivity rules out the use of the exclamative with the force of a declarative or
interrogative clause, and assume that, as a result, widening is assigned as the force.
Therefore, they clearly adhere to a compositional derivation of sentential force
within a broadly dynamic pragmatics approach.

3.4 Theories of clause type systems and sentence mood

Having seen how scholars have sought to explain the nature of sentence mood and
analyze the form and function of the most well-known clause types, we are now
in a position to step back and ask a broader question: Why do languages have the
sentence moods they do? Speech act theory and the dynamic approach both offer
interesting perspectives on this question, and our main goal in this section will be
to understand and evaluate those perspectives. Before we can do this, however, we
must determine what facts a theory of clause type systems should explain. Here is
a list drawn partially from the discussion throughout the chapter:

1. There are powerful crosslinguistic generalizations, including universals, con-

cerning what clause types and sentence moods exist in natural languages.
2. There are three basic clause types, declaratives, interrogatives, and imperat-
ives, and each of them is associated with a sentence mood.
3. The basic types may not all be equally basic. Some evidence suggests that de-
claratives are functionally the most basic, while imperatives are syntactically
the simplest.
(a) Declaratives are often assumed to show the most basic syntactic pattern
of a given language.
(b) Declaratives have the widest range of functions. They can easily be used
to ask questions and issue directives.
(c) On the one hand, morphologically distinct imperatives often have the
least marking of verbal inflection, and this may mean that they have
the simplest syntactic structure. On the other, it is very common for
imperatives to share features of the morphology and syntax of some
subclass of declaratives (e.g. future tense) or of non-indicative clauses (e.g.
infinitives and subjunctives).

 They might retain the idea that the domain is a set of entities, but explicitly widen it according to

the scale of degrees given by the adjectives. Or they might say that the domain being widened is itself a
set of degrees, expanded to include higher degrees.

i i

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

i i

224 sentence mood

(d) Imperatives have the narrowest range of functions. They can barely be
used to assert information or ask questions.
4. The relations among the three basic types are controversial.
(a) Declaratives and interrogatives are alike in many languages in the
morphosyntax of their (finite, indicative, tensed) verbs. This may show
that these types are more closely related to one another than either are to
imperatives.
(b) According to many syntactic theories, declaratives in a given language
show its default sentential syntax, and interrogatives and imperatives are
marked as distinct from declaratives by morphosyntactic means. This
may show that interrogatives and imperatives are alike, and different from
declaratives, in having an explicit marking of force.
5. Members of each clause type can be used in diverse ways, and though much
of this diversity is certainly the result of implicature and other forms of
inference, for some uses it is plausible that the type’s normal sentential force
is absent. Some examples:
(a) Rising declaratives might be better classified as having the force of asking
(like canonical interrogatives).
(b) Rhetorical interrogatives might be better classified as having the force of
assertion (like canonical declaratives).
(c) Imperatives used to give instructions or express idle wishes might be
thought to lack directive force.
6. There are minor clause types as well, and each of them has special properties
which should also be explained:
(a) Exhortatives are related to imperatives, and are fairly common.
(b) Promissives may be related to imperatives or declaratives, and are quite
rare.
(c) Optatives are relatively rare as a distinct clause type, and there is debate
about whether they are closely related to any basic type; constructions
with optative meaning are less rare.
(d) Exclamatives often have much in common with interrogatives, though
types which are more similar to declaratives are also found. Exclamatives
are not especially uncommon.
7. Certain forms of the clause types, such as morphologically distinct imper-
atives and verb-second declaratives and interrogatives, are the standard root
clause forms of their sentence moods, but show restrictions in their ability to
be embedded.

Most work on sentence types and clause type systems has assumed that a sentence
can only be of one basic type; that is, they assume that the types are exclusive.
However, it has been observed that it may be possible for clauses to be simultan-
eously of the imperative and interrogative clause types (Platzack and Rosengren
1994; Reis 1999; Kaufmann and Poschmann 2013). An example is the echo question
imperative (116b), from Kaufmann and Poschmann (2013, (8)):

i i

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

i i

theories of clause type systems and sentence mood 225

(116) a. Maria to Peter:

Gib dem Lehrer mein Buch!
give.imp the teacher my book
‘Give my book to the teacher!’
b. Hans to Peter:
WEM gib ihr Buch?
who.dat give.imp her book
‘WHO should you give her book to?’

As Kaufmann and Poschmann note, the existence of mixed types like this seems
to offer support to the compositional view of clause typing. From the per-
spective of this section, we would wish to determine which combinations of
types are possible, and then ask why just those cases allow for exceptions to
exclusivity.
When considering clause type systems, our overall goal is to understand what
various ideas about the sentence mood can contribute to explaining properties of
clause type systems like these. We will begin with the two major frameworks we
have considered all along, speech act theory and the dynamic approach, in Sections
3.4.1–3.4.2. Each of these has some interesting consequences for understanding the
most obvious properties, Points 1, 2, 5, and 6 above. The other properties require
some basic linguistic analysis to become apparent; we will discuss them, Points 3,
4, and 7, in Section 3.4.3.

3.4.1 Clause type systems in speech act theory

We have already seen in Section 3.2.2 that numerous linguists argue that clas-
sical speech act theory does not provide a strong foundation for explaining the
properties of clause type systems or developing a theory of sentence mood. This
point is based on the assumption that sentential forces are to be identified with
illocutionary forces, and follows from three basic features of the theory: First,
explicit performatives can be associated with a wide variety of—possibly any—
illocutionary force, even though they are grammatically of the declarative clause
type. Second, even if we set aside uses which are plausibly “indirect” or the result
of non-literal language use, every clause type can be used with a very large number
of distinct forces. This fact implies that clause types are not closely associated
with illocutionary forces, and in fact makes it difficult to maintain that there are
such things as sentence moods at all. And third, the taxonomy of illocutionary
forces does not align well with what we know about clause type systems. Notably,
the traditional taxonomy includes a separate class for the very rare promissives
(commissive acts) but groups interrogatives with imperatives (as directive acts),
even though they are not a natural class in structural terms.
In response to these points, followers of speech act theory have pursued a
number of different strategies:

i i

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

i i

226 sentence mood

1. They identify sentential force with a set of illocutionary forces, an illocution-

ary force potential (e.g. Bach and Harnish 1979; Harnish 1994).
2. They decompose the concept of force into more primitive properties, and
identify one of them with sentential force (Searle and Vanderveken 1985;
Vanderveken 1990).
3. They reduce the number of fundamental forces within the theory, as is done
with the dynamic force hypothesis (e.g. Kaufmann 2012; Charlow 2013).

The first strategy above addresses the problem raised by the diversity of illocution-
ary forces with which each clause type may be associated, but by itself does nothing
to help with the other two problems. Concerning explicit performatives, it can be
combined with the idea that explicit performatives have the same literal force as
other declaratives (assertion or declaration), and their apparent force is derived
pragmatically. It does nothing at all to deal with the overall mismatch between the
speech act typology and structure of clause type systems.
The second strategy, decomposing illocutionary force, addresses the first two
problems in useful ways. Concerning explicit performatives, it identifies an illocu-
tionary point of declarations which is inferentially related to assertion; it might
make sense, then, that the declarative clause type is associated with both the
declarative and the assertive illocutionary points. However, it does not answer
the problem posed by the mismatch between the typologies of forces and clause
type systems. The theory still postulates a basic sentential force for promissives
while interrogatives are still grouped with imperatives. Searle and Vanderveken
nevertheless maintain their typology, holding that it is sufficiently well-motivated
by fundamental philosophical considerations that these discrepancies should be
explained away on functional grounds. (Recall from Section 3.2.2 that speech act
theorists have argued without empirical evidence that promises are too rare or too
unimportant to be associated with a grammatically distinct clause type.)
The final strategy for adjusting classical speech act theory is represented by the
dynamic force hypothesis. The idea here is that there are only a few basic forces
because there are just a few simple, natural ways for a speech act to affect the
discourse. The sentential forces of declaratives, interrogatives, and imperatives are
drawn from these simple, natural updates.
Although the strategy based on the dynamic approach (including the dynamic
force hypothesis) appears to provide the best basis for explaining why we have the
basic clause types we do, it is not clear which approach has the advantage when
it comes to minor types. The traditional classification of speech act theory makes
room for all sorts of speech acts, including lamentation, promising, expressing of
wishes, and exclamation; as a result, the meanings of all of the minor types can
be precisely described. The main problem which this approach faces with regard
to the minor types is explaining why certain illocutionary forces such as threats
and lamentations are never (as far as we know) associated with a clause type. This
situation contrasts with the one we find with the dynamic force hypothesis and
dynamic approach. There it is easy to explain why certain forces are not associated
with minor types. The problem is to incorporate into the theory each minor type
that does occur into a framework which is designed to severely restrict the range

i i

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

i i

theories of clause type systems and sentence mood 227

of sentential forces. In the long run, traditional speech act theory will have the
advantage if the minor types are very diverse across languages, so that we conclude
that any illocutionary force can in principle be associated with a syntactic structure,
even if some do so only very rarely. The dynamic approaches will have an advantage
if the set of possible minor types is severely constrained, so that we can extend the
approach which aims to explain the typological properties of the basic types to the
minor ones.

3.4.2 Clause type systems in the dynamic approach

Within the dynamic approach, it is quite common to present an explanation for the
existence of three basic clause types. The core idea is that there is a notion of context
relevant to semantic and pragmatic theory which encompasses less than the full
richness of the situations in which communication happens. Instead, we have a
structured discourse context representing the most fundamental aspects of those
situations. The structured discourse context always contains some component used
to represent conversational commitments concerning the way the world is (e.g.
commitment slates or common ground). As the approach has developed, it is
now assumed to contain a component or structural features which allow it to
encode both the questions under discussion in discourse (e.g. the question under
discussion set) and priorities for action to which the participants in the discourse
are committed (e.g. to-do lists). A simple form of this idea takes the context to be
a triple common ground, question set, to-do list function, and its structure is the
basis for explaining why there are three basic clause types. In this framework, the
sentential forces of the basic clause types are understood to be the most natural
update functions which can be defined for each of the three components. For
example, if the context is such a triple, as on Portner’s (2004) view, each component
has one clause type specialized for updating it. Murray and Starr (2016) develop a
similar idea in a system which extends Starr’s (2013) theory of imperatives.
Although theories within the dynamic approach virtually all offer (or implicitly
assume) this same explanation of the fundamental role played by the basic types
in communication, there is much less consensus about how to deal with non-
canonical functions of the basic types and the meanings of minor types. Concern-
ing the non-canonical functions, the natural impulse within the dynamic approach
is to see diversity in function as arising from implicature and inference. In other
words, the literal force hypothesis remains a powerful draw within the dynamic
approach—as it would have to, given that the sentential forces are so few and so
simple. An example of this tendency is Portner’s (2004) proposal that the various
functions of imperatives differ not in force, but in the social basis which justifies
their being issued; on this view, for example, orders and requests have the same
force (update the to-do list), but orders are justified by authority while requests are
justified by the fact that they would help the speaker achieve a goal. However, there
are cases in which literal force combined with implicature and inferences does not
seem sufficient to explain a sentence’s discourse effects. Two clear examples here
are rising declaratives and rhetorical questions, cases where linguistic features play
an important role in the association of a basic type with a non-standard function.

i i

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

i i

228 sentence mood

Scholars within the dynamic approach have not reached a consensus on the analysis
of rising declaratives and rhetorical questions, but it appears to me that the trend at
present is to follow Gunlogson’s tactic of reanalyzing the basic sentential forces in
ways which allow these peripheral functions to be modeled as natural updates, even
if they are rarer than the canonical functions of asserting, asking, and directing.
When it comes to the minor types, the dynamic approach has not settled
on a standard treatment. In Section 3.3.4, we saw various interesting ideas for
the analysis of specific minor types, but many problems remain. In the cases of
exhortatives and promissives, we have a reasonable explanation within the model
of the structured discourse context for why they are expressed as clause types,
but not for why they are rarer than imperatives. In the cases of exclamatives and
optatives, we do not have any analysis which relates the way they update the context
to models of the structured discourse context used to explain the basic types. The
dynamic approach should seek a theory of exclamative and optative updates which
can be expressed in a natural way within the framework used to analyze the basic
types, yet which makes their updates less essential in a way that explains their
non-basic status.
In sum, then, dynamic theories present a promising route towards explaining
those properties of clause type systems which have to do with the canonical
functions of the basic types. But while there is much interesting work on the
non-canonical functions and minor types, no generally accepted theory of those
functions and types has yet emerged. As mentioned above, the issues facing the
dynamic approach are in many ways the converse of those facing the speech act
theory approach: while speech act theory leads us to expect that very many minor
functions of language are equally natural, dynamic theories lead us to see only a
very few functions as maximally natural. It seems to me that, overall, the dynamic
approach is in a much better position to explain the full range of properties of
clause type systems than speech act theory is, since I think it will prove easier for
the dynamic approach to elaborate upon its simple model of the discourse context
so as to allow natural updates to be associated with the minor types, without every
sort of conceivable update being treated as equally natural. In contrast, apart from
the versions of speech act theory which follow the dynamic force hypothesis (and
which follow a strategy similar to that of the dynamic approach), speech act theory
has few resources internal to the theory for explaining why any particular force
is not associated with a characteristic grammatical form. In the end, an extremely
important difference between the dynamic approach and versions of speech act
theory which accept the dynamic force hypothesis, on the one hand, and other
more standard versions of speech act theory, on the other, is that the latter see
many properties of clause type systems as beyond the explanatory purview of
semantic/pragmatic theory, while the former continue to seek a theory of those
properties.

 Of course, this evaluation accords with my own previous work, but still I think my reasons for this

evaluation are sound.

i i

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

i i

theories of clause type systems and sentence mood 229

3.4.3 Other properties of clause type systems

Several questions about the nature of clause type systems have not received the at-
tention within semantic/pragmatic theory that they deserve, and before concluding
this chapter I would like to call attention to them. These questions have to do with
the status of clause types at the syntax/semantics interface, and for this reason fall
outside of the traditional ways of thinking about sentence moods originating in
philosophy. Nevertheless, they are likely quite important for understanding how
clause types and sentence moods really work, and we find intriguing ideas relevant
to them scattered throughout the linguistics literature.
We can begin with the issue of whether any of the three basic types is more
“basic” at the grammatical level. As we have observed, declaratives are often
assumed to have the most unmarked syntactic structure within a given language;
for example, when we say that English is an SVO language, we say this because
declaratives often have this word order. Syntactic theories will then often derive
interrogatives and imperatives by adding something to declarative syntax; for
example, in certain contemporary theories, a syntactic feature in C triggers the
distinct word order of interrogatives and imperatives.
If interrogatives and imperatives have something extra which differentiates them
from the unmarked declarative, we may ask what consequences this has for clause
typing and sentence mood. It might, for example, have consequences at the level of
force assignment. We see this type of approach in the analysis of Bierwisch (1980);
he proposes that declaratives lack a force marker, in contrast to interrogatives
(with marker Qu) and imperatives (Imp). Moreover, he allows declaratives greater
latitude in force assignment; declaratives, lacking a force marker, are associated
loosely with the force (or “pre-reflexive attitude”) D, while Qu and Imp are more
strictly associated with forces Q and I. Bierwisch’s ideas offer hope for explaining
why declaratives are more flexible in discourse function than interrogatives and,
especially, imperatives.
Although declaratives are assumed to be the simplest basic type in syntactic
terms, imperatives are often simpler in their morphology. In many languages, the
imperative verb is very simple, as simple as or even simpler than the infinitive.
For example, in Italian the canonical second person singular imperative consists
in the verb stem plus the vowel indicating its conjugation class (parl-a ‘speak’). The
fact that imperatives so commonly lack tense, mood, and agreement morphology
suggests that the internal structure of imperative clauses is (at least in some
languages) simpler than that of declaratives. Few theories of sentence mood make
anything of this relative simplicity, however.
This discussion of the respects in which both declaratives and imperatives might
be thought of as the “simplest” basic clause type raises the connected issue of
whether there are subgroups among the three basic types which are important

 Kaufmann’s () theory treats imperatives as containing an unexpressed modal akin to should,

and so the form of the verb could be explained as the result of its being selected by the modal. The
property analysis of Hausser (, ) and Portner () might explain it by building on the
observation that subsentential constituents like VP can denote properties.

i i

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

i i

230 sentence mood

for us to recognize as we seek to understand clause type systems. There is good

reason to assume that declaratives and interrogatives, with their full tense-mood-
agreement paradigms, form an important subgroup apart from imperatives. We
find in the literature many ideas which might connect this way of grouping clause
types to the broader theory of sentence mood. For example, building on the
important German tradition, scholars like Truckenbrodt (2006a) and Lohnstein
(2007) aim to give theories of clause typing which assign an important role to
indicative mood. From a quite different angle, we see the distinction between
declaratives and interrogatives being broken down by the likes of Gunlogson
(2001), Aloni and van Rooy (2002), and Ciardelli et al. (2013) (see Section 3.3.2).
It seems likely that interesting results would follow from trying to link these
two lines of thinking into a deeper view of the relation between declaratives and
interrogatives.
A final feature of clause type systems which is not yet well understood has to do
with the restrictions on embedding we find for certain clause-typing structures. In
particular, researchers have long been interested in the properties of imperatives
and verb-second clauses in Germanic languages:

(117) a. ∗ Mary ordered John [leave].

b. ∗ Maria will, [sie ist in diesem Fall in
Maria wants she is in that case in
Berlin]. (German, Truckenbrodt 2006a)
Berlin.
(118) a. Mary said [leave].
b. Maria sagt [Peter geht nach Hause].
Maria says Peter goes to home.

Although in the past it was sometimes stated that imperatives cannot be embedded,
we know now this is not the case. Similarly, linguists have long known that
it is simplistic to describe German as verb-second in root clauses but verb-final
in embedded clauses. Nevertheless, although imperatives and V2 clauses can be
embedded, the examples in (117) show that they do not do so freely.
One possible view of the data above is that imperatives and verb-second clauses
contain a force-indicating operator, and that this can only be embedded when the
higher predicate is able to combine with such a clause. How this plays out depends
on whether one thinks that the force-indicator actually encodes force (as on Krifka’s
view) or some other meaning associated with force (as on Truckenbrodt’s). It is
also possible that the embedding restrictions are due to semantic or pragmatic
properties not directly tied to force; for example, Pak et al. (2015) argue that

 For the idea that imperatives cannot be embedded, relevant references include Katz and Postal

(), Sadock and Zwicky (), Palmer (), Rivero and Terzi (), Han (), Platzack and
Rosengren (), Aikhenvald (), and Alcázar and Saltarelli (). For the contrary evidence,
we have Rögnvaldsson (), Sheppard and Golden (), Chen-Main (), Rus (), Platzack
(), Pak (), Crnič and Trinh (), Kaufmann and Poschmann (), Thomas (), and
Medeiros ().

i i

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

i i

looking ahead 231

the restrictions on embedded imperatives are due to the unique properties of

the imperative subject and restrictions on the marking of pragmatic features like
speech style and politeness, and not directly to imperatives’ sentential force.

3.5 Looking ahead

In this chapter, we have explored a variety of approaches to the analysis of sentence
moods and related concepts like sentential force and clause types. We have sought
to understand the major approaches to these topics and have looked in some detail
at specific analyses of individual sentence moods. Though I have not touched upon
a great deal of research on these topics, I hope that the presentation will provide
some helpful context as readers look deeper into the literature. Here in conclusion,
I would like to highlight a couple of points about the nature of this particular
subfield of semantics and pragmatics.
First, since clause types and sentence mood have been studied in a serious way
for such a long time and within so many different traditions, it is not an easy field
to get to know. We find research whose primary orientation is syntactic, semantic,
pragmatic, and philosophical. I have found it useful to organize the discussion
around two approaches to discourse meaning, speech act theory and the dynamic
approach, because they each offer a perspective on the nature of sentential force,
and I believe that understanding sentential force is at the heart of the task. This
leads to the second point: there are converging insights within proposals with very
different orientations. For example, in the previous subsection we identified several
analyses which suggest a close relation between declaratives and interrogatives, but
which do so from very different angles. Linguists should seek to build more upon
this type of convergence.
One area in which our knowledge has not developed as far as it should concerns
the crosslinguistic variation of sentence moods and clause type systems. While
we have good work on the typology of clause types, it does not fully incorporate
the vast amount of information about sentence moods embedded in descriptions
of particular languages. Descriptive and typological work often makes use of
concepts of discourse modality broader than sentence mood, such as reality status
and evidentiality. This separation between typological research and the formal
semantics/pragmatics tradition may have begun to narrow due to more accessible
presentations from the typological perspective (e.g. Aikhenvald 2010; Nuyts and
van der Auwera 2016) and analyses within semantic and pragmatic theory well-
informed by that perspective (e.g. Bittner 2011; Krawczyk 2012; Rett and Murray
2013; Murray 2016). Given the current trend in formal semantics towards greater
crosslinguistic research, it is likely that this trend will accelerate.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

4
Core mood, reality status,
and evidentiality

In Chapters 2 and 3, I have reviewed and sought to clarify and sometimes to

extend a range of semantic and pragmatic theories of verbal mood and sentence
mood, and it has emerged that within each of those two major topics there exists a
small number of cutting-edge approaches. I consider the following to be the most
important of them:

1. The theory of verbal mood is dominated by a general perspective which

treats the indicative/subjunctive contrast as marking features of the modal
parameters of interpretation for a clause. These modal parameters can be
modeled using either the method of shifting parameters or the method of
derived contexts, and they track a distinction between subjunctive contexts
with a comparative meaning and indicative contexts which imply truth in a
designated set of worlds (or a related concept like decidedness or contextual
commitment).
2. There are currently two main theoretical approaches to sentence mood.
One of them is the form of the dynamic approach which aims to explain
both the meanings of individual sentence moods and properties of clause
type systems in terms of a structured discourse context. According to this
view, we explain the function and range of the sentential forces because
they are the natural ways of updating the discourse context. The other main
approach is the dynamic force hypothesis, that modern branch of speech act
theory which incorporates ideas from the dynamic approach to represent
illocutionary force.

Verbal mood and sentence mood are by far the most well-studied categories of
mood, and for the most part, they are investigated by linguists as independent,
major subfields in semantics and pragmatics. In this chapter, I aim to offer some
perspective on these two subfields by doing two things: In Section 4.1, I will
investigate whether it is possible to synthesize theories of verbal mood and sentence
mood to produce a unified theory of core mood. Then, in Section 4.2, I will
attempt to situate verbal mood and sentence mood within the broader range
of mood phenomena, in particular looking at the concepts of reality status and
evidentiality. All of the material in this chapter will of necessity be preliminary
and even rudimentary, because there has been little discussion of these issues in

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, 26/9/2017, SPi

i i

prospects for a unified theory of core mood 233

semantics and pragmatics, but I hope this effort helps to lay the foundations for
future research on them.

4.1 Prospects for a unified theory of core mood

In this section I will show how it is possible to synthesize some of the most
important theories of verbal mood and sentence mood into a unified framework.
In particular, I am going to combine simplified versions of the comparison-based
theory of verbal mood and the dynamic approach to sentence mood based on the
structured discourse context. Roughly speaking, the goal is to produce a single
framework in which both the ideas of Section 2.2.2.5 and those of Section 3.4.2 can
be expressed. Doing this will show that there is some prospect for a unified theory
of core mood.
Before going on, I want to emphasize that the exercise undertaken in this section
is not intended to dismiss either past or future research on other approaches to
verbal mood and sentence mood. As has been noted, I am not convinced that the
comparison-based theory will be able to explain all of the relevant patterns (see the
discussion of Portner and Rubinstein’s arguments in Section 2.2.2.5). And there is
most definitely exciting work being done on sentence mood which does not fit into
the unified framework (for example, research on speech acts like Krifka 2014 and
the dynamic force hypothesis like Kaufmann 2012). Readers should understand
that my goal is to show what a unified theory would look like, and not to present
a final theory. Nevertheless, I also want to be clear that I believe that the proposed
synthesis explicates some deep connections between verbal mood and sentence
mood, and that it therefore points in the approximate direction of a correct theory
of how the two types of core mood are related.

The posw framework. Important theories of both verbal mood and sentence
mood are built on theoretical constructs which can be represented as a partially
ordered set of worlds. Within the theory of verbal mood, the idea that the subjunct-
ive marks comparison relies on an ordering relation, either a preference relation
< or an ordering source. Within the theory of sentence mood, imperatives are
either treated as deontic modals which make use of a prioritizing ordering source
or analyzed as updating a component of the structured discourse context which
represents an ordering of worlds or propositions. We can abstract away from some
of these differences and produce a single simplified system in which the essential
ideas of various theories can be stated. We begin by defining the posw:

(1) A partially ordered set of worlds posw is a pair c = csc , <c , where:
a. csc is a set of worlds, and
b. <c is a pre-order over csc .

 I argue in favor of a more fully developed theory in Portner ().

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

234 core mood, reality status, and evidentiality

Now we define the operations and operators for the posw. In doing so, we build
on the techniques of update logic developed by Veltman (1996) and Van der Torre
and Tan (1998). First, the update operations:

(2) Two linguistically natural updates of a posw c are:

a. For any posw c and atomic sentence φ,
c + φ = csc ∩[[ φ ]] c , <c 
b. For any posw c and atomic sentence φ,
c  φ = csc , <c ◦[[ φ ]] c 

We also have two modals:

(3) The basic necessity modals for a posw c are:

a. The informational modal: [[ cs φ ]] c = 1 iff csc ⊆[[ φ ]] c
b. The preference modal: [[ < φ ]] c = 1 iff mc ⊆[[ φ ]] c

In words: + updates the set of worlds in c,  updates the ordering in c, cs φ says
that cs entails φ, and < φ says that the best subset of cs entails φ. If we think
of csc , <c  as a modal base and ordering source, cs φ expresses simple necessity
and < φ expresses human necessity (Kratzer 1981; see Portner 2009, sect. 3.1 for
background on Kratzer’s theory).

The posw framework applied to verbal mood. The posw framework allows us
to capture important ideas in the theory of verbal mood in complement clauses. We
will focus here on the role of comparativity in verbal mood selection. We begin with
a definition of an agent’s cognitive model which can serve to help us express
the meanings of two representative verbs, ‘believe’ and ‘want’:

(4) An agent a’s cognitive model in situation s is a posw m(a, s) where:

a. csm(a,s) = the set of worlds compatible with a’s beliefs in s; and
b. <m(a,s) = the ordering of worlds which represents a’s desires in s.

Now we can give meanings for the representative non-comparative predicate

‘believe’ and the representative comparative predicate ‘want’:

(5) a. [[ A believes that φ ]] c = {w : ∃s[s < w and [[ cs φ ]] m(a,s) = 1]}

b. [[ A wants φ ]] c = {w : ∃s[s < w and [[ < φ ]] m(a,s) = 1]}

The meaning for ‘believe’ uses the informational modal cs , while the meaning
for ‘want’ uses the preference modal < . Both of these entries treat the attitude
predicate as a strong modal, but the one for ‘want’ uses the ordering component of
the cognitive model in a manner similar to an ordering source.

 Compare with the ‘preference structures’ of Condoravdi and Lauer (). Recall that ◦ refines the

ordering <c and mc identifies the best worlds in csc according to <c (Section ..).
 To maintain c + φ as a posw, <
c+φ should be restricted to csc+φ . This issue will be discussed later
in the section.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

prospects for a unified theory of core mood 235

The central idea of the comparison-based approach to verbal mood can now be
stated as follows:

1. Indicative principle: If a clause φ is operated on by the informational modal,

its form is indicative.
2. Subjunctive principle: If a clause φ is operated on by the preference modal,
its form is subjunctive.

An explicit theory would endorse one or both of these principles. If only one
principle is active, the other mood would be selected by default. Of course, these
principles can only be applied to ‘believe’ and ‘want’ so far. In order to apply them
to the full lexicon, we will need a much richer structure for the cognitive model.
See Sections 2.2.2.3 and 2.2.2.5 for discussion of the comparison-based approach.
This posw framework could be combined with many different ideas about the
syntax and semantics of sentence-embedding constructions. For example, we could
incorporate ideas from the approach to verbal mood which states that indicative is
selected by predicates which entail the truth of their argument in a designated set
of worlds. To do this, we have to give the mood morpheme in φ access to the posw
used to interpret it. This might be done by the method of shifting contexts or by
using one of the other techniques discussed in Section 2.2.
The framework is also compatible with more radical ideas about the logical
forms of sentence-embedding constructions. Using techniques from Portner (1997)
or Kratzer (2006, 2013), we might propose that the operators cs and < are not
introduced by the matrix predicate, but rather by a morpheme in the embedded
clause (the complementizer or mood morpheme itself). According to this idea, the
role of the verb is to specify the precise posw used by the modal operators. For
example, we might propose that the indicative introduces the modal operator cs ,
and when embedded under the matrix verb ‘believe’, this verb assigns it the posw
m(a, s). The verb ‘want’ would assign the same posw, but the operator introduced
by the subjunctive would be < .

The posw framework applied to sentence mood. Turning now to sentence

mood, it is not difficult to give an analysis of declaratives and imperatives in
the posw framework. Building on the ideas discussed in Section 3.3, we begin
with definitions of discourse commitments, commitment slate, and priority
slate:

(6) The discourse commitments of an individual a are a posw s where:

a. css = a’s commitment slate, a set of worlds compatible with all of a’s factual
commitments; and
b. <s = a’s priority slate, an ordering of worlds which represents a’s to-do
list.

It would also be possible to pursue this project in terms of a single common ground,
rather than individual commitment slates, but in order to do so we would need
to incorporate multiple priority slates (one for each participant) into the posw
framework.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

236 core mood, reality status, and evidentiality

A discourse context associates each participant with discourse commit-

ments:

(7) The discourse context is a function d from a set of individuals (the parti-
cipants) to discourse commitments.

This conceptualization of the discourse context allows definitions of two basic

sentential forces, assertion and directing:

(8) a. Assertion of a sentence S in discourse context d: apply the update

d(speaker) + S.
b. Directing a sentence S in discourse context d: apply the update
d(addressee)  S.

We might also treat assertion as updating the commitment slate of every parti-
cipant, not just the speaker’s. (The way we have used the commitment slate in (8)
is more in line with Gunlogson’s theory than Hamblin’s.)
Now we can identify the contributions of two of the basic sentence moods:

1. Declaratives: The sentential force of a declarative sentence is assertion.

2. Imperatives: The sentential force of an imperative sentence is directing.

Declaratives update the speaker’s commitment slate, while imperatives update the
addressee’s priority slate.
This sketch of the posw framework leaves open many basic issues in the
theory of sentence mood. Most fundamentally, it does not take a stand on what
grammatical properties constitute the declarative and imperative sentence moods.
It does not say precisely which sentences have a sentence mood—only root clauses,
root clauses and embedded clauses with root-like syntax, or all clauses? It also
does not establish how sentential force is assigned to sentence moods, and it
is compatible with any of the three approaches identified in Section 3.1.3 (the
operator-based, construction-based, and compositional theories). All of these
important issues must be settled in order to give a theory of sentence mood.
What should be focused on is that this sketch gives precise form to the intuition
that there are important connections between declarative sentence mood and
indicative verbal mood, on the one hand, and between imperative sentence mood
and subjunctive verbal mood, on the other. The connection is that declaratives/
indicatives have to do with the first component of a posw, the set of worlds csc ,
and that imperatives/subjunctives have to do with the second component, the
ordering <c .
There is an obvious gap in this discussion of sentence mood—we have not said
anything about interrogatives. The reason for this gap is that theories of verbal
mood rarely touch on how embedded interrogatives fit into their proposals, and
so a system designed to capture what is common to theories of verbal mood and
theories of sentence mood naturally cannot handle them. I do not know of a single
formal theory of mood selection in complement clauses which puts forth a specific

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

prospects for a unified theory of core mood 237

proposal about the meaning of the basic question-embedding verb ‘ask.’ Let me
give a taste of what is at issue: In Italian, the verb chiedere takes the indicative when
it means ‘ask,’ but the subjunctive when it means ‘wonder’ (example based on a
sentence in Eco, La bustina di Minerva):

(9) a. Gli avevo chiesto se ci sono corsi d’inglese.

him have-1sg asked if there be.3pl.indic course of English
‘I asked him whether there are English courses.’
b. Mi chiedo se ci siano corsi d’inglese.
me wonder-1sg if there be.3pl.subj courses of English
‘I wonder whether there are English courses.’

It will be impossible to fully understand the relation between verbal mood and
sentence mood until we have theories of verbal mood that can explain this
difference.
We observed in Section 3.3 that certain dynamic theories of questions minimize
or even eliminate the difference in sentential force between declaratives and
interrogatives. For example, on the dynamic semantics theory of questions which
treats the discourse context as a relation over possible worlds, assertion and asking
both involve intersection; specifically, asking a question involves the update C ∩ q
and asserting a proposition involves the update C ∩ (p × p). We could integrate
this view into the posw framework by replacing the first component of the posw
with a partition or other relation appropriate to the analysis of questions. We would
then have:

(10) A partitioned partially ordered set of worlds pposw is a pair c = pc , <c ,
where:
a. pc is a partition of some set of worlds csc , and
b. <c is a pre-order over csc .

Then both root declaratives and root interrogatives would target the first compon-
ent pc . This might provide the basis for explaining why they both have indicative
verbal mood (see Portner 2017 for an analysis along these lines). Alternatively, we
could add a third component, a question set, to the basic definition of a posw,
resulting in a theory similar to Roberts’ or Portner’s (Roberts 1996, 2012; Portner
2004). But it would not be wise for us to spend too much time considering these
options now. As observed above, if we wish to develop a unified framework for
analyzing verbal mood and sentence mood, what is most needed is a theory of
mood selection by interrogative-embedding predicates.

 Of course there are important theories of embedded interrogatives, some of which are mentioned

in Chapter , but they do not try to explain verbal mood selection.

 This model of context is similar to Starr’s () preference states, but he treats the force of

imperatives and interrogatives somewhat differently.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

238 core mood, reality status, and evidentiality

Embedded updates and embedded speech acts. This sketch of analyses of verbal
mood and sentence mood treats the two as similar only in two very abstract
respects: Both types of mood are analyzed in terms of the posw or pposw and make
a distinction between the mood which relates to csc (indicative and declarative)
and the mood which relates to <c (subjunctive and imperative). But the system
makes room for various hypotheses which imply a tighter relation between verbal
mood and sentence mood. Speaking very roughly, these hypotheses incorporate
sentential force into the semantics of embedding constructions and therefore, to
some extent, aim to derive the contrast between verbal moods from the distinct
forces of declaratives and imperatives.
We can envision analyses which incorporate either sentential force, in the form
of dynamic update potential, or full speech acts into the lexical semantics of attitude
verbs. We see the first of these options, incorporating sentential force, in the
analysis of verbal mood presented by Farkas (2003), as discussed in Section 2.2.2.5.
In the posw framework, we reformulate the meanings of ‘believe’ and ‘want’ as
follows:

(11) a. [[ A believes that φ ]] c = {w : ∃s[s < w and m(a, s) + φ = m(a, s)}]

b. [[ A wants φ ]] c = {w : ∃s[s < w and m(a, s)  φ = m(a, s)}]

These entries use the technique introduced by Heim (1992) for using an update
potential to reproduce the standard semantics for attitude verbs as strong modals.
(11a) is true if csm(a,s) includes the content of φ, because it says that performing a
+-update has no effect on m(a, s), and (11b) is true if <m(a,s) includes the preference
for φ. Assuming that all sentence-embedding constructions can be analyzed in a
similar way, we can consider alternative mood-selection principles:

1. Indicative principle: If a clause φ is the object of a +-update, its form is

indicative.
2. Subjunctive principle: If a clause φ is the object of a -update, its form is
subjunctive.

An analysis based on these principles has the advantage of applying to both root
and embedded clauses, and so can hope to explain the presence of indicative in
declaratives and the occasional use of subjunctive form for the imperative sentence
mood. These principles also imply that we should develop a theory of interrogative
sentence mood which treats it as using the +-update.
The ideas just outlined can be described as proposing that sentential force occurs
in embedded contexts. The posw framework also allows for the more radical
option of incorporating full speech acts into the semantics of embedded clauses.
Suppose that we think ‘say’ has a use which involves embedding a clause with the

 For the reasons discussed in Section ..., we also know that infinitives are closely related to

subjunctives, and so the correct generalization is probably that subjunctives, infinitives, and imperatives
fall under the same principle or closely related principles.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

prospects for a unified theory of core mood 239

illocutionary force of assertion. The posw introduced by ‘say’ is a reported dis-

course context r(s), and the embedded assertion meaning can be (12):

(12) [[ A says that φ ]] = {w : ∃s[s < w and s is a saying situation and the ap-
plication to r(s) of the sentential force appropriate to the sentence mood
of φ yields a transition between contexts which accurately represents A’s
communicative intention in s]}

According to (12), the sentence John said that Napoleon was insane is true in
a world w iff there was a situation s of John saying something in that world,
and his intention in s was to add the proposition that Napoleon is insane to
his commitment slate in the conversation of which s was a part. This type of
meaning might be appropriate to express in the posw framework the intuitions
of authors like Wechsler (1991), Meinunger (2004), and Krifka (2014) concerning
the illocutionary force of embedded verb-second declaratives.
The concept of embedded speech acts might or might not be an attractive
addition to the posw framework. But either way, the issue of embedded speech acts
should be clearly distinguished from the two less radical options discussed here.
Embedded indicatives could be similar to root declaratives for any of the following
reasons (and similarly for subjunctives’ relation to imperatives):

1. Like root declaratives, embedded indicatives are interpreted relative to the

domain of a posw, not the ordering.
2. Like root declaratives, embedded indicatives involve assertive sentential
force, modeled as the +-update.
3. Like root declaratives, embedded indicatives involve a speech act of assertion
relative to a reported context.

More than one of these possibilities could be involved in the overall explanation
of the relation between verbal mood and sentence mood. For example, it’s possible
that a normal indicative under ‘believe’ should be explained in terms of 1 or 2, while
an embedded clause with main-clause syntax, such as an embedded V2 declarative,
might be explained in terms of 3.

Elaborations and alternatives. The posw framework has been developed with
the goal of capturing in a single formal system the ideas of existing theories of
verbal mood and sentence mood, and it is easy to think of many different formal
systems in which we could express the same basic ideas. The reason I set forth
the posw is that it is just about the simplest in which the relevant ideas about
verbal mood and sentence mood can be expressed. Before we close this subsection,
I would like to mention several ways in which the framework could be elaborated.

1. Mutual commitments. The model defines a discourse context as an as-

signment of discourse commitments to each individual, but the analysis
of sentence mood could also be based on the goal of modeling mutual
commitments. The mutual commitments of a discourse would be a posw

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

240 core mood, reality status, and evidentiality

or pposw in which the first component represents the mutual assumptions

about the facts and issues to be answered (the common ground and question
under discussion set) and the second represents shared plans. This system
would be quite similar to that of Murray and Starr (2016). Alternatively,
it might be important to represent separately both mutual commitments
and individual commitments, as suggested by Portner (to appear, 2017). It
seems to me that, while basic declaratives and interrogatives can be analyzed
using mutual commitments, imperatives are more naturally understood as
relating to individual commitments (as encoded in a priority slate, to-do list,
or similar).
2. Multiple orderings. An obvious and probably necessary elaboration of this
framework would be to make room for multiple orderings of csc . We might
think of a “multiply ordered set of worlds,” a mosw = csc , <1c , . . . , <nc .
Clearly it is a dramatic oversimplification to propose that an individual’s
cognitive model or individual commitments can be represented using a single
ordering. But beyond the issue of oversimplification, there are other reasons
to adopt multiple orderings. Doing so would allow us to incorporate the
proposal that modals can be sensitive to multiple ordering sources (von Fintel
and Iatridou 2008), and the idea that, in some cases, ordering sources are
“promoted” to modal base status (Rubinstein 2012). It would also allow us
to adopt the idea that the many discourse functions of imperatives differ in
that they affect distinct “sub-lists” of the to-do list (Portner 2007). We might
well also want to represent expectation in discourse with an ordering relation,
following Veltman (1986).
3. Commitment sets and premise sets. In place of the set of worlds csc and
the ordering over worlds <c , the system might utilize sets of propositions
for these functions. In applying the framework to verbal mood, the set of
propositions which does the job of csc would be a modal base; the set
of propositions which does the job of <c would be the ordering source.
In applying the framework to sentence mood, the former would be an
individual’s commitment set, and the latter would be her to-do list. There are
likely to be advantages in using sets of propositions for these two functions
when it comes to modeling speech acts of retraction and offering permission
(see, for example, Gazdar 1981 and Kamp 1973) and inconsistent priorities.
4. The context of justification. In Section 3.3.3, we saw that Van der Torre
and Tan (1998) make an interesting distinction between the context of de-
liberation and the context of justification in the dynamic logic of deontic
statements. The context of deliberation considers the deontic ordering of
the worlds in csc (the worlds compatible with current information), while
the context of justification considers a deontic ordering over a wider set of
worlds W. In our application of the posw framework to sentence mood, we
were not explicit about whether <c represents the ordering of the context of
deliberation or the ordering of the context of justification. In our definition
of the assertive update in (2a), no change is made to the ordering <c , and this

 In the simplest version, the to-do list would be treated as a set of propositions, rather than as a set

of properties as in Portner’s () proposal.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

prospects for a unified theory of core mood 241

means that it can order worlds not in c + φ. Specifically, if we had w <c v, and
then an assertive update with φ removes w from csc+φ , we still have w <c+φ v.
At this point, c + φ will no longer technically be a posw.
There are two ways we might adjust the definitions to bring them into
alignment. Either we could revise the definition of c+φ so that <c is restricted
to worlds in [[ φ ]] c (<c+φ ⊆([[ φ ]] c ×[[ φ ]] c )). Or we could maintain in our
system both a context of deliberation and a context of justification. Such a
move might have benefits both in the analysis of sentence mood (of optatives
and counterfactual imperatives, for example) and in the analysis of verbal
mood (of the complements of counterfactual attitudes like ‘wish’), since we
sometimes do need to pay attention to what rules, desires, and preferences
have to say about worlds which are ruled out by our information.
5. Ordering and de se. The posw framework is designed to exemplify how we
might capture certain commonalities between theories of verbal mood and
theories of sentence mood. These commonalities are based on the use of
modal parameters to represent both cognitive states (as cognitive models)
and discourse information (as individual commitment slates) in terms of
orderings over possible worlds. There is another, seemingly unrelated, point
of commonality between some of the theories of verbal mood and some of
the theories of sentence mood which we discussed in Chapters 2–3. In both
areas, we from time to time encounter the idea that one subtype of mood
implies special interpretation of its subject, time, or world argument: either a
de se reading or an “unanchored” interpretation which might be understood
as a variety of de se. (At least, de se was my best attempt to understand in
precise terms what the absence of anchoring amounts to.) We specifically saw
ideas of these kinds applied to subjunctives, infinitives, and imperatives. We
should seek ways to develop the idea that this group of clauses has something
of this kind in common. A simple way of thinking about this similarity could
be turned into the hypothesis that subjunctives, infinitives, and imperatives
have a structure which implies that one or more of these arguments is de se.
If it turns out that subjunctives, infinitives, and imperatives all have in com-
mon that one or more of their clausal arguments are obligatorily interpreted
as de se, we will be left with an important but previously-unrecognized puzzle:
• The comparison-de se puzzle: Why do those moods which are associated
with a comparison-based semantics also have arguments which obligatorily
receive a de se interpretation?
We do not yet have a very good grasp of what this link between comparison
and de se amounts to. Consider the case of the de se interpretation of a
complement clause’s subject. A verb like ‘want’ can take either an infinitive or
subjunctive complement, but if the embedded subject is bound by the matrix
subject and de se, only the infinitive is possible. Matters are different with
an indicative-selecting verb like ‘believe’. Both the bound/de se and non-de
se interpretations are available with the same mood form and pronominal
subject. When it comes to tense, we find that the temporal arguments of
both infinitive and subjunctive complements normally must be de se, while
an indicative complement’s temporal argument is optionally de se when it
agrees with the matrix tense (i.e. when it is a sequence-of-tense form). Given

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

242 core mood, reality status, and evidentiality

this complexity, the way I have stated the comparison-de se puzzle may be
incomplete, yet there is clearly something to the point that verbs like ‘want’,
which have a modal semantics of comparison, also select de se-triggering
infinitival and subjunctive complements, while verbs like ‘know’, which do
not make a modal comparison, select the indicative.
6. Probabilities and utilities. As we have discussed in Sections 2.1 and 3.3.1,
the framework of modal semantics based on premise sets (Kratzer) and the
model of discourse meaning based on the common ground (Stalnaker) have
been challenged recently by systems which use mathematical probabilities
and utilities. We should investigate whether such systems can incorporate and
improve upon the most promising hypotheses about the relationship between
verbal mood and sentence mood. Within the posw framework, there is a
single major parameter which distinguishes subjunctives/imperatives from
indicatives/declaratives: the presence or relevance of an ordering relation.
Within the probability-based framework, we have two parameters which
seem equally important: the contrast between contexts which involve probab-
ilities alone (the complements of ‘probable’ and ‘believe’) versus those which
involve both probabilities and utilities (‘good’), and the contrast between
contexts which involve a comparison between probabilities or utilities (‘prob-
able’ and ‘good’) versus those which do not (‘believe’). The success of the
comparison-based approach to verbal mood suggests that a probability-based
theory of verbal mood would see comparison of probabilities and utilities as a
crucial factor. When it comes to sentence mood, however, we could entertain
the idea that what sets imperatives apart from declaratives and interrogatives
is that they relate to utilities. This point of view would be in contrast to the
posw framework, according to which the crucial feature of imperatives is that
they affect a preference or other ordering relation.
It seems likely that one could develop a probability-based theory which
incorporates the ideas of the posw framework. A probability space is simply
a pair c, P consisting of a set of worlds and a probability measure. Since
P gives an ordering on worlds (and on propositions), this already has the
power of a posw. If we add a utility function U, we have a system which
functions very similarly to a multiply ordered set of worlds. The real issue
will be whether this probability-based theory can explain the properties of
verbal mood and sentence mood better than those existing analyses which
provide the motivation for the posw framework. In order to make progress
on this issue, a great deal of work would have to be done, since, as far as
I know, the literature has not yet produced any probability-based analyses
of non-indicative verbal moods or non-declarative sentence moods.

4.2 Reality status and evidentiality

We have seen that verbal mood and sentence mood, the two obvious subcategories
of mood, might be properly seen as subtypes of the same linguistic category. In the
terminology of this book, both are systems of mood, because they both indicate the

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

reality status and evidentiality 243

role a clause plays in the computation of modal meaning. It even seems likely that
they mark similar semantic features of the modal meanings they relate to. Verbal
mood and sentence mood each can be seen as marking the presence or absence of
comparison in that modal meaning, and possibly also some semantic component
related to de se interpretation. If these formal similarities are indeed present, the
reason that they exist is presumably rooted in the fact that both subsentential
and discourse modality build on the same basic architecture of modal semantics.
Because of the fact that they can be understood (in at least their most central
features) using the fundamental concepts of modal semantics, we can describe
verbal mood and sentence mood as “core mood.”
In Chapter 1, we noted that a number of other grammatical categories might
plausibly be recognized either as core mood or as more peripheral types of mood.
Two such categories have been the subjects of important research which would
allow one to consider their relation to the category of mood more generally:
reality status and evidentiality. In this section, we will discuss whether reality status
and evidentiality fall under the definition of mood, and whether they are similar
enough to verbal mood and sentence mood to be considered examples of core
mood. In neither instance will we reach a firm conclusion, but I think that the issues
which get raised in the course of thinking about this topic will prove important in
future research on mood and modality.

4.2.1 Reality status

Within the descriptive/typological tradition, it is common to see the terms “realis”
and “irrealis” used as the labels of morphological forms or paradigms, or in
the explication of forms with more specific meanings. The prevalence of this
terminology, and the ways of thinking about the relation between grammatical
forms and the conceptual domain of modality which go along with it, have led
to an important debate concerning whether such terms should really form part of
our theory of natural language meaning. If one believes that realis and irrealis are
important concepts for linguistics, it will be useful to have a term which labels
the conceptual domain within which they make a distinction. Elliott (2000), a
supporter of the value of these concepts to linguistics, has coined the term reality
status for this category. In this section, we will examine the debate about whether
reality status should be seen as a valid category for semantic/pragmatic analysis,
on a par with such well-motivated categories as tense and aspect, and will consider
the relation between some of the grammatical forms which have been described in
terms of the opposition realis/irrealis and the category of mood as it is understood
in this book.
In order to give the discussion an explicit empirical grounding, I will begin
by giving an outline of some important facts from the Papuan language Amele,
as presented by Roberts (1991). Amele has been brought up repeatedly in the
recent literature on reality status, both by scholars who believe in the value of the
realis/irrealis distinction (such as Elliott 2000) and by those who argue against it
(e.g. de Haan 2012). Other work which can serve as pointers into the literature
on this topic includes Givón (1994), Chafe (1995), Mithun (1995), Van Valin and

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

244 core mood, reality status, and evidentiality

LaPolla (1997), Kinkade (1998), McGregor and Wagner (2006), the papers collected
in Mauri and Sansò (2012), and Matić and Nikolaeva (2014).
In order to understand the marking of reality status in Amele, we must con-
sider three grammatical patterns: the clause chaining construction, subordinate
adverbial clauses, and sentence-level particles. We begin with the clause chaining
construction, a type of sentence appearing as a series of clauses strung together
with only the final one fully marked for tense and mood. Preceding clauses are
dependent on the final one, and have subject agreement marking which indicates
whether they have the same or a different subject from the final clause (conjunct-
disjunct or switch reference marking) and whether their temporal reference is sim-
ultaneous or sequenced with that of the final clause. For example, in example (13),
the initial clause ‘the pig ran out’ is dependent on the verb which follows, ‘hit’
(Roberts 1991, p. 371, (2a)):

(13) Ho bu-busal-en age qo-in.

pig sim-run.out-3sg.ds.r 3pl hit-3pl.rem.p
‘They killed the pig as it ran out.’

The agreement marker glossed as ds.r in (13) is a different subject, simultan-

eous agreement form, indicating that the pig is distinct from the ones who
killed the pig and that the pig’s running out was simultaneous with their
killing it.
There are two sub-paradigms of different subject, simultaneous agreement
forms, with the choice between them determined by the tense, mood, and modal
meaning marked on the final clause. Roberts associates this distinction with reality
status, and so labels them the “realis” and “irrealis” forms. In (13), the realis version
of third singular different subject agreement (-en, ‘ds.r’) is determined by the
fact that the final clause is marked for remote past tense. Other forms of the
final verb which trigger realis agreement are the present and the various other
past tenses. Irrealis agreement is chosen when the final clause is marked as future
tense, imperative, prohibitive, counterfactual, and others (Roberts 1991, pp. 371–2).
Example (14) shows an irrealis form (Roberts 1991, (3a)):

(14) Ho bu-busal-eb age qo-qag-an.

pig sim-run.out-3sg.ds.ir 3pl hit-3pl-fut
‘They will kill the pig as it runs out.’

Turning now to adverbial clauses, Roberts shows that only irrealis agreement
forms appear in adverbial subordinate clauses which convey such meanings as
intention, desire, and purpose; realis agreement never occurs in clauses of this type.
An example is (15) (his (6a)):

(15) Ho bu-busal-eb age qo-qag-a bili tawe-ig-a.

pig sim-run.out-3sg.ds.ir 3pl hit-3pl-rel.fut be stand-3pl-tod.p
‘They stood about to kill the pig as it runs out.’

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

reality status and evidentiality 245

The initial clause here shows irrealis marking, dependent on the overall relative
future tense of the adverbial clause. In the same position as relative future tense in
this construction, Roberts shows examples with imperative and infinitival marking
as well. It appears that reality status marking in adverbial clauses functions in very
much the same way as in clause chaining, but there is no possibility of a tense/mood
form which would trigger realis agreement.
Finally, Roberts shows that sentence-level particles which can be described as
indicating illocutionary force or sentence mood do not affect the marking of realis
or irrealis in chained and adverbial clauses. For example, the polar question marker
fo, the dubitative question marker fa, and negation might be expected to create an
irrealis context, but only the realis form is possible with a realis-triggering tense,
as in (16) (from Roberts 1991, (18)):

(16) Ho bu-busal-en/∗ eb age qo-gi-na fo.

pig sim-run.out-3sg.ds.r/ir 3pl hit-3pl-pres dub
‘Maybe they are killing the pig as it runs out?’

Roberts goes on to describe a number of other Papuan languages in similar terms.

For example, in Anjam reality status is marked (though in different ways) in both
different subject and same subject forms, and in both simultaneous and sequential
temporal meaning. In Menya, according to Roberts (1991) and Whitehead (1986),
reality status is marked on final verbs; for example, an imperative final verb uses
the irrealis subject agreement form.
The range of clauses in Amele which employ the realis and irrealis agreement
forms match up to an impressive extent with the intuitive concepts of realis and
irrealis. Should the pattern found in Amele therefore be taken as evidence in
favor of a linguistically important category of reality status? Elliott (2000) believes
that it should. She accepts Roberts’ description of the meaning of the two sets of
agreement forms, and then points out the important point that, from this perspect-
ive, the marking of reality status is restricted to a specific set of morphosyntactic
environments, namely in combination with the marking of different subjects and
simultaneous temporal reference. De Haan (2012), in contrast, argues that the
failure to mark irrealis in dubitative questions and other sentences marked as non-
assertive by sentence-level particles undermines the description in terms of reality
status. Moreover, Elliott’s description of Amele as a language in which reality status
is marked in a particular set of morphosyntactic environments does not seem quite
right; the “context” can be a clause which marks switch reference and simultaneity,
but if the final verb has present tense, irrealis due to a final particle will not be
marked. Rather, the correct description of the facts seems to be that the morphemes
ds.r and ds.ir are dependent on the tense/mood form of the final verb. They are
dependent on a feature of meaning of that form which is naturally described in
terms of reality status, but this does not mean that the dependent agreement forms
mark the reality status of the clause they are in per se, since if they did, we would
also expect ds.ir in (16).
Following Bybee (1998), de Haan develops a broader argument against the value
of the reality status categories realis/irrealis. Perhaps most importantly, he argues

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

246 core mood, reality status, and evidentiality

that the range of contexts which are argued to trigger realis or irrealis forms is
too diverse across languages to treat any one opposition as the “true” realis/irrealis
opposition. This situation lends support to the idea that realis and irrealis are
prototype categories, as argued by Givón (1994), if they are useful categories at all.
However, as pointed out by de Haan, even such prototypically irrealis constructions
as counterfactuals and imperatives are classified as realis in some languages. It is
difficult to see how to understand the idea that irrealis is a prototype category if
some languages mark less prototypical constructions with the irrealis form but
not more prototypical ones. We might, rather, need to understand the meta-
concept of irrealis as having a prototype structure, with languages representing in
their grammars more or less prototypical versions of the irrealis concept (itself, a
prototype concept).
There is an important issue which has not been raised adequately within this
literature on reality status, namely why the category is marked in the specific range
of contexts that it is. Concerning Amele, why is realis/irrealis marking limited to
different subject, simultaneous clauses? Here it is worthwhile to draw a link to
some of the observations about infinitives and subjunctives discussed in Chapter 2.
First, recall that subjunctives often show an obviation effect, meaning that a
subjunctive complement cannot be used if its subject has the same reference as
that of the superordinate clause; this fact is probably best analyzed in terms of a
preference for subject de se interpretation, realized as a control infinitive, when
it is compatible with the intended meaning (see Section 2.2.2.4). And second, as
pointed out in Section 2.3.1.1, the tense of a complement subjunctive is temporally
dependent on that of the matrix clause, and this temporal dependence might
also be analyzed as a form of temporal de se. Thus, it seems that Amele irrealis
agreement is associated with some of the same semantic properties as the control
infinitive and subjunctive. This point suggests that reality status in Amele might
be analyzed using some of the same ideas as have been used in semantic theories of
verbal mood. A successful analysis along these lines would be quite exciting, since
it is without a doubt important that reality status in Amele is realized in connection
with simultaneity and different-subject marking. Stepping back, work on different
languages has proposed that a wide variety of morphosyntactic categories can
realize reality status, including voice and nominal morphology, in addition to
verbal morphology and agreement (see de Haan 2012, sect. 3). Assuming that all of
these forms really involve the semantics of mood, in the broad sense, the theory of
mood must seek to explain why it is realized in all of these different ways.
This discussion of reality status has of necessity been incomplete and sketchy, but
some important conclusions suggest themselves. It seems likely that the perspective
of Bybee and de Haan is correct: languages do not directly mark clauses for their
reality status. That is, realis marking does not literally mean “this clause is true in
reality,” and nor does irrealis marking mean “this clause is not (known to be) true
in reality.” Instead, reality status should be understood, in some cases at least, as

 Zu’s () work on conjunct marking in Newari and its relation to the subjunctive is relevant here.

Though it does not concern a realis/irrealis system, it highlights the important relations among person
marking, de se meaning, and obviation.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

reality status and evidentiality 247

marking a clause as dependent on another element or construction which itself

might imply something about the truth or falsity of the clause in reality. In such
cases, reality status should be seen as related to verbal mood, and might even be
correctly described as a type of verbal mood, in the general sense in which we have
defined it. Given the wide variety of ways in which the terms ‘realis’ and ‘irrealis’
have been used, it is difficult to predict whether the bulk of phenomena which have
been described in terms of reality status should be treated as core mood—that is,
as examples of mood which can be analyzed using the basic conceptual framework
used to analyze verbal mood and sentence mood.

4.2.2 Evidentiality
Evidentiality is the grammatical encoding of information source. For example, the
Cuzco Quechua example (17) may be used to assert that the speaker’s brother is
working in Italy, and in addition, as a result of the reportative evidential -si, it
conveys that the speaker was told this (from Faller 2002, p. 139):

(17) Tura-y-qa Italia-pi-s llank’a-sha-n kay semana-pi.

brother-1-top Italy-loc-rep work-prog-3 this week-loc
p= ‘My brother is working in Italy this week.’
ev: speaker was told that p.

Although linguists sometimes use the term “evidentiality” to describe any expres-
sion of evidence source, semanticists typically restrict it to closed-class morpho-
syntactic categories. Cuzco Quechua has a typical evidential system, with three
main forms referred to by Faller as the ‘best possible grounds’ (or ‘direct’), ‘con-
jectural,’ and ‘reportative’ evidentials. There is significant diversity in evidential
systems across languages, both with respect to their morphosyntactic realization
and the meanings they express; see Aikhenvald (2006) and de Haan (2012) for
recent descriptive overviews and Krawczyk (2012) for a review of the main semantic
properties of evidentials and an interesting discussion of some problems with
describing evidential meaning solely in terms of the source of evidence a speaker
has for what she says.
As discussed by Portner (2009, sect. 5.3.2), evidentiality has been analyzed as
being closely related to several other linguistic categories, in particular epistemic
modality (for example, McCready and Ogata 2007 and Matthewson et al. 2007),
tense/aspect (Chung 2007), and sentence mood. What is relevant to us here is the
question of whether evidentiality represents a variety of mood, and so we should
look at theories which treat it as related to sentence mood. We find such theories
based on each of the two main theoretical frameworks for analyzing sentence mood
discussed in Chapter 3, namely speech act theory and the dynamic approach. In
this section, we will briefly discuss Faller’s (2002, 2011) theory of evidentiality,
which treats members of this category as affecting the direct, literal illocutionary
force associated with a sentence, as well as Murray’s (2011, 2014, 2016) theory,
which treats them as affecting the update potential of a sentence within a dynamic
semantics framework.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

248 core mood, reality status, and evidentiality

Faller’s speech act theory. In her dissertation, Faller (2002) developed an in-
fluential theory of evidentials based on speech act theory. Focusing on Cuzco
Quechua, she proposes that evidentials modify various features of the illocutionary
force associated with a sentence. She uses a system similar to that of Vanderveken
(1990) (see Sections 1.4.2 and 3.2.2), so that a basic illocutionary act of assertion
with propositional content p can be described in terms of three components: its
illocutionary force, sincerity conditions, and strength (Faller 2002, p.25):

(18) Para-sha-n
rain-prog-3
p= ‘It is raining.’
ill = asserts (p)
sinc = {Bel(s, p)}
strength = 0

Faller proposes that evidentials modify such aspects of the speech act as its sincerity
conditions and strength. The direct (best possible grounds) evidential -mi, for
example, adjusts the speech act as follows:

(19) Para-sha-n-mi
rain-prog-3-bpg
p= ‘It is raining.’
ill =asserts (p)
sinc = {Bel(s, p), Bpg(s, p)}
strength = +1

While the direct evidential can be neatly analyzed as an illocutionary modifier,

Faller finds that the two other main evidentials in Cuzco Quechua raise problems.
According to her analysis, the conjectural -chá affects both the propositional
content and the illocutionary act:

(20) Para-sha-n-chá
rain-prog-3-conj
p= ‘♦It is raining.’
ill = asserts (p)
sinc = {Bel(s, p), Rea(s, Bel(s, p))}
strength = −1

Faller sees -chá as a combination of an epistemic modal, accounting for the ‘♦’ in
(20), and an evidential.
The reportative evidential proves the most difficult for Faller. The problem is that
a reportative sentence like (17) is compatible with the speaker being completely
non-committal as to the truth of its propositional content. For this reason, it does

 The data and descriptions of speech acts in ()–() are modified from Faller (, pp. –).

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

reality status and evidentiality 249

not fall under any of the types of illocutionary point described by Searle (1975b)
and Vanderveken (1990), and Faller proposes a new type of illocutionary point,
presentation.

(21) Para-sha-n-si
rain-prog-3-rep
p= ‘It is raining.’
ill = presents (p)
sinc = {∃s2 [Assert(s2 , p) ∧ s2 ∈ {speaker, addressee}]}

In later work, Faller continues to revise her analysis of the reportative (Faller
2006a,c) and she gives more precise analyses of the sincerity conditions of all three
evidentials within a possible worlds semantics for modality (Faller 2007).
Faller’s theory very clearly places the evidentials of Cuzco Quechua within the
semantic domain of mood as we have defined it here. Each evidential morpheme
indicates how a proposition is used in the expression of modal meaning at the
discourse level. Her theory presupposes that speech act theory is the correct
framework for analyzing sentence mood, and if we grant this presupposition,
it would be accurate to say that evidentiality of the kind exemplified by Cuzco
Quechua forms part of the sentence mood system. This approach to evidentiality
probably implies that evidentiality and sentence mood are not as closely related to
verbal mood as suggested in Section 4.1, since there are no productive theories of
verbal mood based on classical speech act theory.

Murray’s dynamic theory. Murray (2014, 2016) argues that evidentials should
be analyzed as closely related to markers of sentence mood within a dynamic
model of discourse meaning. She focuses on Cheyenne, a language in which both
sentence mood markers and evidentials fall into a morphological paradigm known
as mode, arguing that they are not merely related in their morphosyntax, but also
in their interpretation. Next we will briefly examine how she proposes to give them
a unified semantic analysis.
Murray’s central ideal about evidentiality is that evidentials update the discourse
context in a way similar to, but more “direct” than, the updates associated with
sentence mood. In the case of a direct evidential declarative like the Cheyenne (22),
the declarative mood leads to an update which restricts the context set to worlds
in which Sandy won, while the direct evidential restricts the context set to worlds
in which the speaker has direct evidence for this proposition (Murray 2014, (1a)).

(22) É-hó’tahéva-∅ Sandy.

3-win-dir Sandy
‘Sandy won (I witnessed).’

 To understand the range of forms covered by the category of mode in American linguistics, readers

may review the way it is used by such authors as Boas (), Newman (), Bloomfield (), Parks
(), Axelrod (), and Mithun ().

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

250 core mood, reality status, and evidentiality

The analysis builds on an update semantics closely related to the dynamic semantics
for questions and inquisitive semantics theories outlined in Section 3.3.2, in that
the ultimate function of a declarative is to reduce the context set and that of an
interrogative is to partition it. But, in comparison to other dynamic semantics
models, the update proceeds in several stages, allowing her to differentiate the kind
of update produced by an evidential from the kind produced by a sentence mood.
The key way in which the illocutionary update of sentence mood differs from
the evidential update is that the latter but not the former is “not-at-issue,” and
is introduced into the discourse representation without opening up the issue of
whether it should be accepted. For example, the assertive update associated with
the declarative (22) initially introduces a discourse referent for the proposition that
Sandy won; this step can be thought of as analogous to placing the proposition on
the table and projecting it as a future commitment within the model of Farkas and
Bruce (2010). Only at a later step in the process of updating the discourse context
is the context set restricted to worlds in which Sandy won. In contrast, because
it is not at-issue, the proposition associated with the direct evidential (that the
speaker has direct evidence for the proposition that Sandy won) does not introduce
a discourse referent; instead, the evidential updates the common ground directly,
with no stage at which it is on the table.
Murray’s theory differentiates the illocutionary update from the evidential up-
date in two ways: in the presence or absence of a discourse referent, and in whether
the update proceeds in stages or is direct and immediate. It seems to me that
the former difference is closer to the heart of the matter. As she remarks, the
presence of the discourse referent in the illocutionary update (in the example,
representing the proposition that Sandy won) makes this proposition available for
“denials, objections, and challenges,” while the evidential update does not produce
a discourse referent to be denied or challenged (Murray 2014, p. 230). The difference
between direct and staged update, on the other hand, is less clear to me. Given
our understanding of the context set as representing the mutual presuppositions
within a conversation, one participant can hardly impose an update without
his interlocutor having any say in the matter. The evidential update is certainly
somehow special; it is backgrounded in a way we would like to understand, and
it concerns facts over which the speaker has more epistemic authority (i.e. the
speaker usually knows what he has seen better than the addressee), but these
differences do not seem to me fully captured by the notion of a direct update.
Murray’s theory implies that evidentiality is a variety of mood as we have defined
it in this book, and moreover that evidentials are correctly classified as sentence
moods. Because she employs a dynamic semantics model closely related to the
posw framework, it is possible that her ideas could be incorporated into a theory
of core mood encompassing verbal mood, sentence mood, and evidentiality.

 Other scholars have recently developed similar systems in which update proceeds in stages which

vary in “directness,” for example AnderBois et al. (). We might also draw a connection between
Murray’s direct update and the update of individual commitments which Coppock and Wechsler (to
appear) consider for ego-marked clauses in Newari.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

reality status and evidentiality 251

4.2.3 Final remarks

The literature on mood, in all of its senses, and the literature on the related concepts
of reality status, evidentiality, mode, and modality, are rich and diverse. Because an
analysis of some particular data concerning a language’s mood system (be it a verbal
mood, sentence mood, or other mood-like system) allows a researcher to express
his or her more fundamental commitments about the nature of language, these
analyses are often very difficult to compare, and so to some extent the literature
on these topics has been fragmented into sub-literatures written by individuals
with similar fundamental commitments. In this book, my primary goal has been
to step back and identify many of the important ideas which underlie theories of
verbal mood, sentence mood, and other phenomena which may represent types
of mood, seeking to clarify them, identify their strengths and weaknesses, find
common themes, and ultimately point the way towards new work which builds
on the best ideas of all.
Work on mood can play an important role in helping us gain a better under-
standing of broader issues in linguistics. There appear to be commonalities within
the analyses of verbal mood and sentence mood (and perhaps evidentiality and
reality status as well), and these may point to deep facts about discourse meaning,
grammar, and the relation between them. Because of the close relationship between
mood and modality, a theory of modal semantics cannot be correct unless it
is compatible with reasonable analyses of mood phenomena, and vice versa.
Researchers should make the most of these connections as well.

i i

i i
 i i
OUP CORRECTED PROOF – FINAL, 26/9/2017, SPi

i i

i i

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

i i

References
Abusch, D. (1997). Sequence of tense and temporal de re. Linguistics and Philosophy,
20:1–50.
Abusch, D. (2004). On the temporal composition of infinitives. In Gueron, J. and
Lecarme, J., eds, The Syntax of Time, pages 29–53. MIT Press, Cambridge, MA.
Aelbrecht, L., Haegeman, L., and Nye, R., eds (2012). Main Clause Phenomena: New
Horizons. John Benjamins, Amsterdam.
Aikhenvald, A. (2006). Evidentiality in grammar. In Brown, K., ed., Encyclopedia
of Language and Linguistics, volume 4, pages 320–5. Elsevier, Oxford.
Aikhenvald, A. (2010). Imperatives and Commands. Oxford University Press,
Oxford and New York.
Alcázar, A. and Saltarelli, M. (2012). Why imperative sentences cannot be embed-
ded. Cahiers Chronos, 25:1–23.
Allan, K. (2006). Clause-type, primary illocution, and mood-like operators in
English. Language Sciences, 28:1–50.
Aloni, M. (2007). Free choice, modals, and imperatives. Natural Language
Semantics, 15(1):65–94.
Aloni, M. and Ciardelli, I. (2013). A logical account of free choice imperatives.
In Aloni, M., Franke, M., and Roelofsen, F., eds, The Dynamic, Inquisitive, and
Visionary Life of φ, ?φ, and φ: A Festschrift for Jeroen Groenendijk, Martin
Stokhof, and Frank Veltman, pages 1–17. University of Amsterdam.
Aloni, M. and van Rooij, R. (2004). Free choice items and alternatives. In Bouma,
G., Krämer, I., and Zwarts, J., eds, Proceedings of KNAW Academy Colloquium:
Cognitive Foundations of Interpretation, pages 5–25. Edita KNAW, Amsterdam.
Aloni, M. and van Rooy, R. (2002). The dynamics of questions and focus. In
Jackson, B., ed., Proceedings of Semantics and Linguistic Theory 12, pages 20–39.
CLC Publications, Ithaca, NY.
Aloni, M., Butler, A., and Dekker, P., eds (2007). Questions in Dynamic Semantics.
Elsevier, Oxford.
Alonso-Ovalle, L. (2006). Disjunction in Alternative Semantics. PhD thesis,
University of Massachusetts, Amherst.
Altmann, H. (1987). Zur Problematik der Konstitution von Satzmodi als Form-
typen. In Meibauer, J., ed., Satzmodus zwischen Grammatik und Pragmatik,
pages 22–58. Niemeyer, Tübingen.
Ambar, M. (2016). On finiteness and the left periphery: Focusing on subjunctive.
In Błaszczak, J., Giannakidou, A., Klimek-Jankowska, D., and Migdalski, K., eds,
Mood, Aspect, Modality Revisited: New Answers to Old Questions, pages 125–76.
University of Chicago Press, Chicago, IL.
Anand, P. and Brasoveanu, A. (2010). Modal concord as modal modification.
In Prinzhorn, M., Schmitt, V., and Zobel, S., eds, Proceedings of Sinn und
Bedeutung 14, pages 19–36. Vienna.
Anand, P. and Hacquard, V. (2013). Epistemics and attitudes. Semantics and
Pragmatics, 6(8):1–59.

i i

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

i i

254 references

Anand, P. and Nevins, A. (2004). Shifty operators in changing contexts.

In Watanabe, K. and Young, R. B., eds, Proceedings of Semantics and Linguistic
Theory 14, pages 20–37. CLC Publications, Ithaca, NY.
AnderBois, S., Brasoveanu, A., and Henderson, R. (2013). At-issue proposals and
appositive impositions in discourse. Journal of Semantics, 32(1):93–138.
Anderson, A. R. (1951). A note on subjunctive and counterfactual conditionals.
Analysis, 11:35–8.
Åqvist, L. (1967). Good Samaritans, contrary-to-duty imperatives, and epistemic
obligations. Noûs, 1(4):361–79.
Asher, N. (1987). A typology for attitude verbs and their anaphoric properties.
Linguistics and Philosophy, 10(2):125–98.
Asher, N. and Bonevac, D. (2005). Free choice permission is strong permission.
Synthese, 145(3):303–23.
Asher, N. and Lascarides, A. (1998). Questions in dialogue. Linguistics and Philo-
sophy, 23(3):237–309.
Asher, N. and Lascarides, A. (2001). Indirect speech acts. Synthese, 128(1–2):
183–228.
Austin, J. (1962). How to do Things with Words. Oxford University Press, New York.
Avrutin, S. and Babyonyshev, M. (1997). Obviation in subjunctive clauses and agr:
Evidence from Russian. Natural Language and Linguistic Theory, 15:229–62.
Axelrod, M. (1993). The Semantics of Time: Aspectual Categorization in Koyukon
Athabaskan. University of Nebraska Press, Lincoln, NE.
Bach, K. (1975). Performatives are statements too. Philosophical Studies, 28:229–36.
Bach, K. and Harnish, R. M. (1979). Linguistic Communication and Speech Acts.
MIT Press, Cambridge, MA.
Baker, M. and Travis, L. (1997). Mood as verbal definiteness in a tenseless language.
Natural Language Semantics, 5:213–69.
Ballmer, T. (1978). Logical Grammar. North Holland, Amsterdam.
Barker, C. (2010). Free choice permission as resource-sensitive reasoning.
Semantics and Pragmatics, 3(10):1–38.
Barker, C. (2012). Imperatives denote actions. In Guevara, A. A., Chernilovskaya,
A., and Nouwen, R., eds, The Proceedings of Sinn und Bedeutung 16, pages 57–70.
MITWPL, Cambridge, MA.
Barwise, J. (1981). Scenes and other situations. Journal of Philosophy, 78:369–97.
Beaver, D. and Clark, B. (2008). Sense and Sensitivity: How Focus Determines
Meaning. Wiley-Blackwell, Oxford.
Bell, A. (1984). Language style as audience design. Language in Society, 13(2):
145–204.
Beyssade, C. and Marandin, J.-M. (2006). The speech act assignment problem
revisited: Disentangling speaker’s commitment from speaker’s call on addressee.
In Bonami, O. and Hofherr, P. C., eds, Empirical Issues in Syntax and Semantics 6,
pages 37–68. http://www.cssp.cnrs.fr/eiss6/.
Bhat, D. (1999). The Prominence of Tense, Aspect, and Mood. John Benjamins,
Amsterdam.
Bierwisch, M. (1980). Semantic structure and illocutionary force. In Searle, J. R.,
Kiefer, F., and Bierwisch, M., eds, Speech Act Theory and Pragmatics, pages 1–35.
Reidel, Dordrecht.

i i

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

i i

references 255

Biezma, M. (2011). Optatives: Deriving desirability from scalar alternatives.

In Reich, I., ed., Proceedings of Sinn und Bedeutung 15. Universaar–Saarland
University Press, Saarbrücken.
Biezma, M. and Rawlins, K. (2012). Responding to alternative and polar questions.
Linguistics and Philosophy, 35(5):361–406.
Bittner, M. (2011). Time and modality without tenses or moods. In Musan, R. and
Rathert, M., eds, Tense across Languages. Niemeyer, Tübingen.
Blanco, M. T. (2011). Causatives in Minimalism. John Benjamins, Amsterdam.
Bloomfield, L. (1956). Eastern Ojibwa: Grammatical Sketch, Texts and Word List.
The University of Michigan Press, Ann Arbor.
Boas, F. (1911). Handbook of American Indian Languages, Part 1. Smithsonian
Institution, Bureau of American Ethnology, Washington; Reprint: Thoemmes
Press, 2002. Bristol, England.
Boër, S. E. and Lycan, W. G. (1980). A performadox in truth-conditional semantics.
Linguistics and Philosophy, 4(1):1–46.
Bolinger, D. (1967). The imperative in English. In Halle, M., Lunt, H., and
MacLean, H., eds, To Honor Roman Jakobson: Essays on the Occasion of his
Seventieth Birthday, volume 1 of Janua Linguarum, pages 335–62. Mouton, The
Hague and Paris.
Bolinger, D. (1976). Again—one or two subjunctives? Hispania, 59(1):41–9.
Bolinger, D. L. (1968). Post-posed main phrases: An English rule for the Romance
subjunctive. Canadian Journal of Linguistics, 14:3–30.
Bolinger, D. L. (1974). One subjunctive or two? Hispania, 57(3):462–71.
Brandner, E. (2004). Head-movement in minimalism and V2 as force-marking.
In Lohnstein, H. and Trissler, S., eds, The Syntax and Semantics of the Left
Periphery, pages 97–138. Mouton de Gruyter, Berlin.
Brandt, M., Reis, M., Rosengren, I., and Zimmermann, I. (1992). Satztyp,
Satzmodus, und Illokution. In Rosengren, I., ed., Satz und Illokution, volume 1,
pages 1–90. Niemeyer, Tübingen.
Brasoveanu, A. (2006). Temporal and propositional de se: Evidence from
Romanian subjunctive mood. In Ebert, C. and Endriss, C., eds, The Proceedings
of Sinn und Bedeutung 10, pages 55–70. ZASPIL.
Bresnan, J. (1972). Theory of Complementation in English Syntax. PhD thesis, MIT.
Brown, C. (2011). Narrow mental content. In Zalta, E. N., ed., The Stanford En-
cyclopedia of Philosophy. Fall 2011 edition. http://plato.stanford.edu/archives/
fall2011/entries/content-narrow/.
Brown, M. (1988). On the logic of ability. Journal of Philosophical Logic, 17(1):
1–26.
Büring, D. (2003). On D-trees, beans, and B-accents. Linguistics and Philosophy,
26:511–45.
Bybee, J. L. (1998). “Irrealis” as a grammatical category. Anthropological Linguistics,
40(2):257–71.
Castroviejo, E. (2006a). A degree-based account of wh-exclamatives. In Puig-
Waldmüller, E., ed., Proceedings of Sinn und Bedeutung 11. Universitat Pompeu
Fabra.
Castroviejo, E. (2006b). Wh-exclamatives in Catalan. PhD thesis, University of
Barcelona.

i i

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

i i

256 references

Chafe, W. (1995). The realis–irrealis distinction in Caddo, the northern Iroquoian

languages, and English. In Bybee, J. L. and Fleischman, S., eds, Modality in
Grammar and Discourse. John Benjamins, Amsterdam.
Charlow, N. (2011). Practical Language: Its Meaning and Use. PhD thesis, Univer-
sity of Michigan.
Charlow, N. (2013). What we know and what to do. Synthese, 190:2291–323.
Charlow, S. and Sharvit, Y. (2014). Bound “de re” pronouns and the LFs of attitude
reports. Semantics and Pragmatics, 7(3):1–43.
Chellas, B. (1969). The Logical Form of Imperatives. Perry Lane Press, Stanford, CA.
Chellas, B. (1971). Imperatives. Theoria, 37:114–29.
Chen-Main, J. (2005). Characteristics of Mandarin imperatives. In Brandstetter, C.
and Rus, D., eds, Georgetown University Working Papers in Theoretical Linguist-
ics, volume IV, pages 1–51. Georgetown University Department of Linguistics,
Washington, DC.
Cheng, L. L.-S. (1991). On the typology of WH-questions. PhD thesis, MIT,
Cambridge.
Chierchia, G. (1989). Anaphora and attitudes “de se”. In Bartsch, R., van Benthem,
J., and van Emde Boas, P., eds, Semantics and Contextual Expressions, pages 1–31.
Foris, Dordrecht.
Chierchia, G. (2004). Scalar implicatures, polarity phenomena, and the syn-
tax/pragmatics interface. In Belletti, A., ed., Structures and Beyond. Oxford
University Press, Oxford.
Christidis, A. (1981). Oti/pos–pu: complementizer choice in Modern Greek. Studies
in Greek Linguistics, 2:113–77.
Chung, K.-S. (2007). Spatial deictic tense and evidentials in Korean. Natural
Language Semantics, 15:187–219.
Ciardelli, I., Groenendijk, J., and Roelofsen, F. (2013). Inquisitive semantics: a new
notion of meaning. Language and Linguistics Compass, 7(9).
Clark, B. (1993). Relevance and “pseudo-imperatives”. Linguistics and Philosophy,
16(1):79–121.
Condoravdi, C. and Lauer, S. (2011). Performative verbs and performative acts.
In Reich, I., Horch, E., and Pauly, D., eds, Proceedings of Sinn und Bedeutung
15, pages 149–64. Universaar–Saarland University Press, Saarbrücken.
Condoravdi, C. and Lauer, S. (2012). Imperatives: Meaning and illocutionary force.
In Piñón, C., ed., Empirical Issues in Syntax and Semantics 9. Colloque de
Syntaxe et Sémantique à Paris, Paris.
Coppock, E. and Wechsler, S. (to appear). The proper treatment of egophoricity in
Kathmandu Newari. In Jaszczolt, K. M. and Huang, M., eds, Expressing the Self:
Cultural Diversity and Cognitive Universals. Oxford University Press, Oxford.
Costantini, F. (2006). Obviation in subjunctive argument clauses and the
first-personal interpretation. In Frascarelli, M., ed., Phases of Interpretation,
pages 295–319. Mouton de Gruyter, Berlin.
Cresswell, M. J. and von Stechow, A. (1982). De re belief generalized. Linguistics and
Philosophy, 5(6):503–35.
Crnič, L. and Trinh, T. (2009). Embedded imperatives in English. In Riester, A. and
Solstad, T., eds, Proceedings of Sinn und Bedeutung 13, pages 113–27. University
of Stuttgart.

i i

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

i i

references 257

Crnič, L. and Trinh, T. (2011). Embedding imperatives. In Lima, S., Mullin, K.,
and Smith, B., eds, Proceedings of NELS 39, volume 1. GLSA, University of
Massachusetts, Amherst.
Cross, C. and Roelofsen, F. (2014). Questions. In Zalta, E. N., ed., The Stanford
Encyclopedia of Philosophy. Fall 2014 edition. http://plato.stanford.edu/archives/
fall2014/entries/questions/.
Cui, Y. (2015). Modals in the Scope of Attitudes: A Corpus Study of Attitude–Modal
Combinations in Mandarin. PhD thesis, Georgetown University.
Davidson, D. (1979). Moods and performances. In Margalit, A., ed., Meaning and
Use, pages 9–20. Reidel, Dordrecht.
Davies, E. (1986). The English Imperative. Croom Helm, London.
Dayal, V. (2016). Questions. Oxford University Press, Oxford and New York.
Deal, A. R. (2011). Modals without scales. Language, 87(3):559–85.
de Haan, F. (2012). Irrealis: fact or fiction? Language Sciences, 34(2):
107–30.
Dench, A. (1991). Panyjima. In Dixon, R. and Blake, B. J., eds, The Handbook
of Australian Languages, volume 4, pages 125–244. Oxford University Press,
Melbourne.
Dixon, R. M. W. (1982). Where Have All the Adjectives Gone? And Other Essays in
Semantics and Syntax. Mouton, Berlin.
Dobrovie-Sorin, C. (2001). Head-to-head merge in Balkan subjunctives and local-
ity. In Rivero, M.-L. and Ralli, A., eds, Comparative Syntax of Balkan Languages,
pages 44–73. Oxford University Press, New York.
Donnellan, K. (1966). Reference and definite descriptions. Philosophical Review,
75:281–304.
Downing, B. T. (1969). Vocatives and third-person imperatives in English. Papers
in Linguistics, 1:570–92.
Dudman, V. H. (1970). Frege’s judgment-stroke. The Philosophical Quarterly,
20(79): 150–61.
Dudman, V. H. (1972). Frege on assertion. The Philosophical Quarterly, 22(86):
61–4.
Elliott, J. (2000). Realis and irrealis: forms and concepts of the grammaticalisation
of reality. Linguistic Typology, 4:55–90.
Emonds, J. E. (1969). Root and Structure-preserving Transformations. PhD thesis,
MIT.
Fabricius-Hansen, C. and Saebø, K. J. (2004). In a mediative mood: The semantics
of the German reportative subjunctive. Natural Language Semantics, 12(3):
213–57.
Faller, M. (2002). Semantics and Pragmatics of Evidentials in Cuzco Quechua. PhD
thesis, Stanford University.
Faller, M. (2006a). The Cusco Quechua reportative evidential and rhetorical rela-
tions. Ms., University of Manchester.
Faller, M. (2006b). Evidentiality and epistemic modality at the semantics/
pragmatics interface. Paper presented at the 2006 Workshop on Philosophy and
Linguistics, University of Michigan.
Faller, M. (2006c). Evidentiality below and above speech acts. Ms., University of
Manchester.

i i

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

i i

258 references

Faller, M. (2007). The Cuzco Quechua conjectural, epistemic modal or evidential?

or both? Talk given at SULA 4, Universidade de São Paulo.
Faller, M. (2011). A possible worlds semantics for Cuzco Quechua evidentials.
In Li, N. and Lutz, D., eds, The Proceedings of SALT 20, pages 660–83. Linguistic
Society of America.
Farkas, D. (1985). Intensional Descriptions and the Romance Subjunctive Mood.
Garland, New York.
Farkas, D. (1992a). On obviation. In Sag, I. and Szabolsci, A., eds, Lexical Matters,
pages 85–109. CSLI Publications, Stanford, CA.
Farkas, D. (1992b). On the semantics of subjunctive complements. In Hirschbueler,
P. and Koerner, K., eds, Romance Languages and Modern Linguistic Theory,
pages 69–104. Benjamins, Amsterdam and Philadephia.
Farkas, D. (2003). Assertion, belief, and mood choice. Paper presented at the
workshop on conditional and unconditional modality, Vienna.
Farkas, D. and Bruce, K. B. (2010). On reacting to assertions and polar questions.
Journal of Semantics, 27(1):81–118.
Farkas, D. and Roelofsen, F. (2017). Division of labor in the interpretation of
declaratives and interrogatives. Journal of Semantics, 34(2): 237–89.
Fine, K. (2016). Survey of truthmaker semantics. Ms., New York University.
Finlay, S. (2010). What ought probably means, and why you can’t detach it. Synthese,
177:67–89.
von Fintel, K. and Gillies, A. S. (2010). Must . . . stay . . . strong. Natural Language
Semantics, 18:351–83.
von Fintel, K. and Iatridou, S. (2008). How to say ought in foreign: The composition
of weak necessity modals. In Guéron, J. and Lecarme, J., eds, Time and Modality,
pages 115–41. Springer, Berlin.
von Fintel, K. and Iatridou, S. (2015). A modest proposal for the meaning of
imperatives. Ms., MIT.
Fox, D. (2007). Free choice disjunction and the theory of scalar implicatures. In
Sauerland, U. and Stateva, P., eds, Presupposition and Implicature in Composi-
tional Semantics, pages 71–120. Palgrave-Macmillan, Basingstoke.
Franke, M. (2005). Pseudo-imperatives. Ms., University of Amsterdam.
Franke, M. (2009). Signal to Act: Game Theory in Pragmatics. PhD thesis, Univer-
sity of Amsterdam.
Gärtner, H.-M. (2002). On the force of V2 declaratives. Theoretical Linguistics,
28:33–42.
Gazdar, G. (1979). Pragmatics: Implicature, Presupposition, and Logical Form.
Academic Press, London.
Gazdar, G. (1981). Speech act assignment. In Joshi, A., Weber, B. H., and Sag, I. A.,
eds, Elements of Discourse Understanding, pages 64–83. Cambridge University
Press, Cambridge.
Geach, P. T. (1958). Imperative and deontic logic. Analysis, 18:49–56.
Geurts, B. (2005). Entertaining alternatives: Disjunctions as modals. Natural Lan-
guage Semantics, 13(4):383–410.
Geurts, B. and Huitink, J. (2006). Modal concord. In Dekker, P. and
Zeijlstra, H., eds, Proceedings of the ESSLLI Workshop Concord Phenomena at

i i

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

i i

references 259

the Syntax–Semantics Interface, Malaga, pages 15–20. Available at www.ru.nl/

ncs/janneke/modalconcord-esslli06.pdf.
Giannakidou, A. (1994). The semantic licensing of npis and the modern Greek
subjunctive. In Language and Cognition 4: Yearbook of the Research Group for
Theoretical and Experimental Linguistics, pages 55–68. University of Groningen,
Groningen.
Giannakidou, A. (1995). Subjunctive, habituality, and negative polarity items.
In Simons, M. and Galloway, T., eds, The Proceedings of SALT 5, pages 94–111.
CLC Publications, Cornell University Linguistics Department.
Giannakidou, A. (1997). The Landscape of Polarity Items. PhD thesis, Groningen.
Giannakidou, A. (1998). Polarity Sensitivity as (Non)veridical Dependency. John
Benjamins, Amsterdam.
Giannakidou, A. (1999). Affective dependencies. Linguistics and Philosophy,
22(3):367–421.
Giannakidou, A. (2009). The dependency of the subjunctive revisited: Temporal
semantics and polarity. Lingua, 119(12):1883–908.
Giannakidou, A. (2011). Nonveridicality and mood choice: subjunctive, polarity,
and time. In Musan, R. and Rathert, M., eds, Tense across Languages. Niemeyer,
Tübingen.
Giannakidou, A. (2014). (Non)veridicality, Evaluation, and event actualization:
Evidence from the subjunctive in relative clauses. In Taboada, M. and
Trnavac, R., eds, Nonveridicality and Evaluation: Theoretical, Computational,
and Corpus Approaches, Studies in Pragmatics. Brill, Leiden.
Giannakidou, A. (2016). Evaluative subjunctive and nonveridicality. In
Błaszczak, J., Giannakidou, A., Klimek-Jankowska, D., and Migdalski, K.,
eds, Mood, Aspect, Modality Revisited: New Answers to Old Questions, pages
177–217. University of Chicago Press, Chicago, IL.
Giannakidou, A. and Mari, A. (2015). Mixed veridicality and mood choice in com-
plement clauses. Ms, University of Chicago. Available at home.uchicago.edu/
giannaki/pubs/GM.subjunctive.typology.pdf.
Ginet, C. (1979). Performativity. Linguistics and Philosophy, 3(2):245–65.
Ginzburg, J. (1995a). Resolving questions, part I. Linguistics and Philosophy,
18(5):459–527.
Ginzburg, J. (1995b). Resolving questions, part II. Linguistics and Philosophy,
18(6):567–609.
Ginzburg, J. (1996). Dynamics and the semantics of dialogue. In Seligman, J.,
ed., Language, Logic and Computation, Volume 1, CSLI Lecture Notes. CSLI,
Stanford, CA.
Ginzburg, J. and Sag, I. A. (2001). Interrogative Investigations: The Form, Meaning,
and Use of English Interrogatives. The University of Chicago Press, Chicago, IL.
Giorgi, A. and Pianesi, F. (1997). Tense and Aspect: From Semantics to Morphosyn-
tax. Oxford University Press, Oxford.
Givón, T. (1994). Irrealis and the subjunctive. Studies in Language, 18:265–337.
Gordon, D. and Lakoff, G. (1971). Conversational postulates. In Papers from the Sev-
enth Regional Meeting of the Chicago Linguistic Society, pages 63–84. University
of Chicago Press, Chicago, IL.

i i

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

i i

260 references

Gordon, D. and Lakoff, G. (1975). Conversational postulates. In Cole, P. and

Morgan, J. L., eds, Syntax and Semantics, Volume 3: Speech Acts, pages 83–106.
Academic Press, New York.
Gregory, A. E. and Lunn, P. (2012). A concept-based approach to the subjunctive.
Hispania, 95(2):333–43.
Groenendijk, J. (2009). Inquisitive semantics: Two possibilities for disjunction.
In Bosch, P., Gabelaia, D., and Lang, J., eds, Seventh International Tbilisi Sym-
posium on Language, Logic, and Computation. Springer.
Groenendijk, J. and Roelofsen, F. (2009). Inquisitive semantics and pragmatics.
Presented at the Stanford workshop on Language, Communication and Rational
Agency.
Groenendijk, J. and Stokhof, M. (1982). Semantic analysis of wh-complements.
Linguistics and Philosophy, 5:175–233.
Groenendijk, J. and Stokhof, M. (1984). Studies on the Semantics of Questions and
the Pragmatics of Answers. PhD thesis, University of Amsterdam.
Groenendijk, J. and Stokhof, M. (1990). Dynamic Montague grammar. In Kálman,
L. and Pólos, L., eds, Papers from the Symposium on Logic and Language,
pages 3–48. Adakémiai Kiadó, Budapest.
Groenendijk, J. and Stokhof, M. (1991). Dynamic predicate logic. Linguistics and
Philosophy, 14:39–100.
Groenendijk, J. and Stokhof, M. (1997). Questions. In van Benthem, J. and ter
Meulen, A., eds, Handbook of Logic and Language, pages 1055–124. Elsevier/MIT
Press, Amsterdam and Cambridge, MA.
Groenendijk, J., Stokhof, M., and Veltman, F. (1996). Coreference and modality. In
Lappin, S., ed., The Handbook of Contemporary Semantic Theory, pages 179–213.
Blackwell Publishers, Oxford.
Groenendijk, J., Stokhof, M., and Veltman, F. (1997). Coreference and modality in
multi-speaker discourse. In Kamp, H. and Partee, B., eds, Context Dependence
in the Analysis of Linguistic Meaning, pages 195–216. I.M.S., Stuttgart.
Grosz, P. (2010). Grading modality: A new approach to modal concord and its
relatives. In Prinzhorn, M., Schmitt, V., and Zobel, S., eds, Proceedings of Sinn
und Bedeutung 14, pages 185–201. Vienna.
Grosz, P. (2011). On the Grammar of Optative Constructions. PhD thesis,
Massachusetts Institute of Technology.
Grosz, P. (2014). Optative markers as communicative cues. Natural Language
Semantics, 22(1):89–115.
Gunlogson, C. (2001). True to Form: Rising and Falling Declaratives as Questions in
English. Routledge, Abingdon.
Gutiérrez-Rexach, J. (1996). The semantics of exclamatives. In Garret, E. and Lee,
F., eds, Syntax at Sunset, UCLA Working Papers in Linguistics. Department of
Linguistics, UCLA, Los Angeles, CA.
Hacquard, V. (2010). On the event-relativity of modal auxiliaries. Natural Language
Semantics, 18(1):79–114.
Haegeman, L. (2006). Conditionals, factives and the left periphery. Lingua,
116(10):1651–69. (Special issue: Language in Mind: A Tribute to Neil Smith on
the Occasion of his Retirement.)

i i

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

i i

references 261

Halpern, J. (2003). Reasoning about Uncertainty. MIT Press, Cambridge, MA.

Hamblin, C. L. (1958). Questions. Australasian Journal of Philosophy, 36:
159–68.
Hamblin, C. L. (1971). Mathematical models of dialogue. Theoria, 37(2):
130–55.
Hamblin, C. L. (1973). Questions in Montague English. Foundations of Language,
10:41–53.
Han, C.-H. (1998). The Structure and Interpretation of Imperatives: Mood and
Force in Universal Grammar. PhD thesis, University of Pennsylvania.
Han, C.-H. (2011). Imperatives. In von Heusinger, K., Maienborn, C., and Portner,
P., eds, Semantics: An International Handbook of Natural Language Meaning.
Mouton de Gruyter, Berlin.
Han, C.-H. and Lee, C. (2007). On negative imperatives in Korean. Linguistic
Inquiry, 38(2):373–95.
Hansson, S. O. (1990). Preference-based deontic logic (pdl). Journal of Philosophical
Logic, 19(1):75–93.
Hansson, S. O. (2001). Structures of Values and Norms. Cambridge University Press,
Cambridge.
Harner, H. (2016). The Modality of Desire Predicates and Directive Verbs. PhD
thesis, Georgetown University.
Harnish, R. M. (1994). Mood, meaning, and speech acts. In Tsohatzidis, S. L.,
ed., Foundations of Speech Act Theory: Philosophical and Linguistic Perspectives,
pages 407–459. Routledge, London.
Harper, W. L. (1976). Ramsey test conditionals and iterated belief change. In
Harper, W. L. and Hooker, C. A., eds, Foundations of Probability Theory, Stat-
istical Inference, and Statistical Theories of Science. Reidel, Dordrecht.
Hausser, R. (1980). Surface compositionality and the semantics of mood. In
J. Searle, F. Kiefer, and M. Bierwisch, eds, Speech Act Theory and Pragmatics,
pages 71–95. Reidel, Dordrecht and Boston.
Hausser, R. (1983). The syntax and semantics of English mood. In Kiefer, F., ed,
Questions and Answers. Reidel, Dordrecht.
Haverkate, H. (2002). The Syntax, Semantics and Pragmatics of Spanish Mood. John
Benjamins, Amsterdam.
Heim, I. (1982). The Semantics of Definite and Indefinite Noun Phrases. PhD thesis,
University of Massachusetts, Amherst.
Heim, I. (1983). File Change Semantics and the familiarity theory of definiteness.
In Bauerle, R., Schwarze, C., and von Stechow, A., eds, Meaning, Use, and
Interpretation of Language, pages 164–89. Mouton de Gruyter, Berlin. Reprinted
in Portner, P. and Partee, B. H., eds (2002). Formal Semantics: The Essential
Readings. Blackwell, Oxford.
Heim, I. (1988). On the projection problem for presuppositions. In Barlow, M.,
Flickinger, D., and Wiegand, N., eds, Proceedings of WCCFL 2, pages 114–25.
Stanford University. Reprinted in Portner, P. and Partee B. H., eds (2002). Formal
Semantics: The Essential Readings. Blackwell, Oxford.
Heim, I. (1992). Presupposition projection and the semantics of attitude verbs.
Journal of Semantics, 9:183–221.

i i

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

i i

262 references

Higginbotham, J. (2003). Remembering, imagining, and the first person. In Barber,

A., ed., Epistemology of Language, pages 496–533. Oxford University Press,
Oxford.
Hintikka, J. (1961). Modality and quantification. Theoria, 27:110–28.
Holliday, W. H. and Icard, T. F. (2013). Measure semantics and qualitative semantics
for epistemic modals. In Snider, T., ed., The Proceedings of SALT 23. CLC
Publications, Ithaca, NY.
Hooper, J. and Thompson, S. A. (1973). On the applicability of root transformations.
Linguistic Inquiry, 4:465–97.
Huitink, J. (2012). Modal concord: A case study of Dutch. Journal of Semantics,
29(3):403–37.
Hulstijn, J. (1997). Structured information states: Raising and resolving issues.
In Benz, A. and Jager, G., eds, Proceedings of MunDial97. University of Munich.
Huntley, M. (1980). Propositions and the imperative. Synthese, 45(2):281–310.
Huntley, M. (1984). The semantics of English imperatives. Linguistics and Philo-
sophy, 7:103–34.
Iatridou, S. (2000). The grammatical ingredients of counterfactuality. Linguistic
Inquiry, 31(2):231–70.
Jäger, G. (1996). Only updates: On the dynamics of the focus particle only.
In Dekker, P. and Stokhof, M., eds, Proceedings of the 10th Amsterdam Col-
loquium. ILLC, University of Amsterdam.
Jary, M. (2011). Assertion, relevance and the declarative mood. In Escandell-
Vidal, V., Leonetti, M., and Ahern, A., eds, Procedural Meaning: Problems and
Perspectives, pages 267–89. Brill, Leiden.
Jary, M. and Kissine, M. (2014). Imperatives. Cambridge University Press,
Cambridge.
Jary, M. and Kissine, M. (2016). Imperatives as (non-)modals. In Blaszczak, J., ed.,
Cross-linguistic Approaches to Tense, Aspect and Mood. University of Chicago
Press, Chicago, IL.
Jensen, B. (2003). Syntax and semantics of imperative subjects. Nordlyd, 31:150–64.
Kamp, H. (1973). Free choice permission. Proceedings of the Aristotelian Society,
N.S., 74:57–74.
Kamp, H. (1981). A theory of truth and semantic representation. In Groenendijk,
J., Janssen, T., and Stokhof, M., eds, Formal Methods in the Study of Language,
pages 277–322. Mathematical Centre, Amsterdam.
Kamp, H. and Reyle, U. (1993). From Discourse to Logic: Introduction to Model-
theoretic Semantics of Natural Language, Formal Logic and Discourse Represent-
ation Theory. Kluwer, Dordrecht.
Kanger, S. (1971). New foundations for ethical theory. In Hilpinen, R., ed., Deontic
Logic: Introductory and Systematic Readings, pages 36–58. D. Reidel, Dordrecht.
Karttunen, L. (1971). Some observations on factivity. Papers in Linguistics, 5:55–69.
Karttunen, L. (1973). Presuppositions of compound sentences. Linguistic Inquiry,
4:169–93.
Karttunen, L. (1976). Discourse referents. In McCawley, J., ed., Notes from the
Linguistic Underground, volume 7 of Syntax and Semantics. Academic Press,
New York.

i i

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

i i

references 263

Karttunen, L. (1977). Syntax and semantics of questions. Linguistics and Philosophy,

1:3–44.
Karttunen, L. and Peters, S. (1975). Conventional implicature. In Oh, C.-Y. and
Dinneen, D., eds, Syntax and Semantics, vol. 11: Presupposition, pages 1–56.
Academic Press, New York.
Katz, G. (2005). Attitudes toward degrees. In Maier, E., Barry, C., and
Huitink, J., eds, Proceedings of Sinn und Bedeutung 9. Nijmegen Centre of
Semantics (NCS).
Katz, G., Portner, P., and Rubinstein, A. (2012). Ordering combination for modal
semantics. In Chereches, A., ed., The Proceedings of SALT 22, pages 488–507.
CLC Publications, Stanford, CA.
Katz, J. J. and Postal, P. M. (1964). An Integrated Theory of Linguistic Descriptions.
MIT Press, Cambridge, MA.
Kaufmann, M. (2012). Interpreting Imperatives. Springer.
Kaufmann, M. (2016). Free choice as a form of dependence. Natural Language
Semantics, 24(3):247–90.
Kaufmann, M. and Poschmann, C. (2013). Embedded imperatives: Empirical evid-
ence from colloquial German. Language, 89(3):619–37.
Kempchinsky, P. (1985). The subjunctive disjoint reference effect. In Neidle, C. and
Nuñez Cedeño, R., eds, Studies in Romance Linguistics, pages 123–40. Foris,
Dordrecht.
Kempchinsky, P. (2009). What can the subjunctive disjoint reference effect tell us
about the subjunctive? Lingua, 119(12):1788–810.
Keshet, E. (2011). Split intensionality: A new scope theory of de re and de dicto.
Linguistics and Philosophy, 33(4):251–83.
Kinkade, M. D. (1998). Is irrealis a grammatical category in Upper Chehalis?
Anthropological Linguistics, 40(2):234–44.
Klamer, M. (2012). Reality status in Teiwa (Papuan). Language Sciences, 34(2):
216–28.
Klecha, P. (2012). Positive and conditional semantics for gradable modals.
In Guevara, A. A., Chernilovskaya, A., und Nouwen, R., eds, Proceedings of Sinn
und Bedeutung 16, volume 2, pages 363–76. MITWPL.
Klecha, P. (2014). Bridging the Divide: Scalarity and Modality. PhD thesis, Univer-
sity of Chicago.
Kolodny, N. and MacFarlane, J. (2010). Ifs and oughts. The Journal of Philosophy,
108(3):115–43.
König, E. and Siemund, P. (2007). Speech act distinctions in grammar. In Shopen,
T., ed., Language Typology and Syntactic Description, volume 1, pages 276–324.
Cambridge University Press, Cambridge.
Kratzer, A. (1977). What “must” and “can” must and can mean. Linguistics and
Philosophy, 1(1):337–55.
Kratzer, A. (1981). The notional category of modality. In Eikmeyer, H.-J. and Rieser,
H., eds, Words, Worlds, and Contexts, pages 38–74. de Gruyter, Berlin.
Kratzer, A. (1991). Modality. In von Stechow, A. and Wunderlich, D., eds,
Semantik/Semantics: An International Handbook of Contemporary Research,
pages 639–50. de Gruyter, Berlin.

i i

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

i i

264 references

Kratzer, A. (2006). Decomposing attitude verbs. Talk at The Hebrew University

of Jerusalem honoring Anita Mittwoch on her 80th birthday. Available at
semanticsarchive.net/Archive/DcwY2JkM/attitude-verbs2006.pdf.
Kratzer, A. (2012). Modal and Conditionals: New and Revised Perspectives. Oxford
University Press, Oxford.
Kratzer, A. (2013). Constructing domains for deontic (and other) modals. Slides for
a talk given at the USC deontic modality workshop.
Kratzer, A. and Shimoyama, J. (2002). Indeterminate phrases: The view from
Japanese. In Otsu, Y., ed., The Proceedings of the Third Tokyo Conference on
Psycholinguistics, pages 1–25. Hituzi Syobo, Tokyo.
Krawczyk, L. (2012). Inferred Propositions and the Expression of the Evidence
Relation in Natural Language: Evidentiality in Central Alaskan Yup’ik Eskimo
and English. PhD thesis, Georgetown University.
Krifka, M. (2001). Quantifying into question acts. Natural Language Semantics,
9:1–40.
Krifka, M. (2011). Questions. In von Heusinger, K., Maienborn, C., and Portner, P.,
eds, Semantics: An International Handbook of Natural Language Meaning,
volume 2, pages 1742–85. Mouton de Gruyter, Berlin.
Krifka, M. (2014). Embedding illocutionary acts. In Roeper, T. and Speas, M., eds,
Recursion: Complexity in Cognition. Springer.
Laca, B. (2010). Mood in Spanish. In Thieroff, R. and Rothstein, B. eds, Mood
in the Languages of Europe, pages 199–220. John Benjamins, Amsterdam and
Philadelphia.
Lascarides, A. and Asher, N. (2003). Imperatives in dialogue. In Kühnlein, P.,
Rieser, H., and Zeevat, H., eds, Perspectives on Dialogue in the New Millennium,
pages 1–24. Benjamins, Philadelphia.
Lasersohn, P. (2017). Subjectivity and Perspective in Truth-Conditional Semantics.
Oxford University Press, Oxford.
Lassiter, D. (2011). Measurement and Modality: The Scalar Basis of Modal
Semantics. PhD thesis, New York University.
Lassiter, D. (2016). Graded Modality. Oxford University Press, Oxford.
Lauer, S. (2013). Towards a Dynamic Pragmatics. PhD thesis, Stanford University.
Lemmon, E. J. (1965). Deontic logic and the logic of imperatives. Logique et Analyse,
8:39–71.
Levinson, S. C. (1980). Speech act theory: The state of the art. Language Teaching
and Linguistics: Abstracts, pages 5–24.
Levinson, S. C. (1981). The essential inadequacies of speech act models of dialogue.
In Parret, H., Sbisà, M., and Verschueren, J., eds, Possibilities and Limitations of
Pragmatics: Proceedings of the Conference on Pragmatics, Urbino, July 8–14, 1979,
pages 473–92. John Benjamins, Amsterdam.
Levinson, S. C. (1983). Pragmatics. Cambridge University Press, Cambridge.
Lewis, D. (1972). General semantics. In Davidson, D. and Harman, G., eds,
Semantics of Natural Language. Reidel, Dordrecht.
Lewis, D. (1975). Adverbs of quantification. In Keenan, E., ed., Formal Semantics of
Natural Language, pages 3–15. Cambridge University Press, Cambridge.
Lewis, D. K. (1979a). Attitudes de dicto and de se. Philosophical Review, 88:513–43.

i i

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

i i

references 265

Lewis, D. K. (1979b). A problem about permission. In Saarinen, E., Hilpinen, R.,

Niiniluoto, I., and Hintikka, M. P., eds, Essays in Honour of Jaakko Hintikka,
pages 163–75. Reidel, Dordrecht.
Lewis, D. K. (1979c). Scorekeeping in a language game. Journal of Philosophical
Logic, 8:339–59.
Lohnstein, H. (2000). Satzmodus-kompositionell. Zur Parametrisierung der Modus-
phrase im Deutschen. Akademie Verlag, Berlin.
Lohnstein, H. (2007). On clause types and sentential force. Linguistische Berichte,
209:63–86.
Luján, M. (1979). Clitic promotion and mood in Spanish verbal complements.
IULC, Bloomington, Indiana.
Luján, M. (1980). Clitic promotion and mood in Spanish verbal complements.
Linguistics, pages 381–484.
Lunn, P. V. (1995). The evaluative function of the Spanish subjunctive. In Bybee, J.
and Fleischman, S., eds, Modality in Grammar and Discourse, pages 429–49.
Benjamins, Amsterdam.
McCready, E. and Ogata, N. (2007). Evidentiality, modality and probability.
Linguistics and Philosophy, 30(2):147–206.
McGregor, W. B. and Wagner, T. (2006). The semantics and pragmatics of irrealis
mood in Nyulnyulan languages. Oceanic Linguistics, 45(2):339–79.
McKay, T. and Nelson, M. (2014). Propositional attitude reports. In Zalta,
E. N., ed., The Stanford Encyclopedia of Philosophy. Spring 2014 edition.
http://plato.stanford.edu/archives/spr2014/entries/prop-attitude-reports/.
Maier, E. (2010). Presupposing acquaintance: a unified theory of de dicto, de re and
de se belief reports. Linguistics and Philosophy, 32:429–74.
Malamud, S. and Stephenson, T. (2015). Three ways to avoid commitments: De-
clarative force modifiers in the conversational scoreboard. Journal of Semantics,
32:275–311.
Mascarenhas, S. (2009). Inquisitive Semantics and Logic. Master’s thesis, Univer-
sity of Amsterdam.
Mastop, R. (2005). What Can You Do: Imperative Mood in Semantic Theory. PhD
thesis, University of Amsterdam.
Mastop, R. (2011). Imperatives as semantic primitives. Linguistics and Philosophy,
34(4):305–40.
Matić, D. and Nikolaeva, I. (2014). Realis mood, focus, and existential closure in
Tundra Yukaghir. Lingua, 150:202–31.
Matthewson, L., Rullmann, H., and Davis, H. (2007). Evidentials as epistemic
modals: Evidence from St’át’imcets. In Craenebroeck, J. V. and Rooryck,
J., eds, The Linguistic Variation Yearbook 7, pages 201–54. John Benjamins,
Amsterdam.
Mauri, C. and Sansò, A., eds (2012). Language Sciences, 34.2, Special issue, Papers
selected from the workshop “What do languages encode when they encode
reality status?” at the 41st Annual Meeting of the Societas Linguistica Europaea,
Forlì, Italy, 17–20 September 2008.
Medeiros, D. J. (2015). Embedded Ancient Greek imperatives: A feature transfer
analysis. Syntax, 18(2):124–56.

i i

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

i i

266 references

Meibauer, J. (1989). Ob sie wohl kommt? Zum Satzmodus von selbständigen

Sätzen mit Endstellung des finiten Verbs. In Katny, A., ed., Studien zur kon-
trastiven Linguistik und literarischen Übersetzung, pages 11–33. Lang, Frankfurt
am Main.
Meinunger, A. (2004). Verb position, verbal mood and the anchoring (potential)
of sentences. In Lohnstein, H. and Trissler, S., eds, The Syntax and Semantics of
the Left Periphery, pages 313–41. Mouton de Gruyter, Berlin.
Menéndez-Benito, P. (2005). The Grammar of Choice. PhD thesis, University of
Massachusetts, Amherst.
Merin, A. (1991). Imperatives: linguistics vs. philosophy. Linguistics, 29(4):
669–702.
Merin, A. (1994). Algebra of elementary social acts. In Tsohatzidis, S. L., ed.,
Foundations of Speech Act Theory: Philosophical and Linguistic Perspectives,
pages 234–66. Routledge, London.
Michaelis, L. A. and Lambrecht, K. (1996). Toward a construction-based theory of
language function: The case of nominal extraposition. Language, 72(2):215–47.
Mithun, M. (1995). The relativity of irreality. In Bybee, J. L. and Fleischmann,
S., eds, Modality in Grammar and Discourse, pages 367–88. John Benjamins,
Amsterdam and Philadelphia.
Mithun, M. (1999). The Languages of Native North America. Cambridge University
Press, Cambridge.
Moretti, G. and Orvieto, G. (1981). Grammatica Italiana. Benucci, Perugia.
Morgan, J. L. (1970). On the criterion of identity for noun phrase deletion. In Papers
from the Sixth Regional Meeting, Chicago Linguistic Society, pages 380–9.
Chicago Linguistic Society.
Morgan, J. L. (1977). Two types of convention in indirect speech acts. Technical
Report 52, University of Illinois at Urbana-Champagne.
Morzycki, M. (2012). Adjectival extremeness: Degree modification and contextu-
ally restricted scales. Natural Language and Linguistic Theory, 30(2):567–609.
Moss, S. (2015). On the semantics and pragmatics of epistemic vocabulary.
Semantics and Pragmatics, 8(5):1–81.
Mulder, W. D. (2010). Mood in French. In Thieroff, R. and Rothstein, B. eds, Mood
in the Languages of Europe, pages 157–78. John Benjamins, Amsterdam and
Philadelphia.
Murray, S. (2011). A Hamblin semantics for evidentials. In Ito, S. and Cormany, E.,
eds, Proceedings from Semantics and Linguistic Theory XIX. CLC Publications,
Ithaca, NY.
Murray, S. and Starr, W. B. (2016). The structure of communicative acts. Manu-
script, Cornell University.
Murray, S. E. (2014). Varieties of update. Semantics and Pragmatics, 7(2):1–53.
Murray, S. E. (2016). Evidentials and illocutionary mood in Cheyenne. Inter-
national Journal of American Linguistics, 82(4):487–517.
Newman, S. (1944). Yokuts Language of California. Viking Fund Publications in
Anthropology, Number 2. Johnson Reprint Corporation, New York.
Nuyts, J. and van der Auwera, J., eds (2016). The Oxford Handbook of Modality and
Mood. Oxford University Press, Oxford.

i i

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

i i

references 267

Obenauer, H.-G. (2004). Nonstandard wh-questions and alternative checkers in

Pagotto. In Lohnstein, H. and Trissler, S., eds, Syntax and Semantics of the Left
Periphery, Interface Explorations 9, pages 343–83. Mouton de Gruyter, Berlin.
Ogihara, T. (1996). Tense, Attitudes, and Scope. Reidel, Dordrecht and Boston.
Ogihara, T. (2007). Attitudes without monsters: A Japanese perspective.
In Tancredi, C., Kanazawa, M., Imani, I., and Kusumoto, K., eds, Proceedings of
SALT 16. CLC Publications, Cornell University Linguistics Department.
Pagin, P. (2014). Assertion. In Zalta, E. N., ed., The Stanford Encyclopedia
of Philosophy. Spring 2014 edition. http://plato.stanford.edu/archives/spr2014/
entries/assertion/.
Pak, M. (2008). Types of clauses and sentence end particles in Korean. Korean
Linguistics, 14:113–55.
Pak, M., Portner, P., and Zanuttini, R. (2008a). Agreement and the subjects of
jussive clauses in Korean. In Elfner, E. and Walkow, M., eds, Proceedings of NELS
37, volume 2, pages 127–38. GLSA, Amherst.
Pak, M., Portner, P., and Zanuttini, R. (2008b). Agreement in promissive, imperat-
ive, and exhortative clauses. Korean Linguistics, 14:157–75.
Pak, M., Portner, P., and Zanuttini, R. (2015). Course notes for “The syntax and
semantics of discourse oriented features”. LSA Summer Institute, University of
Chicago.
Palmer, F. (1986). Mood and Modality. Cambridge University Press, Cambridge.
Palmer, F. (1990). Modality and the English Modals. Longman, New York.
Palmer, F. (2001). Mood and Modality, second edition. Cambridge University Press,
Cambridge.
Parks, D. R. (1976). A Grammar of Pawnee. Garland Publishing.
Pearson, H. (2013). The Sense of Self: Topics in the Semantics of De Se Expressions.
PhD thesis, Harvard University.
Pendlebury, M. (1986). Against the power of force: Reflections on the meaning of
mood. Mind, 95(379):361–72.
Percus, O. and Sauerland, U. (2003). On the LFs of attitude reports. In Sinn und
Bedeutung 7, pages 228–42.
Perry, J. (1977). Frege on demonstratives. Philosophical Review, LXXXVI:
474–97.
Pesetsky, D. (1992). Zero Syntax. MIT Press, Cambridge, MA.
Pica, P. (1984). On the distinction between argumental and non-argumental
anaphors. In de Geest, W. and Putseus, Y., eds, Sentential Complementation,
pages 185–93. Foris, Dordrecht.
Picallo, C. (1984). The infl node and the null subject parameter. Linguistic Inquiry,
15:75–101.
Picallo, C. (1985). Opaque Domains. PhD thesis, CUNY.
Pierrehumbert, J. B. and Hirschberg, J. (1990). The meaning of intonational con-
tours in the interpretation of discourse. In Cohen, P., Morgan, J., and Pollack, M.,
eds, Intentions in Communication, pages 271–311. Bradford Books, MIT Press,
Cambridge, MA.
Platzack, C. (2007). Embedded imperatives. In van der Wurff, W., ed., Imperative
Clauses in Generative Grammar: Studies Offered to Frits Beukema, volume 103

i i

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

i i

268 references

of Linguistik Aktuell/Linguistics Today, pages 181–203. John Benjamins, Amster-

dam and Philadelphia.
Platzack, C. and Rosengren, I. (1994). On the subject of imperatives: A minimalist
account of the imperative pronoun and negated imperatives. Sprache und
Pragmatik, 34:26–67.
Platzack, C. and Rosengren, I. (1998). On the subject of imperatives: A minimalist
account of the imperative clause. The Journal of Comparative Germanic Linguis-
tics, 1(3):177–224.
Portner, P. (1992). Situation Theory and the Semantics of Propositional Expres-
sions. PhD thesis, University of Massachusetts, Amherst.
Portner, P. (1997). The semantics of mood, complementation, and conversational
force. Natural Language Semantics, 5:167–212.
Portner, P. (2004). The semantics of imperatives within a theory of clause types.
In Watanabe, K. and Young, R. B., eds, Proceedings of Semantics and Lin-
guistic Theory 14, pages 235–52. CLC Publications, Cornell University Linguistics
Department.
Portner, P. (2007). Imperatives and modals. Natural Language Semantics, 15(4):
351–83.
Portner, P. (2008). Beyond the common ground: The semantics and pragmatics
of epistemic modals. In Yoon, J.-Y. and Kim, K.-A., eds, The Perspectives of
Linguistics in the 21st Century. Hankook Publishing Company, Seoul.
Portner, P. (2009). Modality. Oxford University Press, Oxford.
Portner, P. (2011a). Perfect and progressive. In von Heusinger, K., Maienborn, C.,
and Portner, P., eds, Semantics: An International Handbook of Natural Language
Meaning, volume 1. Mouton de Gruyter, Berlin.
Portner, P. (2011b). Permission and choice. In Grewendorf, G. and Zimmermann,
T. E., eds, Discourse and Grammar: From Sentence Types to Lexical Categories,
Studies in Generative Grammar. Mouton de Gruyter, Berlin.
Portner, P. (2011c). Verbal mood. In von Heusinger, K., Maienborn, C., and
Portner, P., eds, Semantics: An International Handbook of Natural Language
Meaning, volume 2, pages 1262–91. Mouton de Gruyter, Berlin.
Portner, P. (2016). Imperatives. In Aloni, M. and van Rooij, R., eds, Cambridge
Handbook of Semantics. Cambridge University Press, Cambridge.
Portner, P. (2017). On the relation between verbal mood and sentence mood. Ms.,
Georgetown University. http://semanticsarchive.net/Archive/GFiYjAwZ/.
Portner, P. (to appear). Commitment to priorities. In Fogel, D., Harris, D., and
Moss, M., eds, New Work on Speech Acts. Oxford University Press, Oxford.
Portner, P. and Partee, B. H., eds (2002). Formal Semantics: The Essential Readings.
Blackwell Publishers, Oxford.
Portner, P. and Rubinstein, A. (2013). Mood and contextual commitment.
In Chereches, A., ed., The Proceedings of SALT 22, pages 461–87. CLC Public-
ations, Cornell, NY.
Portner, P. and Rubinstein, A. (2016). Extreme and non-extreme deontic modals. In
Charlow, N. and Chrisman, M., eds, Deontic Modals. Oxford University Press,
Oxford.

i i

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

i i

references 269

Portner, P. and Zanuttini, R. (2000). The force of negation in Wh exclamatives and

interrogatives. In Horn, L. and Kato, Y., eds, Studies in Negation and Polarity:
Syntactic and Semantic Perspectives, pages 201–39. Oxford University Press,
New York and Oxford.
Portner, P. and Zanuttini, R. (2006). The semantics of nominal exclamatives. In
Elugardo, R. and Stainton, R., eds, Ellipsis and Non-sentential Speech. Kluwer
Academic Publishers, Dordrecht and Boston.
Potsdam, E. (1996). Syntactic Issues in English Imperatives. PhD thesis, University
of California at Santa Cruz.
Progovac, L. (1993). The (mis)behavior of anaphora and negative polarity. The
Linguistic Review, 10:37–59.
Quer, J. (1998). Mood at the Interface. Holland Academic Graphics, The Hague.
Quer, J. (2001). Interpreting mood. Probus, 13:81–111.
Quer, J. (2009). Twists of mood: The distribution and interpretation of indicative
and subjunctive. Lingua, 119:1779–87.
Quer, J. (2010). Mood in Catalan. In Thieroff, R. and Rothstein, B. eds, Mood
in the Languages of Europe, pages 221–36. John Benjamins, Amsterdam and
Philadelphia.
Quine, W. V. O. (1956). Quantifiers and propositional attitudes. Journal of Philo-
sophical Logic, 53(5):177–87.
Raposo, E. (1986a). On the null object in European Portuguese. In Jaeggli, O. and
Silva-Corvalan, C., eds, Studies in Romance Linguistics, pages 373–90. Foris,
Dordrecht.
Raposo, E. (1986b). Some asymmetries in the binding theory in Romance. The
Linguistic Review, 5:75–110.
Rawlins, K. (2013). (Un)conditionals. Natural Language Semantics, 21:111–78.
Reis, M. (1999). On sentence types in German: An enquiry into the relationship
between grammar and pragmatics. IJGLSA, 4(2):195–236.
Reis, M. (2003). On the form and interpretation of German wh-infinitives. Journal
of Germanic Linguistics, 15(2):155–201.
Rett, J. (2008a). A degree account of exclamatives. In Rett, J., Friedman, T., and Ito,
S., eds, The Proceedings of SALT 18, pages 601–18.
Rett, J. (2008b). Degree Modification in Natural Language. PhD thesis, Rutgers
University.
Rett, J. (2011). Exclamatives, degrees and speech acts. Linguistics and Philosophy,
34(5):411–42.
Rett, J. (2014). Attitude markers and sincerity conditions in an update semantics.
Ms., UCLA.
Rett, J. and Murray, S. (2013). A semantic account of mirative evidentials. In Snider,
T., ed., The Proceedings of SALT 23, pages 453–72. CLC Publications, Cornell
University Linguistics Department.
Rivero, M. L. (1994). Clause structure and V-movement in the languages of the
Balkans. Natural Language and Linguistic Theory, 12(1):63–120.
Rivero, M. L. and Terzi, A. (1995). Imperatives, V-movement and logical mood.
Journal of Linguistics, 31:301–32.

i i

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

i i

270 references

Rizzi, L. (1997). The fine structure of the left periphery. In Haegeman, L., ed.,
Elements of Grammar: Handbook of Generative Syntax. Kluwer Academic Pub-
lishers, Dordrecht.
Roberts, C. (1996). Information structure in discourse: Towards an integrated
formal theory of pragmatics. In Toon, J.-H. and Kathol, A., eds, Papers in
Semantics, OSU Working Papers in Linguistics, Vol. 49, pages 91–136. Depart-
ment of Linguistics, the Ohio State University, Columbus.
Roberts, C. (2004). Context in dynamic interpretation. In Horn, L. and Ward, G.,
eds, The Handbook of Pragmatics, pages 197–220. Blackwell, Oxford and
Malden, MA.
Roberts, C. (2012). Information structure in discourse: Towards an integrated
formal theory of pragmatics. Semantics and Pragmatics, 5(6):1–69.
Roberts, C. (to appear). Speech acts in discourse context. In Fogal, D., Harris, D.,
and Moss, M., eds, New Work on Speech Acts. Oxford University Press, Oxford.
Roberts, J. R. (1991). Modality in Amele and other Papuan languages. Journal of
Linguistics, 26(2):363–401.
Roelofsen, F. and Farkas, D. (2015). Polarity particle responses as a window onto
the interpretation of questions and assertions. Language, 91(2):359–414.
Rögnvaldsson, E. (1998). The syntax of the imperative in Old Scandinavian. www.
academia.edu/2861283/The_syntax_of_the_imperative_in_Old_Scandinavian.
Rooth, M. (1992). A theory of focus interpretation. Natural Language Semantics,
1(1):75–116.
Rosengren, I., ed. (1992a). Satztyp, Satzmodus, und Illokution, volume 1. Niemeyer,
Tübingen.
Rosengren, I., ed. (1992b). Satztyp, Satzmodus, und Illokution, volume 2. Niemeyer,
Tübingen.
Ross, A. (1944). Imperatives and logic. Philosophy of Science, 11(1):30–46.
Rothschild, D. (2012). Expressing credences. Proceedings of the Aristotelian Society,
112(1):99–114.
Rubinstein, A. (2012). Roots of Modality. PhD thesis, University of Massachusetts,
Amherst.
Rubinstein, A., Harner, H., Krawczyk, E., Simonson, D., Katz, G., and Portner, P.
(2013). Toward fine-grained annotation of modality in text. In Proceedings of
IWCS 2013 Workshop on Annotation of Modal Meanings in Natural Language
(WAMM), pages 38–46, Potsdam, Germany. Association for Computational
Linguistics.
Rupp, L. (2003). The Syntax of Imperatives in English and Germanic: Word Order
Variation in the Minimalist Framework. Palgrave Macmillan, Basingstoke.
Rus, D. (2005). Embedded imperatives in Slovenian. In Brandstetter, C. and
Rus, D., eds, Georgetown University Working Papers in Theoretical Linguistics,
volume IV, pages 153–83. Georgetown University Department of Linguistics,
Washington, DC.
Russell, B. (2007). Imperatives in conditional conjunction. Natural Language
Semantics, 15(2):131–66.
Ruwet, N. (1984). Je veux partir / ∗ je veux que je parte: On the distribution of finite
complements and infinitival complements in French. Cahiers de Grammaire,
7:75–138.

i i

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

i i

references 271

Sadock, J. M. (1974). Toward a Linguistic Theory of Speech Acts. Academic Press,

New York.
Sadock, J. M. (1985). On the performadox, or a semantic defense of the performat-
ive hypothesis. In University of Chicago Working Papers in Linguistics, volume 1,
pages 160–9. Department of Linguistics, University of Chicago.
Sadock, J. M. (2004). Speech acts. In Horn, L. and Ward, G., eds, The Handbook of
Pragmatics, pages 53–73. Blackwell, Oxford.
Sadock, J. M. and Zwicky, A. (1985). Speech act distinctions in syntax. In Shopen,
T., ed., Language Typology and Syntactic Description, pages 155–96. Cambridge
University Press, Cambridge.
Šafářová, M. (2005). The semantics of rising intonation in interrogatives and
declaratives. In Maier, E., Bary, C., and Huitink, J., eds, Proceedings of Sinn und
Bedeutung 9, pages 355–69.
Šafářová, M. (2007). Nuclear rises in update semantics. In Aloni, M., Butler, A., and
Dekker, P., eds, Questions in Dynamic Semantics, pages 355–69. Elsevier, Oxford.
Šafářová, M. and Swerts, M. (2004). On the recognition of declarative questions in
English. In Speech Prosody 2004, pages 313–16.
Schlenker, P. (2003). A plea for monsters. Linguistics and Philosophy, 26:29–120.
Schlenker, P. (2005). The lazy Frenchman’s approach to the subjunctive (specula-
tions on reference to worlds and semantic defaults in the analysis of mood).
In Geerts, T., van Ginneken, I., and Jacobs, H., eds, Romance Languages and
Linguistic Theory 2003: Selected Papers from “Going Romance”, pages 269–310.
John Benjamins, Amsterdam.
Schlenker, P. (2010). Presuppositions and local contexts. Mind, 119(474):377–91.
Schlenker, P. (2011). Indexicality and de se reports. In von Heusinger, K., Maien-
born, C., and Portner, P., eds, Semantics: An International Handbook of Natural
Language Meaning, volume 2. de Gruyter, Berlin.
Schmerling, S. (1982). How imperatives are special, and how they aren’t. In
Schneider, R., Tuite, K., and Chameltzy, R., eds, Papers from the Parasession on
Nondeclaratives. Chicago Linguistics Society, The University of Chicago.
Searle, J. R. (1965). What is a speech act? In Black, M., ed., Philosophy in America,
pages 221–239. Unwin Hyman, London.
Searle, J. R. (1969). Speech Acts. Cambridge University Press, Cambridge.
Searle, J. R. (1975a). Indirect speech acts. In Cole, P. and Morgan, J. L., eds, Syntax
and Semantics, volume 3: Speech Acts, pages 59–82. Academic Press, New York.
Searle, J. R. (1975b). A taxonomy of illocutionary acts. In Gunderson, K., ed., Lan-
guage, Mind and Knowledge, Minnesota Studies in the Philosophy of Science,
Vol. VII, pages 344–69. University of Minnesota Press, Minneapolis.
Searle, J. R. (1989). How performatives work. Linguistics and Philosophy, 12(5):
535–58.
Searle, J. R. (2001). Réponses de Searle. Revue internationale de philosophie,
2(216):277–97.
Searle, J. R. (2010). http://youtu.be/-9bxhqiosni.
Searle, J. R. and Vanderveken, D. (1985). Foundations of Illocutionary Logic. Cam-
bridge University Press, Cambridge.
Segerberg, K. (1990). Validity and satisfaction in imperative logic. Notre Dame
Journal of Formal Logic, 31(2):203–21.

i i

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

i i

272 references

Sheppard, M. M. and Golden, M. (2002). (Negative) imperatives in Slovene.

In Barbiers, S., Beukema, F., and van der Wurff, W., eds, Modality and its
Interaction with the Verbal System, volume 47 of Linguistik Aktuell/Linguistics
Today, pages 245–59. John Benjamins Publishing Company, Amsterdam and
Philadelphia.
Siegel, L. (2009). Mood selection in Romance and Balkan. Lingua, 119(12):1859–82.
Simons, M. (2005). Dividing things up: The semantics of or and the modal/Or
interaction. Natural Language Semantics, 13(3):271–316.
Smirnova, A. (2011). Evidentiality and Mood: Grammatical Expressions of Epi-
stemic Modality in Bulgarian. PhD thesis, The Ohio State University.
Smirnova, A. (2012). The semantics of mood in Bulgarian. In Proceedings of the
Chicago Linguistics Society 48. Chicago Linguistics Society.
Soames, S. (1982). How presuppositions are inherited: A solution to the projection
problem. Linguistic Inquiry, 13:483–545.
Sperber, D. and Wilson, D. (1986). Relevance: Communication and Cognition.
Blackwell, Oxford.
Squartini, M. (2004). Disentangling evidentiality and epistemic modality in
Romance. Lingua, 114(7):873–95.
Stalnaker, R. (1970). Pragmatics. Synthese, 22:272–89.
Stalnaker, R. (1974). Pragmatic presuppositions. In Munitz, M. and Unger, P., eds,
Semantics and Philosophy, pages 197–213. New York University Press, New York.
Stalnaker, R. (1975). Indicative conditionals. Philosophia, 5:269–86.
Stalnaker, R. (1978). Assertion. In Cole, P., ed., Syntax and Semantics 9: Pragmatics,
pages 315–32. Academic Press, New York.
Stalnaker, R. (1984). Inquiry. MIT Press, Cambridge, MA.
Stalnaker, R. (1988). Belief attribution and context. In Grimm, R. H. and Merrill,
D. D., eds, Contents of Thought, pages 140–56. University of Arizona Press,
Tucson. Reprinted in: R. C. Stalnaker. Context and Content. Oxford: Oxford
University Press, 1999, 150–66.
Stalnaker, R. (2014). Context. Oxford University Press, Oxford and New York.
Stalnaker, R. (to appear). Dynamic pragmatics, static semantics. In Fogel, D.,
Harris, D., and Moss, M., editors, New Work on Speech Acts. Oxford University
Press, Oxford.
Starr, W. B. (2010). Conditionals, Meaning, and Mood. PhD thesis, Rutgers
University.
Starr, W. B. (2013). A preference semantics for imperatives. Ms., Cornell University.
Starr, W. (2016). Expressing permission. Semantics and Linguistic Theory, 26:
325–49.
Steele, S. (1975). Past and irrealis: Just what does it all mean. International Journal
of American Linguistics, 41:200–17.
Stenius, E. (1967). Mood and language game. Synthese, 17:254–74.
Stowell, T. (1982). The tense of infinitives. Linguistic Inquiry, 13(3):561–70.
Suñer, M. (1986). On the referential properties of embedded finite clause subjects.
In Bordelois, I., Conteres, H., and Zagona, K., eds, Generative Studies in Spanish
Syntax, pages 183–203. Foris, Dordrecht.

i i

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

i i

references 273

Suñer, M. and Padilla-Rivera, J. (1987). Sequence of tenses and the subjunctive,

again. Hispania, 70(3):634–42.
Swanson, E. (2007). The language of subjective uncertainty. Ms., University of
Michigan. Revised version of Ch. 2 of his 2006 MIT dissertation.
Swanson, E. (2011a). How not to theorize about the language of subjective uncer-
tainty. In Egan, A. and Weatherson, B., eds, Epistemic Modality, pages 249–69.
Oxford University Press, Oxford.
Swanson, E. (2011b). Propositional atitudes. In von Heusinger, K., Maienborn, C.,
and Portner, P., eds, Semantics: An International Handbook of Natural Language
Meaning, volume 2. Mouton de Gruyter, Berlin.
Swanson, E. (2015). The application of constraint semantics to the language of
subjective uncertainty. Journal of Philosophical Logic, pages 1–26.
Terrell, T. and Hooper, J. (1974). A semantically based analysis of mood in Spanish.
Hispania, 57(3):486–94.
Thieroff, R. (2010). Moods, Moods, Moods: Mood in the Languages of Europe,
pages 1–29. In Thieroff, R. and Rothstein, B., eds, John Benjamins, Amsterdam
and Philadelphia.
Thieroff, R. and Rothstein, B., eds (2010). Mood in the Languages of Europe. John
Benjamins, Amsterdam and Philadelphia.
Thomas, G. (2014). Embedded imperatives in Mbyà. In Huang, H.-L., Poole, E.,
and Rysling, A., eds, Proceedings of NELS 43, Amherst. GLSA, University of
Massachusetts.
Thomason, R. H. (1981a). Deontic logic and the role of freedom in moral delibera-
tion. In Hilpinen, R., ed., New Studies in Deontic Logic, pages 177–86. D. Reidel
Publishing Co., Dordrecht.
Thomason, R. H. (1981b). Deontic logic as founded on tense logic. In Hilpinen,
R., ed., New Studies in Deontic Logic, pages 165–76. D. Reidel Publishing Co.,
Dordrecht.
Trinh, T. and Crnič, L. (2011). The rise and fall of declaratives. In Reich, I., ed.,
Proceedings of Sinn und Bedeutung 15, pages 645–60. Saarland University Press,
Saarbrücken.
Truckenbrodt, H. (2006a). On the semantic motivation of syntactic verb move-
ment to C in German. Theoretical Linguistics, 32(3):257–306.
Truckenbrodt, H. (2006b). Replies to the comments by Gärtner, Plunze and
Zimmermann, Portner, Potts, Reis, and Zaefferer. Theoretical Linguistics, 32(3):
257–306.
Tsoulas, G. (1996). The nature of the subjunctive and the formal grammar of
obviation. In Zagona, K., ed., Linguistic Theory and Romance Languages,
pages 295–306. John Benjamins, Amsterdam/Philadelphia.
van Benthem, J. and Liu, F. (2004). Dynamic logic of preference upgrade. Journal
of Applied Non-Classical Logics, 14(2):1–26.
van der Auwera, J. and Devos, M. (2012). Irrealis in positive imperatives and in
prohibitives. Language Sciences, 34(2):171–83.
van der Torre, L. W. N. and Tan, Y.-H. (1998). An update semantics for deontic
reasoning. In Proceedings of DEON’98, pages 409–26.

i i

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

i i

274 references

van Fraassen, B. C. (1973). Values and the heart’s command. Journal of Philosophy,
70(1):5–19.
van Rooij, R. (2000). Permission to change. Journal of Semantics, 17:119–45.
Van Valin, R. D. and LaPolla, R. J. (1997). Syntax: Structure, Meaning and Function.
Cambridge University Press, Cambridge.
Vanderveken, D. (1990). Meaning and Speech Acts, Volume 1: Principles of Language
Use. Cambridge University Press, Cambridge.
Vanderveken, D. (1991). Meaning and Speech Acts, Volume 2: Formal Semantics of
Success and Satisfaction. Cambridge University Press, Cambridge.
Vanderveken, D. (2002). Universal grammar and speech act theory.
In Vanderveken, D. and Kuno, S., eds, Essays in Speech Act Theory, pages 25–62.
John Benjamins, Amsterdam.
Veltman, F. (1986). Data semantics and the pragmatics of indicative condition-
als. In Traugott, E., ter Meulen, A., and Snitzer, A., eds, On Conditionals,
pages 147–68. Cambridge University Press, Cambridge.
Veltman, F. (1996). Defaults in update semantics. Journal of Philosophical Logic,
25:221–61.
Villalta, E. (2000). Spanish subjunctive clauses require ordered alternatives. In
Jackson, B. and Matthews, T., eds, The Proceedings of SALT 10, pages 239–56.
CLC Publications, Ithaca, NY.
Villalta, E. (2006). Context Dependence in the Interpretation of Questions and
Subjunctives. PhD thesis, Tübingen.
Villalta, E. (2008). Mood and gradability: An investigation of the subjunctive mood
in Spanish. Linguistics and Philosophy, 31(4):467–522.
von Stechow, A. (1995). On the proper treatment of tense. In Simons, M. and
Galloway, T., eds, The Proceedings of SALT 5, pages 362–86. Cornell University,
Ithaca, NY.
von Stechow, A. (2003). Deletion under semantic binding. Ms., University of
Tübingen; text of NELS 33 talk.
von Stechow, A. (2004). Binding by verbs: Tense, person and mood under attitudes.
In Lohnstein, H. and Trissler, S., eds, The Syntax and Semantics of the Left
Periphery, pages 431–88. Mouton de Gruyter, Berlin and New York.
Vranas, P. (2008). New foundations for imperative logic i: Logical connectives,
consistency, and quantifiers. Noûs, 42:529–72.
Vranas, P. (2011). New foundations for imperative logic: Pure imperative inference.
Mind, 120:369–446.
Wechsler, S. (1991). Verb second and illocutionary force. In Leffel, K. and Bouchard,
D., eds, Views on Phrase Structure, pages 177–91. Kluwer, Dordrecht.
Whitehead, C. R. (1986). Tense, Aspect, Mood and Modality: Verbal Morphology
in Menya. Master’s thesis, University of Manitoba.
Wiklund, A.-L., Bentzen, K., Hrafnbjargarson, G. H., and Hróarsdóttir, T. (2009).
On the distribution and illocution of v2 in Scandinavian that-clauses. Lingua,
119(12):1914–38.
Willer, M. (2014). Dynamic thoughts on ifs and oughts. Philosophers’ Imprint,
14(28):1–30.

i i

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

i i

references 275

Wilson, D. and Sperber, D. (1988). Mood and the analysis of non-declarative sen-
tences. In Dancy, J., Moravcsik, J. and Taylor, C., eds, Human Agency: Language,
Duty and Value, pages 77–101. Stanford University Press, Stanford, CA.
Wiltschko, M. (2016). The essence of a category: Lessons from the subjunctive.
In Błaszczak, J., Giannakidou, A., Klimek-Jankowska, D., and Migdalski,
K., eds, Mood, Aspect, Modality Revisited: New Answers to Old Questions,
pages 218–54. University of Chicago Press, Chicago, IL.
Wittgenstein, L. (1953). Philosophical Investigations. G. E. M. Anscombe and R.
Rhees (eds) and translated by G. E. M. Anscombe. Blackwell, Oxford.
Yalcin, S. (2007). Epistemic modals. Mind, 116(464):983–1026.
Yalcin, S. (2010). Probability operators. Philosophy Compass, 5(11):916–37.
Yalcin, S. (2011). Nonfactualism about epistemic modality. In Egan, A. and
Weatherson, B., eds, Epistemic Modality. Oxford University Press, Oxford.
Yalcin, S. (2012). Context probabilism. In Aloni, M., Kimmelman, V., Roelofsen, F.,
Sassoon, G., Schulz, K., and Westera, M., eds, Proceedings of the 18th Amsterdam
Colloquium, pages 12–21.
Yamada, T. (2007a). Acts of commanding and changing obligations. In Inoue, K.,
Satoh, K., and Toni, F., eds, Computational Logic in Multi-Agent Systems, 7th
International Workshop, CLIMA VII, Hakodate, Japan, volume 4371 of Lecture
Notes in Artificial Intelligence, pages 1–19. Springer, Berlin.
Yamada, T. (2007b). Logical dynamics of commands and obligations. In Takashi-
Washio, Sato, K., Takeda, H., and Inokuchi, A., eds, New Frontiers in Artificial In-
telligence, JSAI 2006 Conference and Workshops, Tokyo, Japan, June 2006, volume
4384 of Lecture Notes in Artificial Intelligence, pages 133–46. Springer, Berlin.
Yamada, T. (2007c). Logical dynamics of some speech acts that affect obligations
and preferences. In van Benthem, J., Ju, S., and Veltman, F., eds, A Meeting
of Minds: Proceedings of the Workshop on Logic, Rationality and Interaction,
Beijing, 2007, volume 8 of Texts in Computer Science, pages 275–89. King’s
College Publications, London.
Zaefferer, D. (1981). On a formal treatment of illocutionary force indicators. In
Parret, H., Sbisà, M., and Verschueren, J., eds, Possibilities and Limitations of
Pragmatics: Proceedings of the Conference on Pragmatics, Urbino, July 8–14,
1979, pages 779–98. John Benjamins, Amsterdam.
Zaefferer, D. (1983). The semantics of sentence mood in typologically differing
languages. In Hattori, S. and Inoue, K., eds, Proceedings of the XIIIth
International Congress of Linguists, pages 553–7. CIPL, Tokyo.
Zaefferer, D. (1986). The grammar of clause type and the pragmatics of illocution
type. In Proceedings of the Twenty-Second Regional Meeting of the Chicago
Linguistic Society, volume 2, pages 29–39. University of Chicago, Chicago, IL.
Zaefferer, D. (1998). On a formal treatment of illocutionary force indicators. In
Kasher, A., ed., Pragmatics: Critical Concepts, Volume II: Speech Act Theory and
Particular Speech Acts, pages 250–67. Routledge, London.
Zaefferer, D. (2001). Deconstructing a classical classification: A typological look
at Searle’s concept of illocution type. Revue internationale de philosophie,
2(216):209–25.

i i

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

i i

276 references

Zaefferer, D. (2006a). Conceptualizing sentence mood—two decades later.

In Brandt, P. and Fuß, E., eds, Form, Structure, and Grammar: A Festschrift
Presented to Günther Grewendorf on Occasion of his 60th Birthday, Studia
Grammatica 67, pages 367–82. Akademie, Berlin.
Zaefferer, D. (2006b). Types, moods, and force potentials: Towards a comprehens-
ive account of German sentence mood meanings. Theoretical Linguistics, 32(3):
335–52.
Zaefferer, D. (2007). Deskewing the Searlean picture: A new speech act ontology
for linguistics. In The Proceedings of BLS 32. Berkeley Linguistics Society.
Zanuttini, R. (1997). Negation and Clausal Structure: A Comparative Study of
Romance Languages. Oxford University Press, New York and Oxford.
Zanuttini, R. (2008). Encoding the addressee in the syntax: Evidence from English
imperative subjects. Natural Language and Linguistic Theory, 26(1):185–218.
Zanuttini, R. and Portner, P. (2000). The characterization of exclamative clauses
in Paduan. Language, 76(1):123–32.
Zanuttini, R. and Portner, P. (2003). Exclamative clauses: At the syntax–semantics
interface. Language, 79(1):39–81.
Zanuttini, R., Pak, M., and Portner, P. (2012). A syntactic analysis of interpretive
restrictions on imperative, promissive, and exhortative subjects. Natural
Language and Linguistic Theory, 30(4):1231–74.
Žarnić, B. (2003a). Imperative change and obligation to do. In Segerberg, K. and
Sliwinski, R., eds, Logic, Law, Morality: Thirteen Essays in Practical Philosophy
in Honour of Lennart Åqvist, volume 51 of Uppsala Philosophical Studies,
pages 79–95. Department of Philosophy, Uppsala University.
Žarnić, B. (2003b). Imperative logic, moods and sentence radicals. In Dekker, P.
and van Rooij, R., eds, Proceedings of the Fourteenth Amsterdam Colloquium,
pages 223–8. ILLC, Amsterdam.
Zeijlstra, H. (2008). Modal concord is syntactic agreement. In Gibson, M. and
Friedman, T., eds, The Proceedings of SALT XVII. CLC Publications, Cornell
University Linguistics Department.
Zimmermann, T. E. (2000). Free choice disjunction and epistemic possibility.
Natural Language Semantics, 8:255–90.
Zu, V. (2016). Competition and obviation from French to Newari. In Proceedings
of NELS 46. GLSA, University of Massachusetts.

i i

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

i i

Index
Note: The most important page references are indicated in bold type.

ability modality (ability modal) 11 185–7, 189, 198–200, 205–7, 214,

accessibility 14–17, 19, 32–3, 50–2, 54–5, 220–1, 223–32
59–60, 78–80, 82, 92, 94, 96–100, command 73, 143, 171–4, 202, 210, 216
103–4, 112, 202 commissive 41, 72, 74, 81, 94, 138, 162–3,
future-possibility accessibility 98 167, 225
historical accessibility 202 commissive point 41, 138, 162–3
realistic accessibility 100 commitment slate 44, 176, 178–9, 183–7,
totally realistic accessibility 94 191–2, 211–12, 235–6, 239
actions 34, 41, 129, 155, 199, 201–2, 206–7, common ground 22–3, 25, 33–4, 78, 83–6,
214, 215–17, 219, 227 147, 155, 170–1, 174–5, 178–81,
adverbial clauses 69, 112–13, 244–5 183–4, 186–8, 193–5, 204, 207, 213–14,
Amele 243–6 227, 240, 242, 250
anchor (intensional, extensional) 95–9, 103 comparison 68, 70–1, 80–93, 96, 98, 103–7,
asking 35–6, 40–1, 45, 121, 123, 129–30, 233–5, 241–3
138–40, 146, 151–2, 162, 177, 181, 185–7, comparison-based approach 70, 80–93,
189–99, 215, 224, 228, 237 96, 106, 235, 242
assertion 21, 23–7, 32–4, 36, 39, 42–3, 45, complement clause (argument clause) 5, 8,
72–5, 78, 81, 85, 89, 94, 98, 105, 114, 37, 48, 53, 55–6, 59–61, 63, 65, 69–70,
137–40, 146, 148–52, 154–5, 158, 162–3, 71–80, 83–4, 87–8, 90, 92–3, 96–7,
165–6, 169–72, 175, 177, 179–81, 183–9, 99–100, 102–3, 107, 109, 117, 119, 141,
191–8, 202–3, 207, 212, 214, 224, 226, 234, 236, 241
236–7, 239, 248 compositional approach 40, 140–1,
146–7, 152–7, 169, 173, 182, 185, 214–17,
basic types (basic clause types, major 221–3
clause types) 127, 129, 134, 138–9, 143, conditional 20, 54–5, 83, 86–7, 96, 107,
155, 160, 162, 168, 177, 214, 221, 223–4, 112–13, 114, 117, 129, 200, 209, 215
226–9 see also adverbial clauses, if clause
belief predicates 55, 73, 94 conjunction 42
buletic modality (buletic modal) 11, 52, 61, constative 34–5, 37, 201
82, 89, 108 construction 43, 127, 140–1, 148–52, 157,
buletic accessibility 52, 103 167, 169, 182, 185, 221, 236
buletic individual model 108 construction-based approach 140–1,
buletic ordering source 61, 82, 89 148–52, 157, 167, 182, 221, 236
context 7, 14–16, 18–27, 29–35, 37–9, 44–5,
causative 72, 75, 82, 86–7 55, 70–1, 78–80, 82–9, 91–8, 100–7,
Cheyenne 249 110–11, 140, 143–4, 146–7, 152, 154–6,
circumstantial modality (circumstantial 161, 165, 169–89, 192–8, 202–6, 208–15,
modal) 11, 106 217–22, 227–8, 232–3, 235–7, 239–42,
circumstantial accessibility 106 245, 249–50
circumstantial modal 11 common ground of the context
clause type 34, 38, 47, 93, 121–36, (common ground), see common
137–45, 147–53, 155–7, 160–8, 177, ground

i i

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

i i

278 index

context (cont.) deontic modality (deontic modal) 7, 11–12,

context change potential (CCP) 27–30, 15, 19, 81, 83, 105, 172–3, 202–3, 208–12,
32, 105, 147, 152, 178, 180–2, 192, 216 233, 240
context index 146–7 deontic accessibility 15
context of deliberation 209–10, 215, deontic conversational background 83
240–1 deontic ordering 81, 210, 240
context of evaluation (domain of deontic state 208–12
evaluation) 82–7, 96 derived context 79–80, 104–6, 232
context of justification 209, 215, 240–1 method of derived contexts 79, 104, 232
context parameter 94–6 desire predicates 53, 55, 87, 90, 104–5
context set 22–5, 32, 79–80, 84–6, 91, desideratives 7, 90–1, 115–16
102–3, 105, 156, 170, 173, 175–6, diagonal proposition 94–6
183–4, 202, 204, 212–13, 215, direction of fit 41–2
249–50 directives 41, 72–4, 81, 90–1, 115–16, 134–5,
context situation 14–16, 18 138, 145–6, 154–6, 162, 167, 201, 206–7,
contextual commitment 107, 232 216, 223–5
credal context 188, see also information directive predicates 72, 74, 81, 90, 116
state discourse modality (discourse
derived context, see derived context modal) 8–10, 13, 49, 231, 243
discourse context 7, 21, 32, 104, 140, 144, discourse referent 25–31, 33, 174, 179,
155–6, 175, 177, 180–6, 192–3, 195–7, 183, 250
204–6, 212, 214, 220, 227–8, 232–3, doxastic modality (doxastic
236–7, 239, 249–50 modal) 51, 54–5, 59, 61, 67, 79–80, 82,
non-defective context 22–3, 26 84, 92, 97–8, 102, 105–6, 108, 112
structured (discourse) context 180–6, doxastic accessibility 51, 54–5, 59, 67,
193, 195, 197, 204, 212, 220, 227–8, 79–80, 92, 97–8, 102, 105–6, 108, 112
232–3 doxastic conversational background 83
control construction 60, 115–16, 246 doxastic modal base 61, 82, 84
conversational background 18–20, 82–3, DRT (Discourse Representation
91, 156, 205 Theory) 28–9, 32, 155
core mood 7, 20, 47, 114, 232–3, 243, dynamic approach 21, 27–9, 33–4, 40, 44,
247, 250 46–7, 79, 121, 124, 136, 138–9, 143, 152,
CP 51, 100, 141, 147–9, 156, 222 158, 169, 171, 174, 176–85, 188–9, 197,
credence 61, 188 199, 201–7, 211, 215, 218, 220, 223,
Cuzco Quechua 9, 247–9 225–8, 231–3, 247
dynamic force hypothesis 44–6, 138, 160,
de re 55–61 169–77, 184–5, 201, 226, 228, 232–3
de se 48, 55–61, 94, 96, 101, 116–18, dynamic logic 21, 29–33, 206–7, 240
241–3, 246 dynamic modality (dynamic modal) 11
decidedness 106, 232 dynamic pragmatics 27–8, 44, 124, 155, 158,
declarative 1, 3, 5, 7, 9, 35–6, 38–9, 41, 43–4, 175, 181–2, 188, 194, 197, 202, 213,
47, 72, 78–9, 84–5, 104, 123, 126–7, 215, 223
129–30, 133–9, 141–5, 147, 150–6, 158–9, dynamic semantics 13, 21–4, 27–30, 32–3,
160, 162–8, 171–3, 180–8, 196–202, 204, 44, 46, 78–9, 85, 121, 123, 143, 158, 178,
212, 214–16, 223–30, 235–40, 242, 180–3, 185, 188, 195–9, 206–8, 210–12,
249–50 215, 218, 221, 237, 247, 250
declarative point (declaration) 41, 43,
137–8, 162, 166, 226 embedded clause 15, 47, 74, 78, 92, 98, 101,
root declarative 9, 78, 84–5, 104, 113, 133, 112, 115, 117, 123, 134, 144, 147–8, 156,
141–2, 147, 237, 239 230, 235–6, 238–9

i i

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

i i

index 279

embedded imperative 231 future tense 93, 135, 223, 244–5

emotive factives (factive emotives) 73, 75,
81, 86, 93, 106 German 50, 74–6, 85, 93, 101–2, 113–14, 126,
epistemic modality (epistemic modal) 10, 142, 146–8, 156, 220, 230
14–15, 29, 32, 91–2, 247–8
epistemic accessibility 32 hortative 127
evaluative 70, 72–3, 86, 105–6
evaluative factives 86 if clause 69, 107, 112–13
evaluative meaning 73 see also adverbial clauses, conditional
evidentiality (evidential) 7, 9–10, 34, 47, illocutionary force 35–6, 39–46, 123–4,
130, 231–2, 242–3, 247–51 136–9, 143–5, 147, 149, 151–2, 156–8,
exclamation 125, 150, 164–6, 226 160–9, 171, 173–4, 177, 202, 225–7, 232,
exclamative 125–7, 129–31, 138, 150–2, 162, 239, 245, 247–50
164–6, 220–3, 224, 228 illocutionary point 40–2, 123, 138, 161–3,
exhortative 127–9, 167, 199, 220, 224, 228, 166, 226, 249
see also hortative imperative 1, 5–7, 10, 34–5, 38, 46–7, 72–3,
expressive 41, 61, 65, 138, 162, 165, 167 95, 105, 114, 123, 126–9, 132–5, 137–9,
142–7, 149–60, 162, 167–9, 171–4, 177,
factivity 52, 82, 87, 101, 106, 222–3 180–2, 185–6, 189, 199–208, 210–20,
factive predicates 72–5, 81 223–31, 233, 235–42, 244–6
factive emotives, see emotive factives implicature 39, 218, 224, 227
features 36, 114, 121, 124, 126, 129, 131, 133, indexical 48, 94–5, 101, 154, 202
141–2, 144–5, 151–3, 156, 159–60, 162, shiftable indexicality 101–3
180, 187, 193, 200, 202, 204, 206–7, 215, indicative 1–2, 4–7, 36–9, 42–3, 47–8,
223, 227, 229, 231–2, 242–3, 245 68–75, 77–8, 80–7, 90–1, 93–110,
felicity condition 23–4, 27, 36, 45, 138, 158, 112–14, 117–19, 133, 142, 147, 154, 156,
165, 170 192, 223–4, 230, 232, 235–9, 241–2
File Change Semantics 21, 23–7, 28–33, indirect speech act 37–40, 43, 136
181, 183 individual model 96–8, 103, 108
focus 87, 89–91, 194 infinitive 4, 7–8, 47, 49, 60, 74, 102, 114–18,
force 15, 35–47, 99–100, 112, 121–31, 136–62, 133, 145, 154, 156, 207, 223, 229, 238,
164–71, 173–4, 176–7, 180–8, 194–7, 241, 246, see also future;
199, 201–2, 204–7, 210–12, 214–18, future-orientation of infinitives
221–33, 236–9, 245, 247–8 information state 30–3, 66–7, 178, 184, 188,
force marker 42–3, 159, 212, 229 208
illocutionary force, see illocutionary credal information state 66–7, 188,
force see also context; credal context
modal force 16–17, 50–1, 53, 71, 78, 86, information structure 130, 193–4
99–100, 112 inquisitive semantics 157, 181, 185, 187,
sentential force, see sentential force 197–9, 218, 250, see also dynamic
formality 114, 134, 202 semantics
free choice 200, 205–6, 212, 216–18 interrogative 1, 5–7, 9, 34–9, 44, 47, 72, 110,
French 4–5, 74–7, 82–3, 85–6, 90, 95, 101–2, 113, 123–7, 129–36, 138–9, 142–5, 147,
106, 111, 113 150–4, 156, 158–9, 162–4, 166–9, 173,
future 5–6, 9–10, 41, 93, 97–8, 115–17, 135, 177, 180–2, 185–7, 189–90, 192, 194–5,
176, 199, 223, 244–5, 250 197–200, 204, 214–15, 223–6, 229–31,
future-orientation of imperatives 199 236–8, 240, 242, 250
future-orientation of infinitives 115–17 intonation 130, 135–6, 140, 161, 164,
future-orientation of subjunctive 115–16 186–7, 198
future possibility relation 97–8 irrealis 7, 70, 93, 114, 145–6, 207, 243–7

i i

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

i i

280 index

Italian 9, 70, 73, 75–6, 81–7, 91, 100–1, 113, obligation 207–9, 216
133–4, 145, 149, 222, 229, 237 obviation effect 102, 115–16, 118, 246
operator 9, 19, 32–3, 50, 64, 66, 85, 87, 90,
judgment type 12, 13–15, 17, 32–3, 48–9 96–7, 100–2, 108–11, 117–18, 140–7,
jussives 127, 129, 220 151–3, 156–7, 159–60, 164, 169, 174–7,
182, 185, 187, 198, 207–10, 212, 215,
Korean 123, 133–4 221–2, 230, 234–6
assertion operator 33, 175, 185, 187
language game 38, 201–2 diagonalization operator 96, 101–2
literal meaning hypothesis (literal force E-force 164–6
hypothesis) 36–7, 39, 45, 123, 139, 145, focus sensitive operator 90, see also
162, 227 focus
locutionary act 35–6, 170 force-indicating operator (force
logical form 2, 5, 26, 29, 35, 36–8, 42, 44, operator) 143–7, 153, 156, 164–6, 174,
57–9, 102, 108, 118–19, 140, 142–3, 147, 176, 187, 207, 230
148, 152, 157–8, 160, 166, 177, 202, 204, modal operator 19, 32–3, 50, 64, 85, 100,
210–11, 217, 235 110, 118, 151–2, 209, 235, see also
Logical Form 2, 24, 28, 135, 158, 198 modality (modal)
non-veridical operator, see veridicality
main clause phenomena 147–8 probability operator 64, 66
major clause types, see basic types veridical operator, see veridicality
metaphysical modality (metaphysical operator-based approach (operator
modal) 10 approach) 140, 142–8, 152, 177, 182,
method of shifting parameters 79, 91, 97, 185, 236
102, 110, 112, 119, 232 opportunity modality (opportunity
minor types (minor clause types) 127–8, modal) 11
130, 134, 162, 189, 199, 220–1, optative 96, 114, 117, 127, 129, 220–1, 224,
226–8 228, 241
modal base 18–19, 61, 70–1, 78–9, 82–6, 89, ordering relation 55, 61–2, 105–6, 211–12,
104, 107, 234, 240 214–15, 219, 233, 240, 242
non-realistic modal base 83–4 ordering semantics 13, 17–20, 50, 55,
(weakly) realistic modal base 83–4, 89 62–5, 68, 70–1, 78, 80, 82–3,
modal concord 118–19 104, 205
modal logic 13–15, 17, 19, 32–3, 47, 49–52, ordering source 18–20, 55, 64, 68, 78–80,
72, 78–9, 200–1 82–7, 89, 104, 107, 233–4, 240
modality (modal) 4, 7–18, 20–1, 29, 32–3,
47–50, 52, 55–6, 61–4, 68–9, 98, 112, partition 156, 170–1, 173, 190–1, 195–8,
118–19, 130, 188, 205, 231, 243, 247, 237, 250
249, 251 perception predicates 72, 74, 82
mood-indicating modal 118, 124 performative 9, 34–5, 37–43, 46, 136–8,
mood marker 159, 249 151–2, 157–62, 171–2, 174, 201, 203, 222,
225–6
necessity 10, 12–13, 16, 48, 50, 64, 89, 99, explicit performative 37–43, 136–8,
112, 144, 203, 205, 234 157–60, 162, 174, 225–6
necessity modal 12–13, 16, 89, 99, 205 performative hypothesis 151–2,
negation 63, 72, 81, 100, 110–12, 116, 128, 158–60
130, 142, 149–50, 207, 212 performative verbs 37, 43, 137, 159–61
non-defective context, see context perlocutionary act 35
(non-defective context) permission 139, 155, 173, 186, 200, 204–6,
non-veridical 97, 111 210–12, 216, 217–20, 240

i i

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

i i

index 281

person (first person, second person, third realis 7, 70, 93, 128, 243–7
person) 101, 114, 129, 133, 154, 171, realistic 17, 82–7, 89, 94, 100
229, 246 realistic modal base 83–4, 89
politeness 114, 133, 200, 231 reality status 7, 47, 69, 231–2, 242–7,
polite imperative 133–4, 149 251
possibility 10, 12, 27, 30–1, 33, 49–50, 68, relative clause 69, 107–10
71, 82, 89, 93–4, 98, 128, 204 Relevance Theory 155
possible worlds 3, 13, 16, 18, 21–2, 25–6, 30, reportative evidential 247–9
33, 47, 49–50, 52–3, 55, 66–7, 70, 79, reportative subjunctive 74
91–3, 95, 99, 176, 185, 188, 190–1, 196, retraction 183–5, 191–3, 212, 215, 219, 240
205, 208–9, 211–12, 215, 237, 241, 249 root clause 4, 47, 69, 78, 102, 107, 113–14,
POSW (partially ordered set of worlds) 7, 134, 142, 147–8, 151–2, 158, 177, 224,
20, 233–42, 250 230, 236
premise set 240, 242
preparatory condition 41–2, 161 sentence mood 1, 5–7, 9, 20–1, 29, 34, 37–8,
presupposition 21–7, 32–3, 41–2, 52, 55, 80, 40, 44–5, 47, 80, 93, 113–14, 121–5,
88, 102, 172–4, 178–80, 202–3, 208–9, 127–31, 135–43, 149–50, 153, 156–8,
214, 249–50 160–9, 171, 173–4, 177–84, 186, 188–9,
speaker presupposition 21, 25, 27 192, 194, 196–7, 199, 201, 207, 210–12,
priority modality (priority modal) 10, 214–17, 220–5, 229–33, 235–43, 245,
48, 119 247, 249–51
probability-based approach 61–8, 187–8 sentence radical 38, 142, 207, 212–13
prohibitive 127–8, 149 sentence type 34, 123–8, 130–1, 134,
promissive 127, 128–9, 138, 162–3, 167, 199, 144, 148, 152, 154, 160–2, 169, 174,
220, 224–6, 228 221, 224
property 8, 59–61, 117, 125, 154, 157, 165–7, sentential force 47, 121–36, 137–52, 154–8,
172, 181, 214, 229 161–2, 164, 167–8, 173–4, 176–7, 180–6,
proposition 3, 6, 18, 22, 34, 44–5, 54, 59, 61, 188, 194–7, 199, 207, 210, 214–16, 218,
64–8, 83–5, 93, 117, 152–6, 170–1, 221–8, 231–2, 236–9
173, 176, 179–80, 183–5, 189–90, 196, sentential modality (sentential
198–9, 207, 211–14, 216, 218–19, 233, modal) 8–10, 13, 49–50, 52, 55, 67
240, 242 sincerity condition 42, 161, 165, 248–9
propositional attitude 8, 50, 52–3, 57–8, 61, Spanish 50, 74, 76–7, 87, 145–6, 149, 207
71–2, 80, 91, 95, 99, 168 speech act theory 21, 34–40, 43–7, 121,
propositional content 36–7, 40–4, 57, 135, 123–4, 136–9, 142–3, 157–64, 166–9,
137–8, 143, 157, 160–2, 164–7, 169–70, 171, 176–7, 182, 185, 189, 201, 223,
172–5, 177, 186, 207, 210–12, 248 225–8, 231–2, 247–9
sphere of accessibility 202
question 34, 36, 39–40, 121, 123, 129–30, sphere of permissibility 202
132, 139, 149, 151–2, 155–6, 162, 166, strength (of an illocutionary act) 41–2, 138,
170–1, 173–4, 179–81, 184–7, 189–99, 161, 248
204, 212, 214, 220, 223–4, 227–8, 236–7, strength (modal strength) 12–13, 15–17,
240, 245, 250 19–20, 49–50
rhetorical question 130, 227–8 structured discourse context 180–1, 182,
yes/no question (polarity question) 130, 184, 186, 193, 195, 197, 204, 212, 220,
149, 156 227–8, 232–3
question set 155, 180–1, 193–5, 204, 214, subjunctive 1, 4–7, 47–8, 68, 69–71, 73,
227, 237 74–7, 78, 80–7, 89–91, 93–100, 102–3,
question under discussion (QUD, or 105–20, 133, 145–6, 149, 156, 207, 223,
table) 186–7, 193–4, 220, 227, 240, 250 232–3, 235–9, 241–2, 246

i i

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

i i

282 index

subsentential modality (subsentential truth condition 9, 14, 26, 56, 61, 64, 66,
modal) 13, 16, 33, 47–8, 49–50, 52, 55, 80, 94, 202, 221
56, 61–2, 64–5, 67–9
suppletive imperatives 114, 207 update potential 30, 40, 43–4, 139, 173,
176–7, 178, 181–2, 188, 195, 198, 238, 247
table, see question under discussion
teleological modality (teleological verb second (V2) 93, 142, 147–8, 156,
modal) 11 230, 239
temporal 55, 96, 115, 116–18, 241, 244–6 verbal mood 1, 4–8, 20–1, 33–4, 36–7, 47–9,
tense 9, 50, 70, 93, 111–12, 116–17, 223, 55, 57, 59, 61, 65, 67–74, 77–80, 89–90,
229–30, 241, 243–7 93–6, 98–104, 106–9, 113–14, 116–20,
sequence of tense 111, 241 122, 133, 146, 149, 232–43, 246–7,
to-do list 155, 180–1, 213–15, 218–19, 227, 249–51
235, 240 veridicality (veridical) 97–8, 111
truth 9, 14, 26, 54, 56, 61, 64–6, 70–3, 80,
87, 93–6, 98, 102–4, 106, 139, 155, 172, weak necessity 13, 16, 50, 205
174–5, 200, 202, 221, 232, 235, 247–8 wh-clause (wh-word, wh-question,
truth-based approach (truth in a wh-movement) 36, 127, 131–2,
designated set) 70, 93–103, 104–7, 232 150, 222

i i

i i