Writing the Book of the World

Writing the Book of the World

Theodore Sider

CLARENDON PRESS • OXFORD

Great Clarendon Street, Oxford ox2 6DP
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 in
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi
New Delhi Shanghai Taipei Toronto
With offices in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
© Theodore Sider 2011
The moral rights of the authors have been asserted
Database right Oxford University Press (maker)
First published 2011
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, 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 book in any other binding or cover
and you must impose the same condition on any acquirer
British Library Cataloguing in Publication Data
Data available
Library of Congress Cataloging in Publication Data
Data available
ISBN 978-0-19-969790-8
For Jill
Preface
The central theme of this book is: realism about structure. The world has a
distinguished structure, a privileged description. For a representation to be
fully successful, truth is not enough; the representation must also use the right
concepts, so that its conceptual structure matches reality’s structure. There is
an objectively correct way to “write the book of the world”.
Realism about predicate structure is fairly widely accepted. Many—
especially those influenced by David Lewis—think that some predicates (like
‘green’) do a better job than others (like ‘grue’) at marking objective
similarities, carving nature at the joints. But this realism should be extended,
beyond predicates, to expressions of other grammatical categories, including
logical expressions. Let “there schmexists an F” mean that the property of
being an F is expressed by some predicate in some sentence of this book.
‘Schmexists’ does not carve at the joints; it is to the quantifier ‘there exists’
as ‘grue’ is to ‘green’. Likewise, the question of joint-carving can be raised
for predicate modifiers, sentential connectives, and expressions of other
grammatical categories. (Structure is a generalization and extension of
Lewisian naturalness.)
I connect structure to fundamentality. The joint-carving notions are the
fundamental notions; a fact is fundamental when it is stated in joint-carving
terms. A central task of metaphysics has always been to discern the ultimate
or fundamental reality underlying the appearances. I think of this task as the
investigation of reality’s structure.
Questions about which expressions carve at the joints are questions about
how much structure reality contains. Whether reality contains causal, or
ontological, or modal structure is a matter of whether causal predicates,
quantifiers (or names), and modal operators carve at the joints. Such
questions lie at the center of metametaphysics. Those who say that questions
of ontology are “merely verbal”, for example, are best regarded as holding
that reality lacks ontological structure. Such deflationary metametaphysical
stances are thus themselves metaphysical stances. There is no ametaphysical
Archimedean point from which to advance deflationary metametaphysics,
since any such metametaphysics is committed to at least this much
substantive metaphysics: reality lacks a certain sort of structure.
A subsidiary theme is: ideology matters. There is an unfortunate tendency,
perhaps encouraged by bad terminology, to psychologize Quine’s notion of
ideology: to regard a theory’s choice of primitive notions—its ideology—as a
merely psychological or linguistic or conventional matter (in contrast to the
entities it postulates—its ontology—which is part of its objective content).
Philosophers reject their opponents’ ideology in psychological/semantic
terms: “I don’t understand what you mean by that.” And when introducing
their own ideology, the hurdle to be passed is again psychological/semantic:
primitive notions must be “intelligible”. But there is a squarely metaphysical
issue concerning any proposed piece of ideology (including logical and
quasi-logical ideology such as modal operators or second-order quantifiers):
does reality contain the requisite structure? If it does, then “intelligibility” in
previously “understood” terms is not required for successful reference to and
theorizing about that structure, no more in metaphysics than in physics.
A shift of focus from psychological/semantic to metaphysical constraints
on ideology is at times liberating for metaphysics, but it also keeps our feet
on the ground, by restraining the tendency to evade ontological commitments
by adding to ideology. A fundamental theory’s ideology is as much a part of
its representational content as its ontology, for it represents the world as
having structure corresponding to its primitive expressions. And the world
according to an ideologically bloated theory has a vastly more complex
structure than the world according to an ideologically leaner theory; such
complexity is not to be posited lightly.
Fixating on ontology while ignoring ideology is both too narrow and
incautious.1 It is too narrow because the goal of metaphysics is to give a
fundamental description of the world, and doing so requires more than
merely saying what there is. It is incautious because it uncritically assumes
that quantificational structure is fundamental. If quantificational structure is
indeed fundamental (as I think it is), ontology deserves its place in
fundamental metaphysics. But if quantificational structure is not fundamental,
then ontological inquiry deserves little more attention within fundamental
metaphysics than inquiry into the nature of catcher’s mitts.
A final theme is a “pure” conception of metaphysics, free of certain
encumbrances. One encumbrance is doing metaphysics primarily in modal
terms. Against this, there is a growing consensus that modal notions are too
coarse for metaphysics, and that notions in the vicinity of “fundamentality”,
“in virtue of”, and the like, should not be understood in modal terms. A
second encumbrance is linguistic entanglements. Here too, there is a growing
consensus: that it is not so important for metaphysical and linguistic theory to
neatly mesh. The fundamental metaphysics underlying a discourse might
have a structure quite unlike that suggested by the discourse. Whereas a good
linguistic theory must fit the suggested structure, good metaphysics must fit
the underlying structure.2
This book presented an organizational challenge. Theory-then-
applications would have been neatest, but the concept of structure is
unfamiliar enough that readability demanded early applications. My
compromise was to intermingle. Chapter 1 introduces the concept of structure
and describes in a preliminary way how it will be applied. Chapter 2 begins
to present the theory, arguing that structure is primitive and objective, and
defending an epistemology of structure. Chapters 3–5 turn to applications,
showing how structure illuminates explanation and laws, reference,
epistemology, physical geometry, substantivity, and metametaphysics.
Chapters 6–8 return to theory, arguing that expressions of any grammatical
category (not just predicates) can be evaluated for structure, addressing
various abstract questions about how structure behaves, and criticizing
certain rival concepts (such as truthmaking and ground). Chapters 9–12
return to applications, showing how the metaphysics of four domains—
ontology, logic, time, and modality—looks when conceptualized in terms of
structure. Chapter 13 concludes with a sketch of a “worldview”: a
comprehensive metaphysics cast in terms of structure. As a guide to those
who wish to read selectively:
The metaphysics of structure: chapters 1, 2, 6–8;
Applications: chapters 3–5, 9–12;
Metametaphysics: chapters 4–5, 9, and (to a lesser extent) 10–12;
Mix of first-order and meta- metaphysics: chapters 9–13.

I am grateful to many people for helpful discussions and feedback: Frank

Arntzenius, Elizabeth Barnes, Paul Boghossian, Craig Callender, Ross
Cameron, David Chalmers, David Copp, Troy Cross, Louis deRosset, Janelle
Derstine, Cian Dorr, Ant Eagle, Andy Egan, Matti Eklund, Adam Elga, Matt
Evans, Delia Graff Fara, Mike Fara, Hartry Field, Hilary Greaves, Liz
Harman, Allan Hazlett, Eli Hirsch, Thomas Hofweber, Paul Horwich, Alex
Jackson, Carrie Jenkins, Boris Kment, Tora Koyama, Uriah Kriegel, Heather
Logue, Ofra Magidor, Ishani Maitra, Colin Marshall, Farid Masrour, Andy
McGonigal, Ian McKay, Joseph Melia, Ulrich Meyer, Alan Musgrave, Daniel
Nolan, Jill North, Tim O’Connor, Laurie Paul, Zach Perry, Agustín Rayo,
Tony Roark, Dan Rothschild, Stephen Schiffer, Michael Schweiger, Adam
Sennet, Alan Sidelle, David Sosa, Ernie Sosa, Joshua Spencer, Jason Stanley,
Irem Kurtsal Steen, Steve Steward, Sharon Street, Zoltán Gendler Szabó,
Amie Thomasson, Jason Turner, Ryan Wasserman, Brian Weatherson, Ralph
Wedgwood, Bruno Whittle, Tim Williamson, Tobias Wilsch, Chris Wüthrich,
Stephen Yablo, and Dean Zimmerman. I’m especially grateful to Karen
Bennett, Gideon Rosen, Jonathan Schaffer, and Robbie Williams for
extensive and challenging comments (which, I fear, I have not fully
addressed). Thanks also to Oxford University Press and to Blackwell
Publishing for permission to include bits of Sider (2003), Sider (2009), and
Sider (2007a).
I’d also like to thank Kit Fine, John Hawthorne, and Phillip Bricker. I’ve
learned much from talking to Kit about fundamentality in the past few years,
and from thinking through his writings on the subject. John read large
portions of the manuscript and gave me many insightful comments, as well as
pushing me, years ago, to go beyond the predicate. Phil directed my
dissertation, which was on Lewisian naturalness. He taught me the power of
this idea, how to apply it to the philosophy of space and time, and much,
much more. My intellectual debt to Phil is massive.
Finally, it should be obvious how much this book owes to David Lewis.
His ideas on natural properties and relations have always seemed to me
among his best: powerful, correct, revolutionary yet deeply intuitive.

1 Dorr (2004, section 1) and Schaffer (2009a) make related complaints.

2 Kit Fine’s (1994a; 2001) recent work has been especially influential in forging both consensuses.
Contents
1 Structure
1.1 Structure: a first look
1.2 Philosophical skepticism about structure
1.3 Structure in metaphysics: a preview
2 Primitivism
2.1 Understanding
2.2 Primitivism supported
2.3 Epistemology
2.4 Against reduction
2.5 Against subjectivity
2.6 The privilege of the physical
3 Connections
3.1 Explanation and laws
3.2 Reference magnetism
3.3 Induction and confirmation
3.4 Intrinsic structure in physical spaces
4 Substantivity
4.1 Nonsubstantive questions
4.2 Substantivity characterized
4.3 Conventionality
4.4 Subjectivity
4.5 Epistemic value
4.6 Objectivity of structure
5 Metametaphysics
5.1 The challenge of metaphysical deflationism
5.2 Personal identity, causation
 5.3 The metaphysics room
5.4 Substantivity in nonfundamental disputes
5.5 A test case: extended simples
5.6 Metametaphysics as just more metaphysics
6 Beyond the predicate
6.1 The reason to generalize
6.2 Inapplicability of the similarity test
6.3 No entities
6.4 Unclear epistemology?
6.5 Logical conventionalism
7 Questions
7.1 Complete?
7.2 Pure?
7.3 Purity and connection
7.4 Metaphysical semantics
7.5 Completeness and purity reformulated
7.6 Metaphysics after conceptual analysis
7.7 Metaphysical semantics for quantifiers
7.8 Metaphysics and the study of language
7.9 Nonfundamental ground
7.10 Subpropositional?
7.11 Absolute?
7.11.1 Absolutism and comparative structure
7.11.2 Absolutism and infinite descent
7.12 Determinate?
7.13 Fundamental?
8 Rivals
8.1 Fine’s concepts
8.2 First Finean view: grounding and reality
8.2.1 Ground and purity
8.2.2 Ground and infinite descent
8.3 Second Finean view: reality
8.3.1 Explanation of fundamental truths
8.3.2 Combinatorialism about fundamentality
8.3.3 Combinatorialism about determinacy
 8.3.4 No fundamental truths
8.3.5 Nihilism and deflationism
8.4 Truthmaking
8.5 Truthmaking as a theory of fundamentality
8.6 Schaffer: entity-grounding
8.7 Entity-fundamentality
9 Ontology
9.1 Ontological deflationism
9.2 Ontological realism
9.3 Ontologese
9.4 Predicates not the issue
9.5 Quantifier variance
9.5.1 Quantifier variance and domain restriction
9.6 Objections to quantifier variance
9.6.1 The semantic argument
9.6.2 No foundation
9.6.3 No epistemic high ground
9.6.4 Indispensability
9.7 Easy ontology
9.8 Analyticity
9.9 Against easy ontology
9.10 Other forms of easy ontology
9.11 Metaontology and conceptions of fundamentality
9.12 Ontological commitment
9.13 Quantifiers versus terms
9.14 NonQuinean first-order ontology
9.15 Higher-order quantification
10 Logic
10.1 Fundamental logic
10.2 Hard choices
10.3 Nonfundamental metalogic
10.4 Logical pluralism
10.5 Objectivity in model theory
10.6 Classical logic and fundamentality
11 Time
 11.1 Presentism
11.2 Presentist ideology: quantified tense logic
11.3 Is the dispute over presentism substantive?
11.4 Passage
11.5 Williamson on saturation and contingency
11.6 Change and saturation
11.7 Talk of saturation is legitimate
11.8 The metaphysics of saturation
11.9 Varieties of passage
12 Modality
12.1 No fundamental modality
12.2 A Humean strategy for reduction
12.3 Logical consequence and mathematical truth
12.4 Analyticity
12.5 Laws of metaphysics
12.6 Determinates and determinables
12.7 Contextualism
12.8 The necessary a posteriori
12.9 Micro-reduction
12.10 De re modality
12.11 Family resemblances
12.12 Spreading arbitrariness
13 A Worldview
References
Index
1 Structure
Metaphysics, at bottom, is about the fundamental structure of reality. Not
about what’s necessarily true. Not about what properties are essential. Not
about conceptual analysis. Not about what there is. Structure.
Inquiry into necessity, essence, concepts, or ontology might help to
illuminate reality’s structure. But the ultimate goal is insight into this
structure itself—insight into what the world is like, at the most fundamental
level.

1.1 Structure: a first look

Discerning “structure” means discerning patterns. It means figuring out the
right categories for describing the world. It means “carving reality at its
joints”, to paraphrase Plato. It means inquiring into how the world
fundamentally is, as opposed to how we ordinarily speak or think of it.
Consider three objects: two electrons in identical intrinsic states, and a
cow. It is the most natural thing in the world to say that the electrons are
perfectly similar to each other, and that neither is perfectly similar to the cow.
The three objects should be divided into two groups, one containing the
electrons, the other containing the cow. The electrons go together, and
neither goes with the cow.
Or imagine a universe that is entirely full of fluid. A plane divides the
universe into two halves, one in which the fluid is uniformly red, the other in
which the fluid is uniformly blue (figure 1.1). Now imagine a group of people
who encounter this universe, but accord no special status to the dividing blue-
red plane. Instead of thinking of the universe as divided into the red and blue
halves, they think of it as being divided in half by a different plane, marked
by the dashed line in figure 1.2. And they do not use predicates for red and
blue. Instead, they have a pair of predicates that they apply uniformly within
the two regions separated by their dividing plane. These predicates (whose
extensions are indicated by diagonal hash lines in the diagram) cut across the
predicates ‘red’ and ‘blue’. The regions to the left of the dashed line they call
“bred”; the regions to the right they call “rue”.

FIG. 1.1. The red–blue world

FIG. 1.2. Bizarre carving of the red–blue world

It is almost irresistible to describe these people as making a mistake. But

they’re not making a mistake about where the red and blue regions are, since
they make no claims about red or blue. And they make no mistakes when
they apply their own concepts. The regions that they call “bred” are indeed
bred, and the regions they call “rue” are indeed rue. The problem is that
they’ve got the wrong concepts. They’re carving the world up incorrectly. By
failing to think in terms of the red/blue dividing plane, they are missing
something. Although their beliefs are true, those beliefs do not match the
world’s structure.

1.2 Philosophical skepticism about structure

All is well until we encounter a philosopher, who, as usual, asks some
uncomfortable questions. Why do the two electrons “go together”, the
philosopher wants to know? Yes, they share many features in common: each
has 1.602 × 10−19 C charge, 9.109 × 10−31 kg mass, and so on. But there are
plenty of features that the electrons do not share. They are in different
locations, travel at different velocities, and are parts of different wholes. And
why doesn’t the cow go together with the electrons? If all three are located in
North America, then all three share the feature being located in North
America. And all three share the feature: being an electron or a cow.
The philosopher continues: what is wrong with carving the red–blue world
along the diagonal plane? What is wrong with grouping the bred things
together and the rue things together? All bred things really are bred; they all
share the feature of being on the left side of the diagonal plane. One might
protest that not all bred things are alike, since some are red and some are
blue; but the philosopher will reply that carving the world along the vertical
plane is no better on this score. Not all red things are alike, since some are
bred and some are rue.
In fact, once we get the hang of the philosopher’s way of thinking about
“features”, we can see that any two objects share infinitely many features,
and also differ with respect to infinitely many features. For consider any
objects x and y. Where Fx and Fy are any features of x and y, respectively, x
and y share the feature: being either Fx or F y. And they share the feature
being either Fx or F y or 1 kg mass. And they share the feature being either
Fx or Fy or 2 kg mass. And so on. So they share infinitely many features. As
for the infinitely many features with respect to which they differ, consider:

being Fx, and located at L

being Fx-or-1-kg-mass, and located at L
being Fx-or-2-kg-mass, and located at L
etc.

where L is some location occupied by x but not y. Object x has each of these
features; object y lacks each.
The crux is obviously the philosopher’s willingness to allow such
“features” as being either an electron or a cow, and to treat them on a par
with features like being an electron and being a cow. If we had nothing but
the philosopher’s features to go by, then indeed, we wouldn’t be able to make
any sense of a “correct” way to group our three objects, or of the electrons
being more similar to each other than to the cow. If, on the other hand, we
could make a distinction between genuine features—features that are
fundamental, that carve nature at the joints, whose sharing makes for
similarity—and the rest, then we could say what we want. Can we make this
distinction?
Concepts and distinctions that resist definition in terms of the popular
philosophical ideology of the day tend to be viewed with suspicion. Thus it
was that throughout much of the twentieth century, philosophers tended not
to speak of genuine features. Quine’s extensionalism, for example, which
dominated the 1950s and 1960s, allowed only a meager set of concepts to be
used in drawing distinctions (roughly, those of first-order logic plus an array
of scientific predicates). Noticing the presence of disjunction in the
definitions of many philosopher’s features, an extensionalist might begin an
attempt to characterize genuineness by disqualifying features defined in this
way. But what language do we use to evaluate whether a feature is “defined
using disjunction”? Speakers of English must use ‘or’ to define the feature:
being an electron or cow, but speakers of a language with a primitive
predicate for this feature—‘blurg’, call it—can define the same extension
without using ‘or’. Indeed, if the language is strange enough, its speakers
would need to use ‘or’ and other logical connectives to say things that in
English may be said using simple predicates like ‘cow’ and ‘electron’, just as
we must use logically complex predicates of English to say what they say
using ‘blurg’. The extensionalist attempt fails to characterize an appropriately
language-independent notion of genuineness.1
In the 1970s, modality became kosher ideology, and there were renewed
attempts to define concepts in the vicinity of structure. For instance, Roderick
Chisholm (1976, p. 127) and Jaegwon Kim (1982, pp. 59-60) tried to give a
modal definition of the notion of an intrinsic property—a property that an
object has just by virtue of what it’s like in itself, independently of how it is
related to other objects. They proposed, roughly, that a property is intrinsic if
and only if it is possibly instantiated by an object that is alone in the world.
But this definition was shown to be unacceptable. The property of being
alone in the world, and the property of either (being alone in the world and
being green) or (not being alone in the world and being blue), satisfy the
definition but are extrinsic (Lewis, 1983a).
(The 70s’ fixation on modality was doubly unfortunate. Not only are
modal tools too crude;2 they’re also distant from the subject matter of most of
metaphysics. It is needlessly indirect to approach the question of what the
world is like by asking what it must be like and what it might have been
like.3)
Since the 1980s many philosophers have become comfortable with a
richer ideology, one that includes notions in the vicinity of “genuine feature”,
“intrinsic property”, and the like. The zeitgeist has been that these notions are
legitimate even if they cannot be defined in other terms. Two Davids have led
the way.4 David Armstrong (1978a; b) used the traditional doctrine of
universals to draw the distinction between genuine and nongenuine features.
Some predicates, like ‘is an electron’, perhaps, stand for universals,
Armstrong said; but others do not: there simply is no universal of “being
either a cow or an electron”. Through sheer force of will as much anything,
he put realism about genuine features on the map. But as our second David,
David Lewis (1983b) showed, Armstrong embedded this insight in a quite
independent dialectic: the traditional debate over the existence of universals
and their role in a general analysis of predication. According to Lewis, we
can incorporate Armstrong’s insight by admitting a notion of “natural
properties and relations” (those properties and relations that carve nature at
the joints) without thinking of these as universals in the traditional sense, and
without taking on the (misguided, according to Lewis) project of giving a
general analysis of predication. The notion of a genuine feature was thus
freed from unwanted entanglements.
Of course, everyone can agree that there is some difference between being
an electron and being either an electron or a cow. If nothing else, ordinary
English has a single word for the former attribute. What distinguishes
Armstrong and Lewis is that they regard the distinction as objective.
Structure, too, is to be understood as objective. There are hard questions
about what objectivity amounts to (some of which will be discussed in
chapter 4), but the intuitive idea is clear: whether a property, word, or
concept5 carves at the joints has nothing to do with the place of the concept in
human languages, conceptual schemes, biology, or anything like that. Thus
“fundamental” (which I use more or less interchangeably with “joint-carving”
and “part of reality’s structure”) signifies a metaphysical, rather than
conceptual, sort of fundamentality. Humans may need to acquire other
concepts first before they grasp joint-carving ones; and conversely, those
concepts we acquire first, or most easily, may fail to carve at the joints.

1.3 Structure in metaphysics: a preview

The goal of this book is to push forward the front of realism about structure. I
want to expand our conception of structure’s importance, generalize the
concept of structure, investigate its nature, use it as the foundation of
“metametaphysics”, and reconceptualize metaphysics in terms of it.
The connection to similarity is only the beginning of the importance of the
notion of structure. As we will see, structure pops up throughout philosophy,
in our thinking about reference, epistemology, spacetime, objectivity, and
other matters.
Structure is particularly central to metaphysics. The heart of metaphysics
is the question: what is the world ultimately, or fundamentally, like? And
fundamentality is a matter of structure: the fundamental facts are those cast in
terms that carve at the joints.
The truly central question of metaphysics is that of what is most
fundamental. So in my terms, we must ask which notions carve perfectly at
the joints. By using ‘red’ and ‘blue’, we carve more closely to reality’s joints
than do the speakers of the ‘bred’/‘rue’ language. But we do not thereby
carve perfectly at the joints; colors are presumably not perfectly fundamental.
To carve perfectly, one must use the most fundamental concepts, expressing
the facets of reality that underly the colors.
Which concepts are the perfectly fundamental ones? In my view, certain
concepts of physics, logic, and mathematics.6 But this thesis about structure
is not built into the idea of structure, and defending it is not one of the main
goals of this book. The great metaphysical disputes concern which theses of
this sort are true; my goal is to explain what is at stake in such disputes, not
to settle them. Is mentality part of reality’s fundamental structure? (Modal
theses in the philosophy of mind, such as psychophysical supervenience, are
crude ways of getting at what clearly was the issue all along: whether reality
is fundamentally mental.) Do mathematical entities exist, in the fundamental
sense of ‘there exist’, and if so, what are the fundamental features of those
entities? Do causal or nomic notions have any place in a fundamental
description of the world? These are questions about structure.
Metametaphysics—inquiry into the status of metaphysics—will be central
in this book. Is the pope (or Robinson Crusoe, or a twelve-year-old boy) a
bachelor? Intuitively, the question is merely verbal or conceptual. To answer
it, all we need to do is investigate our concept of a bachelor; intuitively, all
that is at stake is how we use the word ‘bachelor’. In contrast, the question of
whether there is any lithium in a certain region on Mars has nothing to do
with word use or concepts; it is substantive. This rough and ready notion of
substantivity needs to be clarified; after all, the statement that Robinson
Crusoe is a bachelor is no more about our concept of a bachelor than the
statement that there is lithium in the region is about our concept of lithium.
Nevertheless, there is a strong intuitive contrast between the two questions.
The opponents of metaphysics (and even some renegade practitioners)
tend to regard many metaphysical questions as being—to some extent,
anyway—like the question of whether the pope is a bachelor. True believers,
on the other hand, tend to think of their favorite metaphysical questions as
being substantive, like the question about lithium. In my view, whether a
question is substantive—in one important sense of ‘substantive’—depends
largely on the extent to which its terms carve at the joints; to the extent, that
is, that the question concerns the world’s fundamental structure. The central
metametaphysical questions are about how much structure the world
contains.
Consider two properties:
Being an unmarried male
Being an unmarried adult male eligible for marriage.
It may well be that exactly one of these properties is (determinately) what we
mean by ‘bachelor’. So it may be that the question of the pope’s
bachelorhood has an answer. But neither of these two properties carves
nature at the joints better than the other. The unmarried males don’t go
together any more than do the unmarried males eligible for marriage. A
linguistic community that used the word ‘bachelor’ for the first property
would not be getting at the world’s structure any better than a community that
used the word for the second property. And since the pope is an unmarried
male who is ineligible for marriage, speakers of the first community speak
truly when they say ‘The pope is a bachelor’, whereas speakers of the second
community speak truly when they say ‘The pope is not a bachelor’. So,
intuitively, the only question facing us is: which sort of linguistic community
do we inhabit? Which of two equally good ways to talk is our way to talk?
The question of whether there is lithium in the region near Mars has a
very different status. Suppose that the region does indeed contain lithium. We
can imagine a linguistic community that uses the word ‘lithium’ exactly as
we do, but with one exception: their word does not apply to the lithium (in
our sense of the word) in the region. So ‘There is lithium in the region’
counts as true in our language, and false in theirs. But here the parallel with
the previous paragraph ends. The lithium in the region is just like the lithium
elsewhere, so the imagined linguistic community fails badly to carve nature at
the joints. They fail to group together things that, objectively, go together.
The question of whether there is lithium in the region is not just a question of
which of two equally good ways to talk is our way to talk.
Few would deny that the question of the pope’s bachelorhood is
insubstantial in a way that the question about lithium is not. But in
metaphysics, things are far less clear. Consider questions of ontology, for
example. There has been much discussion recently of whether tables and
chairs and other composite material objects exist. It is generally common
ground in these discussions that there exist subatomic particles that are
“arranged tablewise” and “arranged chairwise”; the controversy is over
whether there exist in addition tables and chairs that are composed of the
particles. Is this really a substantive debate about the world? Most of the
ontologists engaged in the debate think so—or really, presuppose so. But Eli
Hirsch, Hilary Putnam, and other “ontological deflationists” have argued that
the debate is in some sense merely verbal or conceptual. The
“metaontological” question here ultimately boils down, I think, to a question
of structure: whether quantificational notions like ‘there exists’ carve at the
joints. What the ontological deflationists have in effect been saying is that
reality would need fundamental quantificational structure in order for the
question of whether there exist tables and chairs to be worth asking, and that
this structure is, in fact, missing. I oppose ontological deflationists in chapter
9, but they deserve credit for raising an important and difficult question—a
question that is in a way more foundational than the first-order question of
what there is.
There are similar foundational questions throughout metaphysics. Do
modal concepts carve at the joints (chapter 12)? (Here my answer is no;
modality is not the core of metaphysics that some take it to be.) Do tensed
concepts (chaper 11)? (Again, no; but seeing the issue as concerning reality’s
temporal joints helps to illuminate what are otherwise extremely perplexing
questions.) Do logical concepts (chapter 10)? (Here I say yes. Certain debates
over the “correct” logic are genuine, and are not linguistic or conceptual; they
are as substantive as ontological debates.)
More generally, metaphysicians regularly speak of what is “really” or
“genuinely” the case. (Often they feel guilty about it, but don’t know how to
stop.) As Kit Fine (2001; 2009) has emphasized, such talk is central to
metaphysics, but in dire need of explication. When a nominalist says that
there do not really exist abstract entities like properties, while granting that
frogs share more properties in common with crocodiles than they share with
humans, the ‘really’ is essential; otherwise she contradicts herself. Those
who think that “time is like space” say that there is no “genuine” or
“objective” distinction between past, present, and future, but they do not deny
that there once were dinosaurs. Again, if ‘genuine’ and ‘objective’ are
dropped then the position becomes incoherent. These claims are not merely
about what is true; they are about what is true at the fundamental level.
If the concept of structure is to play this role in metametaphysics, it must
be generalized beyond Armstrong’s notion of a universal and Lewis’s notion
of natural properties and relations. For many metaphysical questions are not
about universals, properties, and relations. The crucial expressions in
ontology, logic, and modality do not stand for universals, properties, or
relations; these expressions are quantifiers and operators, not predicates. Our
conception of structure, therefore, must allow us to ask, of expressions of any
grammatical category, whether they carve at the joints.
Call a language “fundamental” if all of its expressions carve at the joints.
Realism about structure leads to realism about fundamental languages. On the
generalized conception of structure, in order to be fundamental, it is not
enough that a language have the right predicates. It must also have the right
logical apparatus. Will a fundamental language contain quantifiers? The
sentential connectives of propositional logic? Modal or tense operators? The
realist about structure thinks that these questions have objective answers.
There is a privileged way to “write the book of the world”.

1 The paradigm of first-order logic had perhaps the following additional influence. The standard
model theory of first-order logic treats the semantic values of (n-place) predicates as subsets of the (n-
place Cartesian product of the) domain. Viewed from a purely set-theoretic perspective, the semantic
values of the predicates ‘is an electron’ and ‘is an electron or cow’ are on a par: each is a subset of the
domain.
2 On which see, for instance, Fine (1994a); Restall (1996).
3 I also suspect that the right account of how the world might have been and must be defers to how
the world is (chapter 12).
4 Earlier relevant work includes Quinton (1958); Quine (1969); Putnam (1975c); Bealer (1982,
chapter 8).
5 Subtleties will come later, but to forestall misunderstanding: 1. Structure is a worldly, not
conceptual or linguistic, matter (my informal talk of “notion/word/concept X carves at the
joints”notwithstanding). 2. ‘Structure’ is not a noun; structure is not an entity or stuff (this very
sentence, and phrases like “how much structure the world contains”, notwithstanding). 3. ‘Structure’
and its variants are not predicates—not of properties, nor of any other sorts of entities (“charge carves
at the joints” notwithstanding). 4. My most basic notion of structure is absolute, although I allow a
derivative notion that comes in degrees. 5. Structure includes distinguished monadic features (such as
charge), not just relational ones (despite what may be suggested by the term ‘structure’).
6 Plus the concept of structure itself! See section 7.13.
2 Primitivism
I cannot define ‘structure’. As we will see, a rich characterization can be
given: connections to other concepts, theses about its behavior, and an
official regimentation for talking about it. But none of this will add up to a
definition. Indeed, I will argue in section 7.13 that structure is perfectly
fundamental.

2.1 Understanding
I know from bitter experience that philosophers are wary of this primitivism.
Many times I have been asked (to murmuring general approval): “What on
earth do you mean by ‘structure’??”.
Let’s be realistic about the extent and value of definitions. Philosophical
concepts of interest are rarely reductively defined. Still more rarely does our
understanding of such concepts rest on definitions.
On what does our understanding of philosophical concepts rest?
Sometimes there is a perceptual basis: we directly experience space and time,
perhaps. But the perceptual model is mostly unhelpful (think of modality,
logic, laws of nature, identity over time, morality justice, knowledge,
justification, …). We generally “understand” philosophical concepts to the
extent that we know what role they play in our thinking. (Understand “role”
here very broadly, so as to include particular “cases”—we judge Gettier’s
(1963) Smith not to know that Jones owns a Ford or Brown is in Barcelona—
as well as general inferential patterns—we think of identity as a transitive
relation; we think of inability to refrain from action as an excuse.)
Philosophers sometimes slip into a magical-grasp picture of
understanding. An opponent wields a crucial term. She will not be bullied
into equating it with some combination of preferred terms. An inward search
for a mystical mental state of UNDERSTANDING comes up empty. The
opponent is pronounced confused or obscure.
 Philosophical terms can be unclear: when they have been given no clear
theoretical role to play. But ‘structure’ has a relatively clear role—given in
this book and elsewhere. What more is wanted? The perceived magical grasp
of more familiar concepts like modality, in-virtue-of, or law of nature, is due
solely to the fact that we’ve become accustomed to talking about them. The
theoretical roles backing those concepts are no richer or better specified than
the role backing structure. Philosophy is not just the building of theories on
previously existing concepts. We also build new concepts, by building
theories that use them.
This is not to say that all there is to meaning and reference is inferential
role. Meaning and reference may well be determined by external factors that
transcend inferential role (see section 3.2). So even if structure’s inferential
role is richly specified, the concept may nevertheless fail to refer to anything.
But that’s true of any philosophical concept: the world may simply fail to
contain anything—or any unique thing—fitting the inferential role associated
with the concept. My hope is that this unhappy possibility is not realized.

2.2 Primitivism supported

And my argument that the unhappy possibility is not realized is simply the
overall argument of this book: recognizing structure improves our
understanding of the world—our understanding of:
(Objective) similarity (chapter 1);
Intrinsic properties (Lewis, 1986b, pp. 61–2);1
Laws of nature and explanation (section 3.1);
Reference (section 3.2);
Induction and confirmation (section 3.3);
The intrinsic structure of space and time (section 3.4);
Substantivity (chapter 4);
Epistemic value (section 4.5);
Metametaphysics (chapter 5);
Disputes about time, modality, ontology, logic (chapters 9–12)
(In the first four items I follow David Lewis.) Structure is a posit, a posit that
is justified by its ability to improve our theories of these matters.
Posits are most justified when they’re unifying. When a single posit can
be viewed as underlying multiple phenomena, this counts in favor of the
posit. But here we must distinguish two kinds of unification. Compare two
unifying features of Newtonian mechanics. Newtonian mechanics contains
two fundamental laws governing mass: the second law of motion, which
specifies the acceleration of a body as a function of its mass and the net force
acting on it, and the law of gravitation, which specifies the gravitational force
between a pair of bodies as a function of their masses and spatial separation.
The first unifying feature is the identification of inertial and gravitational
mass: the very same notion of mass is asserted to be involved in both laws.
This first unification is one of ideology, of the set of our undefined
words/concepts/notions. Instead of containing two notions of mass, one for
the second law (“resistance to acceleration”) and one for the law of
gravitation (“tendency to produce gravitational force”), Newtonian ideology
contains just one notion: mass. The second unification is the derivation of the
planets’ elliptical orbits. Here there was unification of fundamental
principles. The orbiting of the planets was shown to require no new
fundamental laws, since elliptical orbits were shown to follow from the the
second law and the law of gravitation. While both sorts of unification seem to
count in favor of a posit, too much of the former sort without any of the latter
seems rarely to be pursued. We like to keep our posits few in number, but we
also want them to obey a small number of fundamental laws, from which
much else can be derived.
To be sure, metaphysics isn’t rocket science. In metaphysics we lack
precisely formulated, sufficiently specific fundamental laws from which
myriad important and precise consequences may be rigorously derived. Still,
we can be guided by physics here. The posit of structure will be unjustified if
its alleged “applications”—its connections to similarity, intrinsicality, and the
rest—are in the end unified by nothing beyond a bare assertion that a single
notion of structure plays the needed role in each case. That would be
unification of the first sort without unification of the second sort. But positing
structure is not like this sort of bare assertion, since the applications are all
intertwined. For example, many of the applications can be seen as flowing
from a single principle connecting structure to epistemic value (section 4.5).
Also, the applications just seem intuitively appropriate. Similarity, intrinsic
structure of spacetime, substantivity, and the rest really do seem to be
connected to fundamentality in the ways to be proposed.2 This claim is
difficult to justify, but I hope the reader will agree once the applications have
been laid out.

2.3 Epistemology
A typical follow-up to “What do you mean by ‘structure’?” is “How are we
supposed to know—or even, reasonably believe—anything about structure?”
Unless structure is defined in more familiar terms, it’s thought, facts about
structure become epistemically inaccessible.
The epistemic worry leads to a further worry about understanding.
Suppose we reject the magical-grasp picture of understanding, as urged
above. Perhaps we replace it with a more inferentialist picture. But then, if
facts about structure are unknowable, structure-talk becomes inferentially
isolated, in which case our understanding of such talk is again threatened.
But why think that primitivism about structure has such drastic epistemic
consequences? The dialectic here is a familiar one. A “realist” resists
downsizing the facts in some domain—reducing them or regarding them as
subjective—and so is accused of making the facts in that domain
unknowable. Her response is fallibilist epistemology: the subjective or
reductive facts that her opponents offer as replacements are instead fallible
guides to the upsized facts.
Most philosophers are comfortable with taking the realist side of this
dialectic in the most familiar case: they reject Berkeleyan idealism,
phenomenalism, and other downsized conceptions of the external world, and
instead regard ordinary evidence as a fallible guide to upsized facts. But their
ranks dwindle as the subject matter becomes more metaphysical. The reason
for this is simple: many of our models of the nature of reasonable—albeit
fallible—belief about the external world do not apply straightforwardly to
beliefs about more metaphysical matters. For example, we do not seem to be
in causal contact with the facts debated by metaphysicians in the same way
that we are in causal contact with more familiar facts about the external
world.
But the models that immediately disallow reasonable belief in
metaphysics are too simplistic, and as a result are in trouble anyway. Our
causal contact with the facts of logic, mathematics, and particle physics, for
example, is quite unlike our causal contact with the facts of the everyday
external world. The ray of hope for the metaphysician is this: when the
models become more sophisticated, allowing for reasonable belief in logic,
mathematics, and particle physics, perhaps they will also allow for reasonable
belief in metaphysics as well.
The epistemology of metaphysics is far from clear; this any metaphysician
should concede. For what it’s worth, as a general epistemology of
metaphysics I prefer the vague, vaguely Quinean, thought that metaphysics is
continuous with science. We employ many of the same criteria—whatever
those are—for theory choice within metaphysics that we employ outside of
metaphysics. Admittedly, those criteria give less clear guidance in
metaphysics than elsewhere; but there’s no harm in following this argument
where it leads: metaphysical inquiry is by its nature comparatively
speculative and uncertain.
This Quinean thought suggests an epistemology for structure in particular.
Quine’s advice for forming ontological beliefs is familiar: believe the
ontology of your best theory. Theories are good insofar as they are simple,
explanatorily powerful, integrate with other good theories, and so on. We
should believe generally what good theories say; so if a good theory makes
an ontological claim, we should believe it. The ontological claim took part in
a theoretical success, and therefore inherits a borrowed luster; it merits our
belief. This all is familiar; but a believer in structure can say more. A good
theory isn’t merely likely to be true. Its ideology is also likely to carve at the
joints. For the conceptual decisions made in adopting that theory—and not
just the theory’s ontology—were vindicated; those conceptual decisions also
took part in a theoretical success, and also inherit a borrowed luster. So we
can add to the Quinean advice: regard the ideology of your best theory as
carving at the joints. We have defeasible reason to believe that the conceptual
decisions of successful theories correspond to something real: reality’s
structure.
The term ‘ideology’, in its present sense, comes from Quine (1951a;
1953). It is a bad word for a great concept. It misleadingly suggests that
ideology is about ideas—about us. This in turn obscures the fact that the
confirmation of a theory confirms its ideological choices and hence supports
beliefs about structure. A theory’s ideology is as much a part of its worldly
content as its ontology.
The familiar Quinean thought is that we search for the best—simplest, etc.
—theory that explains our evidence. My addition to this thought—though it
may have been implicit all along—is that this search is ideological as well as
doctrinal; we search simultaneously for a set of concepts and a theory stated
in terms of those concepts. We solve for the best and most explanatory pair
〈I, T1〉 of ideology I and theory TI in terms of that ideology. We do not hold
fixed our initial ideological choices (‘fire’, ‘air’, ‘water’ …) since there may
be limits to how good a theory can be formulated in those terms. Many of the
most dramatic advances in science are ideological; a new ideology (such as
that of Minkowskian spacetime) can dissolve intractable problems and enable
new, more powerful theories.
(Sometimes our evidence does not support a unique pair 〈I, TI〉. This is not
in itself worrisome; we do not know everything, after all. But in some cases,
it is hard to see what evidence could be mustered in favor of one pair rather
than another. For example, should our fundamental theory of part and whole
take ‘part’ or ‘overlap’ as primitive? Should our fundamental logical theory
take conjunction and negation, or instead, disjunction and negation, as
primitive? In such cases it’s hard to see how to choose, and indeed, hard to
believe that there could be a single correct choice. We will return to this issue
in section 10.2.)
The Quinean thought rationalizes commonly held beliefs about what is
fundamental. Nearly everyone agrees that physical notions like mass and
spatiotemporality are fundamental. These beliefs are reasonable because
those notions are drawn from highly successful theories.
The Quinean thought also rationalizes changes in beliefs about what is
fundamental. The special theory of relativity led to (at least) two such
changes. First, we came to regard electromagnetism as a single fundamental
force, rather than regarding electricity and magnetism as separate
fundamental forces.3 And second, we came to regard spacetime as lacking
absolute spatial and temporal separation. These changes weren’t ontic:
changes in which entities are accepted. Nor were they merely doctrinal:
changes in view, but phrased in the old terms. The changes were rather
ideological: we revised our fundamental ideology for describing the world.
The changes are rationalized by the Quinean thought because the
fundamental ideology of the special theory of relativity differs from the
fundamental ideology of Newtonian physics: in place of electrical, magnetic,
spatial, and temporal ideology, the special theory has unified ideology for
electromagnetism and unified ideology for spatiotemporal metrical structure.
The Quinean thought about ontology is sometimes put in terms of
indispensability: believe in the entities that are indispensable in your best
theory. The analogous thought about ideology may be similarly put: regard as
joint-carving the ideology that is indispensable in your best theory. This is
fine provided “indispensable” is properly understood, as meaning: “cannot be
jettisoned without sacrificing theoretical virtue”. The indispensability
argument for mathematical entities is not refuted by just any nominalistic
alternative to platonist mathematical physics. The nominalistic alternative
must be attractive as a scientific theory; it must compete with the platonist
theory for being simple, explanatory, and so on. Similarly, consider rewriting
a given theory of mass and charge in terms of schmass and charge, where the
schmass of an object is its mass if it has unit negative charge and twice its
mass otherwise. The rewritten theory has the same consequences about
charge and mass as the original, so ‘charge’ and ‘mass’ are in a sense
dispensable in physics. But the resulting theory is far worse as a theory. What
were syntactically simple generalizations in the old theory are no longer
simple in the new.
We have been exploring the positive side of the Quinean approach to
ideology: we can support claims about joint-carving by showing that the
ideology in question is part of a good theory. The approach has a negative
side too. Good theories must be as simple as possible, and part of simplicity
is having a minimal ideology. So we can oppose claims about joint-carving
by exhibiting good theories that do not contain the ideology in question.
The demand for minimal ideology recalls a familiar trade-off between
ontology and ideology. We often face a choice between reducing our
ontology at the cost of ideological complexity, or minimizing ideology at the
cost of positing new entities.4 If ideology is psychologized, the trade-off is
one of apples for oranges: whether to posit a more complex world or a more
complex mode of expression.5 But on the present approach, both sides of the
trade-off concern worldly complexity. A theory with a more complex
ideology posits a fuller, more complex, world, a world with more structure.
Thus ideological posits are no free lunch.
“Believe the ontology and ideology of your best theory” is schematic in
various ways. One in particular is worth mentioning: should the special
sciences be counted as part of our “best” theory? Saying yes leads to an
expansive conception of the fundamental; saying no—my preferred answer—
leads to a more austere conception. The defender of the latter answer must
concede that our understanding of the world would be severely impoverished
without the special sciences, but will insist that since facts about the special
sciences hold “in virtue of” more fundamental facts in some sense (chapter
7), they needn’t be cited in our “best” theory. This is a difficult issue, which I
will not attempt to resolve here.
We have, then, an epistemology for structure. Claims about structure can
be supported by evidence, and so are not inferentially isolated, and so are not
in danger of unintelligibility. This epistemology is admittedly superficial and
birds-eye. Then again, so are the epistemological models that are claimed to
preclude reasonable belief in metaphysics. What is needed for progress in
these issues is a more sophisticated and detailed understanding of the
epistemology of our more theoretical endeavors, such as mathematics and
theoretical physics, including their foundations.

2.4 Against reduction

Primitivism about structure would be unnecessary if structure could be
reduced. This section will argue briefly against a few reductive approaches.
But the matter cannot really be decided by a few quick arguments, since it
turns on the question of which global metaphysics is most attractive. My
primary aim in this book is to exhibit the attractions of my own approach,
rather than to refute others. (Also, much of my approach could be adopted by
a reductionist, provided the reduced notion of structure is sufficiently
objective and capable of broad application.)
Consider first reductions to putatively fundamental concepts. One can
object to such reductions in two ways: extensionally and systematically. One
can argue that a reduction fails to generate a reasonable extension for
‘structure’. Or one can argue, on global, systematic grounds, that structure
itself, rather than the proposed reducing concept, should be taken to be
fundamental.
Consider, for example, the proposal that a structural property or relation is
one that figures in some law of nature, where the notion of a law of nature is
taken to be fundamental. In my view, logical and mathematical notions, as
well as the notion of structure itself, carve at the joints; but it is unclear
whether these notions figure in the laws of nature. A notion of structure that
is too closely tied to lawhood will not be general enough to do all the work it
needs to do. This is an extensional objection. But there is also a systematic
consideration: it is better to posit fundamental structure than fundamental
lawhood. (One could posit both, but that would be overkill.) Later on I will
say more against all forms of fundamental modality, including fundamental
lawhood; but in brief: modal notions are generally of dubious explanatory
value. Adding the notion of law to physical theory, for example, doesn’t seem
to enhance its explanatory power. Also, the claims of physicists do not
bifurcate naturally into laws and mere facts (which is “spacetime is a four-
dimensional Lorentzian manifold”?), so introducing a fundamental notion of
law imposes a distinction on their inquiry that isn’t there to begin with.6
For similar extensional and systematic reasons, we should not define
structure in terms of a fundamental notion of causation. Causation is a
particularly unsavory fundamental posit—at least if the posit is intended to
closely match our ordinary concept of causation. It takes only a glance at the
recent literature on causation to appreciate how arbitrary and baroque our
ordinary concept of causation is.7 One might posit a fundamental sort of
causation that is only distantly related to our ordinary concept, perhaps on the
grounds that the sciences demand recognition of some sort of fundamental
scientific causation. The case for fundamental causation is, I suspect, strictly
weaker than the case for fundamental laws of nature; but at any rate, the
objections to defining structure in terms of fundamental scientific causation
are the same as those in the previous two paragraphs.
Consider, next, a proposal to define structure in terms of supervenience,
where supervenience is then defined in terms of a putatively fundamental
notion of metaphysical necessity.8 Call a set of properties and relations
complete iff all properties and relations supervene (globally, across all
metaphysically possible worlds) on it; and call a set minimally complete iff it
is complete and contains no proper subset that is also complete. The proposal
I want to attack is this: the set of structural properties and relations can be
defined as the minimally complete set of properties and relations.
Here too there is the systematic objection that structure is a better
fundamental posit than any modal notion. But there are also extensional
objections. In Sider (1996b, section 3.3) I gave two.9 First, it is unclear
whether the structural properties and relations are complete: nonqualitative
properties and relations (such as the property of being identical to Ted) may
not supervene on them, and it’s hard to exclude nonqualitative properties and
relations in the analysis without helping oneself to some notion in the vicinity
of structure. Second, and more importantly, even if the set of structural
properties and relations is complete, it may not be minimally complete.
Perhaps both earlier-than and later-than are structural relations, for example.
Third, and still more importantly, even if the set of structural properties and
relations is minimally complete, it won’t be the only such set: certain other
sets containing “grueified versions” of the structural properties and relations
will also be minimally complete. Begin with a set containing charge and
mass, say, and replace mass with schmass (see above). Since an object’s mass
supervenes on its charge and schmass, any set that supervenes on the former
set will supervene on the latter set; so the latter set is complete. And
minimally so: since it doesn’t contain mass properties, removing the schmass
properties would disrupt its completeness.
There is a response to my third argument that I did not consider in my
(1999b) paper. Perhaps we can single out the set of structural properties and
relations, from amongst all the minimally complete sets, as that set that
“enables the best Lewisian laws”, in the following sense. Lewis used his
notion of naturalness to give a reductionist, “Humean” theory of laws of
nature, a souped-up version of the old regularity theory.10 Neglecting
complications involving chance, he defined a law as any generalization in the
“best system”; that is, in the deductive system, stated in a language whose
predicates stand for natural properties and relations, that best balances the
virtues of strength and simplicity. A system is strong depending on how
much (in some suitable measure) information it entails; a system is simple
depending on how simply (in some suitable measure) it can be axiomatized.
Now, Lewis pursued this project under the assumption that naturalness was
primitive (or near enough). But the suggestion I want now to consider is that
we reductively define structural properties and relations in terms of goodness
of deductive systems and supervenience, as follows. Say that a deductive
system is based on a set of properties and relations iff each primitive
predicate in that system expresses a property or relation in the set. And say
that one set, S1, enables better laws than another set, S2, iff some system
based on S1 is better—i.e., better balances strength and simplicity—than any
system based on S2. (Perhaps the laws of the systems should also be required
to nontrivially utilize all the members of the sets.) With any luck, there exists
some minimally complete set that enables better laws than every other
minimally complete set. If so, structural properties and relations may be
defined as the members of this set.
This proposal is worth thinking about more, but the other objections given
above remain. The systemic objection remains: better to posit basic structure
than basic modality. And the first two of my (1996b) objections remain. The
first might be addressed by simply taking ‘qualitative’ as primitive, and the
second might be addressed by claiming that neither earlier-than nor later-
than is structural (those who deny that time has an intrinsic direction will
already want to say this). But other objections in the vicinity of the second
objection remain. Since supervenience is defined modally, mathematical
properties and relations will supervene trivially on any set whatsoever, given
the common dogma that mathematical facts are necessary. Thus, these
properties and relations will not be present in any minimally complete set.
But our conception of structure would be impoverished by their exclusion.
The distribution of structural properties and relations is supposed to give the
fundamental facts about the world, and we might well want to say that the
fundamental facts include mathematical facts.11
We have considered reductions to putatively fundamental law, cause, and
necessity. What of reduction to one of these notions construed
nonfundamentally? The systematic objections, anyway, would be avoided.
But the reduction would be circular if the nonfundamental notion were in turn
reduced to structure. And I suspect it would need to be. (Consider, for
example, Lewis’s account of law and cause, and my account of necessity in
chapter 12.)

2.5 Against subjectivity

Primitivism would also be unnecessary if structure were tied to us in some
way—to human language, biology, history, or psychology. According to this
view, what distinguishes the class of the electrons, as opposed to the class of
the electron-or-cows, is just that humans have a simple predicate for the
former class, find it psychologically more natural to think in terms of
‘electron’, and so on; it doesn’t go any deeper than that. This kind of
subjectivism about structure could be taken as a reduction or, alternatively, as
expressivism or some other form of noncognitivism.
Speaking just for myself, this is incredible. It is really, really hard to
believe that the fact that electrons go together, in a way that electrons-or-
cows do not, is merely a reflection of something about us.
But this is autobiography, not argument. The argument here, such as it is,
is that any subjectivity in the notion of structure would infect all the domains
in which structure is applied. If structure is just a reflection of our language
(or whatever) then so are the facts about similarity intrinsicality, laws of
nature, the intrinsic structure of space and time… And this is incredible.
At its last step the argument again reverts to autobiography. Certain
philosophers will rightly remain unconvinced, for example “antirealists” of
various stripes—pragmatists, Kantians, logical positivists, and so on.
A certain “knee-jerk realism” is an unargued presupposition of this book.
Knee-jerk realism is a vague picture rather than a precise thesis. According to
the picture, the point of human inquiry—or a very large chunk of it anyway, a
chunk that includes physics—is to conform itself to the world, rather than to
make the world. The world is “out there”, and our job is to wrap our minds
around it. This picture is perhaps my deepest philosophical conviction. I’ve
never questioned it; giving it up would require a reboot too extreme to
contemplate; and I have no idea how I’d try to convince someone who didn’t
share it.

2.6 The privilege of the physical

A final consideration in favor of primitivism about structure may be
advanced. It is based on knee-jerk realism.
Let Г be the set of true sentences in the language of completed physics,
and consider two sets of propositions. The first set, P, is the set of
propositions expressed by the members of Г, under their intended
interpretation. Thus P is the set of physical truths. The second set, S, consists
of “scrambled” propositions. To arrive at S, reinterpret all nonlogical symbols
of the language of physics under some arbitrary permutation µ of the totality
of objects (see section 3.2 for a full description of this sort of
reinterpretation), and let S be the propositions expressed by the members of
Г.thus reinterpreted. The members of S are all true, since the members of Г
are true under their intended interpretation, and reinterpretation under a
permutation preserves truth (again see section 3.2).
The consideration is then this. Even though both P and S consist of true
propositions, knee-jerk realism requires recognizing that there is something
better about P. Believing the members of P would be better than believing
the members of S; P constitutes a better description of reality than S. To deny
these things would be to admit that there’s nothing mandatory about physics,
that other perspectives on the world are “just different”, not worse. Knee-jerk
realism is incompatible with the thought that it’s just optional to think in
physical terms, that it would be just as good to pick wholly arbitrary carvings
of the world (meanings under µ) and think in those terms.
Knee-jerk realism further requires that the betterness be objective. It isn’t
merely that the propositions in P have a simpler description in our language,
or are more useful if one is trying to stay alive, let alone build rocket ships
(though this usefulness is good evidence for the betterness.) The betterness, it
is natural to conclude, is that only the propositions in P are cast in joint-
carving terms.
Are there less metaphysical terms in which one might characterize the
inferiority of S? It’s hard to see what they might be. For example, P is not
more “complete” than S in any interesting sense. Neither set contains all the
true propositions. Granted, S is silent on questions of physics, but P is silent
on questions of “shmysics” (the subject matter of S); what we need is some
objective sense in which the first omission is worse.
At this point an opponent might try out some of the moves of section 2.4.
She might say that P is better because it contains laws, or because it is
associated with a complete set of properties and relations (“associated” in the
sense that P’s members are expressed by simple sentences in a language with
predicates for the members of the complete set; “complete” in the modal
sense introduced in section 2.4). Systematic reasons against appealing to
modality—whether nomic or metaphysical—were given; but also, the appeals
are intuitively off-target. The deficiency of S, surely, inheres in its
categorical, real-world, nature, and should be explicable without bringing in
the ghosts of what might have been.
Knee-jerk realism, then, requires that the physical description of reality be
objectively privileged. And a natural account of the privilege is that physical
notions carve at the joints.

1 Lewis defined “duplicates”—intuitively, perfectly intrinsically similar objects—as pairs of objects

whose parts can be mapped one-to-one preserving parthood and perfectly natural properties and
relations; and he then defined intrinsic properties as those that can never differ between a pair of
duplicate objects, whether in the same or in different possible worlds.
2 This vague claim raises difficult issues. The intuitive conception of structure as involving
fundamentality doesn’t seem inert; if the entire theory of this book were replaced with its ramsey
sentence, omitting all mention of fundamentality, something would seem to be lost. But in what does
this intuitive conception consist? Is it simply further principles connecting fundamentality to other
notions? Or does it somehow resist being captured in principles?
3 See Maudlin (1996, pp. 131–3).
4 See, for example, Quine (1976a).
 5 See Oliver (1996, section 2).
6 See section 3.1 on laws and chapter 12 on modality.
7 This literature, it seems to me, needs to make up its mind whether it is about fundamental
metaphysics or conceptual analysis. (Hall (2006) and Lewis (2004) are notable for being clear which
they are pursuing—the latter.) Above all, it is important not to shift uncritically between the two
conceptions of the project, since they call for different methodologies. If the project is conceptual
analysis, then heavy reliance on thought experiments is appropriate; but it is far less clear that one can
insist on premises like: absences cannot be causes; abstract entities like facts cannot be causes; there
must be uniform causal relata; causation cannot be contextual or normative or extrinsic. One could
claim that these premises are part of our ordinary concept; but conceptual analysts generally regard
intuitive judgments about particular cases as being far more diagnostic than intuitive judgments about
such general principles. At the very least, one should evaluate these premises as alleged components of
our concept, and not as if they are supported by considerations of simplicity or fit with a favored
fundamental metaphysics. If, on the other hand, the project is to investigate the fundamental
metaphysics of the causal relation, then it must be argued that there is such a thing—that there is a
fundamental relation of causation; and heavy reliance on thought experiments must be abandoned, or
else somehow justified. See also Paul (2009, section 1; 2010, sections 1–3).
8 To say that some properties supervene on others is to say that in some sense, the former properties
cannot vary independently of the latter properties. Supervenience is normally defined in terms of
possible worlds and individuals, which are then defined in terms of necessity (and ancillary notions).
For an overview of these issues, see McLaughlin and Bennett (2005) on supervenience and Divers
(2002) on possible worlds.
9 The proposal was there taken as a definition of Lewisian perfect naturalness. See also Lewis
(1983a); Marshall (2009).
10 See Lewis (1973b, pp. 73–4; 1983b, pp. 366–8; 1986c, pp. 121–4; 1994).
11 See also section 10.2.
3 Connections
… as I bear [the distinction between natural and unnatural properties] in mind
considering various topics in philosophy, I notice time and again that it offers
solutions to my problems.—David Lewis (1983b, p. 343)

Structure is central in and around metaphysics. We saw in chapter 1 how it

connects to similarity. The next three chapters explore further connections.
Each connection is a proposal, stated in terms of structure, for how to
think about a certain topic. Although none of these proposals is mandatory
for the believer in structure, each is natural and attractive.
This chapter discusses explanation and laws, metasemantics, induction,
and physical geometry. These connections (especially induction and physical
geometry) will play only a peripheral role in the rest of the book. The
following two chapters discuss more central connections. Chapter 4 connects
structure to questions about the status (substantivity, conventionality,
objectivity) of disputes in general, and chapter 5 connects structure to
questions about the status of metaphysical disputes in particular—to
metametaphysics.

3.1 Explanation and laws

Many connect laws of nature with fundamentality, in one way or another.
Primitivists about lawhood sometimes define fundamental properties as those
that are involved in the laws, or else claim that “fundamental properties are
those that are involved in the laws” is a substantive principle connecting two
primitive notions. Conversely, a primitivist about fundamentality—in my
case, about structure—might define laws in terms of structure.
The Lewisian approach to laws of nature is an example of the final option.
As we saw in section 2.4, Lewis defined laws as generalizations in the best
system—the deductive system, cast in a language whose predicates express
natural properties and relations, that best balances the virtues of simplicity
and strength. The restriction on the language of the best system is essential;
otherwise, as Lewis (1983b, p. 367) points out, a simple and maximally
strong theory could be given with a single, simple axiom, ∀xFx, where F is a
predicate true of all and only things in the actual world. All true
generalizations would be counted as laws.
The metaphysical core of Lewis’s theory is its Humean, negative part: the
rejection of metaphysically fundamental laws of nature. Fundamentally
speaking, the world is anomic. The best reason to accept this Humean core is
parsimony.1 We ought, other things being equal, to keep fundamental
theories simple; and metaphysically fundamental laws would add complexity
without adding to understanding. The claims of physics aren’t explanatorily
enhanced by adding that those claims are laws.2
The positive part of Lewis’s theory is its definition of lawhood. This part
is attractive, even more so if it is augmented as follows. Adding more content
to a system generally makes it stronger but also more complex. Since the best
system must balance strength and simplicity, additions are justified only if
the benefit of the added strength outweighs the demerit of the added
complexity. Question: how much complexity can be tolerated to gain a given
amount of strength? That is, how much does complexity “cost”? Different
answers correspond to different notions of law. The more expensive
complexity is made, the simpler the best system will need to be, thus making
fewer generalizations count as laws. Special-science generalizations, for
example, are more complex than those of physics when they involve notions
that carve at the joints less well than do physical notions, and when they
include ceteris paribus conditions (whatever that means exactly, it surely
detracts from simplicity). So let a “middling” assignment of cost to
complexity be one that counts the generalizations of physics as laws, but only
barely. This corresponds to a sense of ‘law’ in which there are laws of
physics, but in which certain special-science generalizations do not count as
laws.3 If complexity is instead made cheaper, then those special-science
generalizations will count as laws. (And if complexity is made expensive, so
that very little complexity is tolerated, then even physical generalizations no
longer count as laws; all that remain are laws of metaphysics and logic. See
sections 10.3 and 12.5.)
Once augmented, Lewis’s theory of law is good as far as it goes. But I
doubt that the notion of law is quite as central as philosophers (or
metaphysicians, anyway!) think. Consider these claims: spacetime is a four-
dimensional Lorentzian manifold; the universe began with an initial
singularity; the universe began in a state of very low entropy. It’s a stretch to
call these laws, yet they’re perfectly central to physics. If the point of the
notion of law is its role in a rational reconstruction of science, then we should
broaden our focus, to include these non-laws as well.4 And the centrality of
laws arguably diminishes once we move beyond physics to the special
sciences. It has been argued, for example, that there are no laws of biology.5
The core insight of Lewis’s account of laws can be generalized beyond the
case of laws: good scientific theories, whether or not they cite laws, must be
cast in joint-carving terms. We may put this in terms of explanation:
“theories” based on bizarre, non-joint-carving classifications are
unexplanatory even when true.6 Theories whose basic notions fail badly to
carve at the joints fail badly as theories, even if they are exemplary from an
“internal” point of view, for their inner workings fail to mirror the inner
workings of the world. We know on cardinality grounds that there are
functions from the motions of the planets to past stock market performance
under which the motions correspond to the fluctuations of the Dow Jones
industrial average to date. But if someone were actually to produce such a
function, no one would regard it as being explanatory (and no one would
expect the correlation to continue). In contrast, explanations that cite facts
about the geometry of spacetime or the initial singularity are genuine
(whether or not the cited facts are laws), in part because the cited facts are
stated in joint-carving terms.
This dimension of theoretical excellence is best exemplified by theories
cast in perfectly fundamental terms—theories of fundamental physics, for
example. But it is exemplified, to varying degrees, by special-science theories
as well, since the notions of the special sciences carve at the joints reasonably
well.

3.2 Reference magnetism

One of the “problems” Lewis used his notion of naturalness to solve was the
problem of radical semantic skepticism (1983b; 1984). The problem is one in
metasemantics. How do words (or thoughts—but let’s stick to words) get
their meanings? What “semantic glue” attaches them to the world? There are
different views about the nature of the semantic glue, but on nearly all of
them, the glue doesn’t seem to be sticky enough; it apparently cannot secure
meaning with sufficient determinacy. Most roughly put: what I mean by ‘pig’
is surely determined by such facts as that I’ve always said ‘pig’ when in the
presence of pigs; but why do such facts determine that by ‘pig’ I mean pigs,
rather than pigs-I’ve-encountered-in-the-past, or pigs-in-my-immediate-
vicinity, or pigs-before-2011-A.D.-or-cows-afterwards or … ?7

FIG. 3.1. Interpretation 1

To bring out the problem more fully, we must consider concrete proposals
for the nature of the glue. Following Lewis (1984), I’ll illustrate how the
problem goes for a toy descriptivist theory of the nature of semantic glue.
According to this descriptivism, there is a set of sentences, S, such that our
words mean whatever they must in order for the sentences in S to come out
true. (This theory is schematic; different versions specify different sets S. S
might, for example, be taken to include “definitional sentences”, whatever
that means exactly.)
The sentences in S do not, on their own, provide enough semantic glue.
Let (F) be an intuitively false sentence of our language that is logically
consistent with S—“Some pigs have wings”, say. We had better be able to
say that (F) is false (otherwise, as Fodor would say, it’s the end of the world).
Now, (F) turns out false in an interpretation iff nothing is both in the
extension of ‘pig’ and also in the extension of ‘has wings’ in that
interpretation (figure 3.1). And (F) turns out true in an interpretation iff these
extensions overlap in that interpretation (figure 3.2). So if (F) is to turn out
false, the correct interpretation of our language must be of the former sort;
interpretations like interpretation 2 are incorrect—such interpretations do not
reflect what we really mean.8 But if—as the descriptivist says—all that is
required of a correct interpretation is that the sentences in S come out true
under that interpretation, then we are pretty much guaranteed to be able to
construct a correct interpretation like interpretation 2. All we need to do is
assign extensions to predicates so that every sentence in S, plus (F) as well,
turns out true. We might, for instance, begin by assigning the set of hard-
boiled eggs to ‘pig’ and assigning the set of edible things to ‘has wings’. This
makes (F) true since some hard-boiled eggs are edible. Now, suppose (S)
contains the sentence ‘Every pig is an animal’. The interpretation we are
constructing must count this sentence true as well. But this is easy to
accomplish: simply assign the set of eggs

FIG. 3.2. Interpretation 2

to ‘is an animal’. Since every hard-boiled egg is an egg, ‘Every pig is an

animal’ comes out true: the bizarre assignment to ‘pig’ is “cancelled out” by
the equally bizarre but compensating assignment to ‘is an animal’. We then
continue in this way, assigning compensating bizarre extensions so that every
other member of S comes out true as well.
The extensions assigned by this bizarre interpretation to ‘pig’, ‘has
wings’, ‘animal’, and other predicates are clearly not what those predicates
really mean, by any ordinary standard of “really mean”. But our descriptivist
says that to be correct (i.e., to reflect what we really mean), an interpretation
need only make every member of S come out true. So the bizarre
interpretation 2 would count as a correct one, and so we could not say that
‘Some pigs have wings’ is false! Something has plainly gone wrong, and it is
pretty clear what that is. Descriptivism is false as stated; there must be more
to the correctness of an interpretation than merely making certain specified
sentences come out true.
The argument assumes that it is possible to continue selecting meanings
for the nonlogical expressions in our language so that every member of S (and
(F) as well) turns out true. But this will always be possible, except in special
cases. Let C be the class of entities that we are using to interpret our words,
and consider any abstract model—in the logician’s sense—in which (F) as
well as all members of S are true, and in which the domain is no larger than
C. (The only cases in which there would not exist such a model are: i) if (F)
is inconsistent with S; or ii) if S and (F) are consistent but only have models
that are larger than C. And we lose no important generality by stipulating that
the sentences S are consistent with (F) and do not, in concert with (F),
logically require the universe to be larger than C.9) Since the domain of this
abstract model is no larger than C, it can be mapped by some one-to-one
function µ onto some subset, D, of C, as in figure 3.3. But
 FIG. 3.3. Abstract domain mapped one-to-one onto D

then we can use µ to construct our desired interpretation 2 in D. Each

nonlogical symbol in our subject’s language has a meaning, m, in the abstract
model; let the meaning of each such symbol in interpretation 2 be the object
or set of ’tuples µ[m] in D to which m corresponds under the mapping. For
example, ‘pig’ has an extension, E, in the abstract model, so the meaning of
‘pig’ in interpretation 2 will be the set µ[E] of members of D to which
members of E are mapped by µ (figure 3.4). Now, it is an elementary fact
from model theory that if a sentence is true in a model, it is true in any model
constructed via a one-to-one mapping as above. So, since all the members of
S, plus (F), are true in the abstract model, they are true in interpretation 2 as
well.10
The argument shows that a correct interpretation must do more than make
specified sentences come out true. What more? Lewis’s answer is that correct
interpretations must, as much as possible, assign natural properties and
relations (or their extensions) to predicates. The meanings under
interpretation 1 are (or correspond to) at least somewhat natural properties:
being a pig, being edible, and so on. But there is no guarantee that the
meanings assigned by interpretation 2 are natural to any significant degree,
since they were constructed from the arbitrarily chosen function µ.
Lewis’s proposal must be distinguished from the proposal that the
sentence ‘Predicates stand for natural properties and relations’ is to be
included in the set, S, of meaning-determining sentences. This proposal
would not solve the problem. For so long as this sentence is consistent with
(F) and the rest of the sentences in S,

FIG. 3.4. Interpretation 2 induced by µ.

we will again be able to construct our interpretation 2. The resulting

interpretation will misinterpret ‘predicate’, ‘stand for’, ‘natural’, ‘property’,
and ‘relation’ so that ‘Predicates stand for natural properties and relations’,
like the other sentences in S, come out true. Lewis’s idea is not to stick with
the original descriptivist constraint on correct interpretations (“correct
interpretations are those that make every sentence in S true”) and add more
sentences to S; it is rather to modify descriptivism by adding an additional
constraint. That constraint is not that ‘Predicates stand for natural properties
and relations’ must come out true on a correct interpretation; it is rather, and
more simply, that predicates must stand for natural properties and relations
in a correct interpretation.
Lewis’s constraint on reference is “externalist”; reference is not
determined merely by us. We, of course, do part of the work. Our patterns of
use of language determine which sentences go into S. But the degree to which
an interpretation counts sentences in S as being true is just one factor in
determining whether that interpretation is correct. The other factor is the
naturalness of that interpretation’s semantic values. And this second factor
has nothing to do with us. Natural properties and relations are “intrinsically
eligible meanings” they are “reference magnets”.
The doctrine of reference magnetism has been received warily. Some
regard it as an ad hoc response to a problem, with no independent backing.
Others regard it as occult metaphysics, as the postulation of an irreducible
“semantic force”. (Perhaps the term ‘reference magnetism’, never used by
Lewis but increasingly the doctrine’s customary name, encourages this
thought.11) Neither charge is founded. Following J. Robert G. Williams
(2007, section 2), we can derive the doctrine of reference magnetism from a
well-motivated and more general doctrine about theoretical virtue. This
doctrine is the one defended in section 3.1: explanatory theories must be cast
in joint-carving terms.
As I will develop it, the crucial assumption of the derivation is that
reference is an explanatory relation—one can explain certain facts by citing
what words refer to. But if reference were given by a bizarre interpretation,
then reference-involving “explanations” would not in fact be explanatory,
since they would be cast in badly non-joint-carving terms. Hence reference is
not given by a bizarre interpretation.
Thus the exclusion of bizarre interpretations is of a piece with something
we already do. We already regard “theories” based on bizarre classifications
as being explanatorily useless; reference magnetism is just the extension of
this point to metasemantics. Attempting to explain a community’s linguistic
behavior by citing a relation based on the permuted interpretation 2 would be
like attempting to explain the behavior of the stock market in terms of an
arbitrary correlation with the motions of the planets.
Let’s look at the derivation of reference magnetism in more detail.
Reference is a theoretical concept of a certain special science: semantics. A
special science, quite generally, attempts to explain a certain target domain of
facts by means of certain theoretical concepts. In the case of semantics the
target domain is certain aspects of human thought, behavior, and
communication, and the theoretical concepts include reference and truth. To
use an example of Vann McGee’s (2005, section 4), suppose a high-school
teacher writes on the chalkboard the sentence ‘Maiasaurs were highly social
animals that traveled in herds of as many as 10,000.’ Why did precisely that
pattern of marks appear on the chalkboard? It clearly has something to do
with the connection between those marks and maiasaurs. In particular, it
surely involves the fact that the marks constitute words that refer to
maiasaurs and their properties.
Some philosophers reject reference-based explanation. This is a big issue,
not to be resolved here. But two small points. First, some claim that
conventional meaning encodes far fewer referential properties than
mainstream semantics thinks, but even they usually grant some explanatory
role to reference—or at least to reference-by-a-speaker.12 Second, granting
the existence of deflationary concepts of truth and reference that cannot take
part in explanations of the sort described above is compatible with also
accepting nondeflationary concepts of truth and reference that can take part in
those explanations.13 At any rate, I will assume that some reference-based
explanations are indeed explanatory.
If reference-based explanations are to be explanatory, then the reference
relation must be a joint-carving one. Not a perfectly joint-carving relation,
presumably; but at the very least, not a wildly non-joint-carving relation. And
this excludes the bizarre interpretations. For only a wildly non-joint-carving
relation would relate a linguistic population to the semantic values of a
bizarre interpretation.
Caveat: suppose there were a perfectly joint-carving reference relation.
Suppose, that is, that part of the absolutely fundamental story of the world
included a specification of a relation of reference between certain sounds or
inscriptions made by human beings and parts of the world. Such a
fundamental relation could perhaps, as a brute fact, relate words to bizarre
semantic values.14
But reference is surely not metaphysically fundamental.15 As a result, it’s
hard to see how a reasonably joint-carving reference relation could relate us
to the bizarre semantic values. The point may be seen initially by making two
strong, crude assumptions about “reasonably joint-carving”. Assume first that
a notion is reasonably joint-carving iff it has a reasonably simple and
nondisjunctive definition in terms of the perfectly joint-carving notions, and
second that the perfectly joint-carving notions are those of physics. Then
surely no reasonably joint-carving relation that is to play the role of reference
could relate a human population to bizarre semantic values. For the bizarre
semantic values themselves have no simple basis in the physical, nor do they
stand in any physically simple relations to human populations. Given any
relation that does relate us to bizarre semantic values, there is surely some
other relation with a simpler basis in the physical that relates us to nonbizarre
semantic values.
The two assumptions of the previous paragraph are undoubtedly too
crude, but the point is independent of them. Whether a notion is reasonably
joint-carving—enough to take part in special-science explanations—has
something to do with how it is based in the fundamental. So reference must
have the right sort of basis in the fundamental if it’s to be explanatory. It’s
highly unclear what exactly the “right sort” of basis is (this is in essence the
question of the relation between special-science concepts and physics), but
it’s quite clear that a relation connecting us to bizarre semantic values would
have the wrong sort of basis—for the same reason that arbitrary correlations
between the motions of the planets and the stock market have the wrong sort
of basis.
Conceiving of reference magnetism in Williams’s way risks restricting its
scope. In addition to replying to the Putnamian semantic skeptic we discussed
earlier, Lewis also used reference magnetism to reply to Kripke’s (1982)
Wittgensteinian skeptic about the meaning of mathematical language, who
asks why by ‘plus’ we mean plus rather than quus, where quus is a function
like plus except that it assigns to all pairs of sufficiently large numbers the
value 5. Lewis’s answer is that plus is a more natural function than quus; this
becomes, in Williams’s hands, the claim that a semantic theory assigning
quus as a semantic value is less explanatory. But now, some will claim that
abstract entities cannot take part in genuine explanations, perhaps because
abstracta are causally inert. For myself, I reject this conception of
explanation. Explanations come in many sorts, not all of them of the
“pushing-and-pulling” variety. But at any rate, the constraint on reference
magnetism’s scope here seems like welcome discipline, not unwanted
restriction.
The problem of radical semantic skepticism was presented above as it
confronts descriptivist metasemantics. But other metasemantic theories face
the problem as well, and reference magnetism—understood in Williams’s
way—can defend these other metasemantic theories as well, so long as they
conceive of reference as an explanatory relation.
For example, Ruth Millikan (1989), who thinks of reference as a
biological phenomenon, also faces the problem. Why does the true reference
relation, rather than numerous gerrymandered alternatives we could
construct, count as the relation that has been selected by evolution for the
purpose of storing and communicating information? Answer: any hypothesis
which says that a gerrymandered relation plays a role in an evolutionary
explanation is to be rejected simply because the relation is gerrymandered.
The correct answer on behalf of Millikan to the skeptical worry is thus akin
to any biologist’s answer to the skeptical worry that it is facts about
gruegenes, rather than genes, that explain inheritance. (The array of
gruegenetic properties a thing has is determined by the physical makeup of
the spatiotemporally nearest pair of bluejeans; on cardinality grounds, there
are sufficiently many different gruegenetic properties to enable us to concoct
“laws” of gruegenetics which really do correlate gruegenetic properties with
inheritance as fine-grainedly as the real laws of genetics do with real genetic
properties.) Similar remarks apply to causal theorists like Jerry Fodor (1987)
(unless causation itself is taken to solve the problem16).17
When understood in the Williamsian way, we might think of reference
magnetism as a thesis of meta-metasemantics, rather than a metasemantic
theory in its own right: reference magnetism may be combined with any
metasemantics you like. A metasemantics is a metaphysical proposal about
the nature of the reference relation; schematically: “reference is a relation of
such-and-such a type”. Millikan holds that reference is a relation that was
chosen by natural selection to achieve a certain goal; the descriptivist says
that reference is a relation that assigns values under which sentences in S
—“definitional” sentences, let us say—come out true; and so on. Each
proposal faces the problem of semantic skepticism: there are many relations
of the proposed type. How to cut down on this multiplicity? The answer of
reference magnetism is to reinterpret a metasemantic proposal that reference
is a relation of type T as the proposal that reference is a reasonably joint-
carving relation of type T.
One way a relation could fit one of the modified proposals is for it to be
defined in terms of the concept of joint-carving. (Assuming, that is, that the
concept of joint-carving itself is reasonably joint-carving—see sections
7.11.1 and 7.13.) For example, the modified descriptivist proposal says that
reference is a reasonably joint-carving relation that assigns values under
which definitional sentences come out true; here is a relation that fits that
proposal: R(w, x) =df w is a word and x is the semantic value assigned to w by
the interpretation I that maximizes the combination of i) joint-carvingness of
assigned semantic values; and ii) truth assigned to the sentences that are
definitional in the linguistic population that uses w. But this is not the only
way that a relation could fit one of the modified proposals: the relation could
be defined in terms of reasonably joint-carving concepts other than the
concept of joint-carving itself.
If reference magnetism is merely a thesis of meta-metasemantics, it can
survive the demise of individual metasemantic theories based on it. Consider
again descriptivism. To be more concrete, consider what we might call
“simple charity-based descriptivism”, a close cousin of the reference-
magnetism-enhanced descriptivism discussed in the previous paragraph.
According to simple charity-based descriptivism, the correct interpretation of
a language is that interpretation that maximizes the combination of two
factors, each of which comes in degrees: “eligibility”—determined by the
degree of naturalness of the semantic values it assigns; and “charity to use”—
the number (on some suitable measure) of sentences believed (or reasonably
believed, or …) by the speakers of the linguistic community that come out
true under that interpretation.18 Since it’s the combination of eligibility and
charity that must be maximized, this theory implies that an imperfectly
charitable interpretation can nevertheless be the correct interpretation, if it
assigns sufficiently joint-carving properties. A highly eligible interpretation
can “trump” the superior charity of rival interpretations. Now, this prediction
is in some cases correct, especially for “theoretical” terms—terms that are,
intuitively, intended to stand for joint-carving meanings. When a term like
‘mass’ is introduced in physics, it’s intended to stand for a fundamental
physical magnitude, and so if there’s a joint-carving property in the vicinity
then that property is meant by ‘mass’, even if it doesn’t quite fit the
physicists’ theory of ‘mass’. But for nontheoretical terms, the prediction of
trumping sometimes seems wrong.
For example, imagine a linguistic community that uses the word ‘amulet’
in the same way we do, except that it just so happens that—with a single
exception—the objects they call ‘amulets’ are all and only the instances of
some highly joint-carving property—gold, say.19 The single exception is that
they call one silver ornament an amulet as well. Now, the property gold
carves very well at the joints—much more so than the property gold or silver,
or the property ornament or small piece of jewelry thought to give protection
against evil, danger, or disease, to quote the dictionary. Moreover, nearly
everything the community says about “amulets” comes out true if ‘amulet’ is
assigned the property gold. So such an assignment seems to best combine the
virtues of eligibility and charity; its superior eligibility trumps the superior
charity of interpretations that also count the silver ornament as an “amulet”.
The simplistic account thus seems to imply that ‘amulet’ means gold in this
linguistic community. But this is absurd. The silver ornament obviously
counts as an “amulet”; the fact that it lacks a joint-carving property shared by
all other objects called ‘amulets’ is irrelevant.20
Now, this particular example is easily addressed with minor refinements.
For example, charity to use might be strengthened to require counterfactual
robustness: charitable interpretations must make sentences that would be
believed in certain counterfactual circumstances come out true in those
circumstances. Since the community in question is disposed to call
appropriate ornaments ‘amulets’ whether or not they’re made of gold, the
assignment of gold to ‘amulet’ no longer counts as charitable. But the
underlying point of the example cannot be so easily dismissed: the fact that
gold carves much better at the joints than being an ornament or small piece
of jewelry thought to give protection against evil, danger, or disease just
seems irrelevant to the question of which is meant by ‘amulet’. No matter
how “charity to use” is understood, examples of this sort will surely emerge.
Consider, for instance, this example from John Hawthorne: if there just
happens to be a physically special line through the Ural mountains, that line
isn’t thereby the determinate boundary of Europe. The candidacy of a
meaning for ‘Europe’ that takes this line into account is not enhanced by its
superior eligibility.
It would seem, then, that for nontheoretical terms, the joint-carvingness of
candidate semantic values plays a complex role. Highly joint-carving
candidates don’t automatically trump; simple charity-based descriptivism is
too simplistic. But on the other hand, joint-carving isn’t completely irrelevant
either: the gruesome candidates of the semantic skeptic must still be
excluded. Fortunately, even if we reject simple charity-based descriptivism,
we can still appeal to reference magnetism to exclude the gruesome
candidates. We can exclude them simply because they’re gruesome—simply
because they carve so badly at the joints. No reference relation connecting us
to the gruesome candidates could take part in an explanatory theory. And we
can say this even if we’re uncertain what the true metasemantic theory is. In
fact, we can say this even if we doubt that the truth about metasemantics can
be captured in any simple formula—not an overly pessimistic doubt, given
how few simple reductive theories of complex macro-phenomena have ever
been given.
 I have argued that reference magnetism solves the problem of radical
semantic undetermination, is a consequence of a more general claim that
explanatory theories must be stated in joint-carving terms, and does not
presuppose descriptivism, but rather may be combined with any
metasemantic theory that regards reference as explanatory. These conclusions
collectively provide a reason to recognize structure: we need structure to
answer the semantic skeptic. But there is a challenge to this reasoning: a
causal metasemantics also lets us answer the skeptic. According to causal
metasemantics, ‘pig’ means pigs rather hard-boiled eggs, say, because it is
pigs rather than hard-boiled eggs that cause our uses of ‘;pig’.21 See? No
need for reference magnetism!
But a pure causal theory is likely to be insufficiently general. We need to
rule out incorrect interpretations of mathematical and logical language, for
example—as put forward by Skolemite skeptics about set-theoretic language,
say, or Kripke’s Wittgenstein—just as we need to rule out incorrect
interpretations of ‘pig’ and ‘has wings’; and it is hard to see how a pure
causal account could do this. Likewise for terms of theoretical physics for
hypothesized properties that are instantiated only under conditions present
only at the time of the big bang, or even conditions that have never been (and
will never be) present at all.22 Likewise for predicates for spacetime structure.
Likewise for many of the concepts of philosophy, to be discussed in
subsequent chapters (though how much structure there is in these areas is
debatable). Reference magnetism, on the other hand, can be combined with a
broader, not purely causal, metasemantics to rule out bizarre interpretations in
these cases.23 This is not to say that causation has no role to play. It’s
perfectly compatible with reference magnetism, construed as a
metametasemantics, that causation is one ingredient of the semantic glue.
Further, the notion of causation required by causal metasemantics may
presuppose the notion of structure.24 To bring this out, consider an intuitively
incorrect interpretation of our language that matches the correct interpretation
up until some specified time, 3000 A.D., say, but then goes haywire
afterward. For the sake of definiteness, here is one such interpretation.
Indulge in the harmless simplifying assumption of “super-substantivalism”,
according to which the physical world consists purely of spacetime, and let v
be a one-to-one mapping over points of spacetime that maps each point at or
before (the first instant of) 3000 A.D. to itself, and maps each point after
3000 A.D. to the simultaneous point 1010 km in some one chosen direction.
Thus, v leaves everything before 3000 A.D. alone, and applies a Leibnizian
shift to everything afterward. Now, the intuitively correct interpretation of
our physical vocabulary assigns semantic values involving points of
spacetime to physical vocabulary. So v induces a “shifted” interpretation:
simply replace each point p in each semantic value with v(p). What makes the
intuitively correct interpretation rather than the shifted one correct? The
shifted interpretation is derived from the correct one by a one-to-one
mapping, and therefore renders the same sentences true. Furthermore, the
shifted semantic values are instantiated by exactly the same spacetime points
before the year 3000 A.D. as are the correct semantic values. The
interpretations diverge only in the far future. So on the face of it, the shifted
values seem to be just as causally responsible for past usage as the correct
ones, in which case the pure causal metasemantics fails.
Veterans of the literature on natural kinds will be quick to notice that
according to a “robust” conception of causation, the shifted semantic values
will not be causally responsible for past usage—indeed, not causally
responsible for anything at all. On this conception, even if a property P is
involved in relations of counterfactual dependence, relations of nomic
necessity or sufficiency, or what have you, it still may not be causally
efficacious. If I hold a grue emerald in front of you, you will have green
sensations; my holding the grue emerald plus background conditions
necessitates those sensations; if I hadn’t held up the grue emerald you
wouldn’t have had those sensations; and so on. (An object is “grue” iff it is
green and first observed before 3000 A.D. or blue and not first observed
before 3000 A.D. (Goodman, 1955, chapter III).) Nevertheless, so the story
goes, my holding up the grue emerald did not cause you to have the
sensations. Only my holding up of the green emerald caused you to do this.25
What we have learned is that causation must be robustly conceived, if it’s
to play the needed role in metasemantics. But how is robust causation to be
understood? A robust cause might be defined as a joint-carving property
involved in an event that causes something. Or instead as any property
involved in an event that causes something, where our theory of causation
invokes structure.26 Or instead as any property that figures in a law of nature,
where our theory of lawhood invokes structure.27 Each of these approaches
invokes the notion of structure. Only a primitivist, structure-free approach to
causation (or laws) yields a structure-free approach to metasemantics; and
against such approaches, recall section 2.4.
3.3 Induction and confirmation
We need structure to make sense of learning from experience.
The simplest model of learning from experience is that we remember past
experiences, we expect the future to be like the past, and so we form
appropriate expectations about the future. This model requires the notion of
structure to be plausible. For the philosophical skeptic of section 1.2 will be
quick to point out that any possible future is “like” the past along some
dimension of similarity. The model had better say that the future is like the
past in some genuine dimension, some dimension that respects nature’s
joints.
“The future is like the past” is too crude a concept on which to base a
theory of learning from experience. We do a little better with the following
concept: “observation o confirms sentence S”. Analogs of the above worries
about similarity then confront confirmation-based theories of learning from
experience.
Which observations confirm a generalization ‘All F s are Gs’? A natural
answer is the “Nicod principle”: observations of Fs that are Gs confirm ‘All
Fs are Gs’. But suppose that an observation confirms any logical equivalent
of any sentence that it confirms. Then, as Hempel (1945) pointed out, the
observation of red roses confirms ‘All ravens are black’ (given the Nicod
principle it confirms ‘All nonblack things are nonravens’, which is logically
equivalent to ‘All ravens are black’.) And as Goodman (1955, chapter III)
pointed out, Nicod’s principle implies that observations of green emeralds
before 3000 A.D. confirm ‘All emeralds are grue’ (since green emeralds
observed before 3000 A.D. are grue.) But anyone who believed that all
emeralds are grue would expect emeralds observed after 3000 A.D. to be
blue.
These conclusions can be avoided by restricting Nicod’s principle in some
way—most crudely, to predicates that carve at the joints.28 Since ‘is
nonblack’, ‘is a nonraven’, and ‘grue’ fail to carve at the joints, the restricted
principle does not apply to generalizations containing them. In Goodman’s
terminology, only terms that carve at the joints are “projectible”.
‘Observation o confirms sentence S’ is less crude than ‘the future is like
the past’, but it is still too crude a concept for a realistic theory of learning
from experience. A realistic theory will need to consider relative
confirmation, and so surely must consider quantitative measures of
confirmation, and so must surely employ the concepts of probability theory.29
It is natural to do this within a Bayesian framework, according to which i) a
rational subject’s beliefs at a time consist in a probability distribution over all
propositions, whose values measure the subject’s “degrees of belief”
(“subjective probabilities”, “credences”) in the propositions; and ii) the
rational subject updates her degrees of belief by conditionalizing on
propositions describing her experiences.30 In this framework one can
introduce various quantitative measures of confirmation, and can characterize
relative confirmation.31
Bayesianism proper tells the rational subject what method she should use
for updating whatever degrees of belief she began with: conditionalization.
But it says very little about what those initial degrees of belief ought to be
like (beyond the minimal demand that they must satisfy the axioms of the
probability calculus, and perhaps be “nondogmatic” in never assigning the
values 0 and 1 except to necessary falsehoods and truths). Now, subjective
Bayesians say that Bayesianism proper is all there is to rationality. That is,
provided a subject has updated her beliefs by conditionalization (and at each
moment obeyed the minimal synchronic demands), she has done all that
rationality requires. Objective Bayesians, on the other hand, add further
synchronic requirements on credences beyond the minimal ones. One’s
credences must, perhaps, assign higher probabilities to “simple” hypotheses,
other things being equal, or obey some sort of principle of indifference,
dividing credence equally over symmetric possibilities.
This difference between objective and subjective Bayesianism matters
greatly, since the effect that evidence has on a Bayesian conditionalizer
depends heavily on her prior probability distribution. Bizarre prior
probability distributions will result in bizarre responses to evidence.
Consider, for example, making a series of pre-3000 A.D. observations of
green emeralds. Intuitively, this should result in increasing confidence that
emeralds observed after 3000 A.D. will likewise be green. This increasing
confidence is indeed forthcoming for a Bayesian if she begins with an
appropriate prior probability function Pr—one that assigns high probability to
emeralds observed after 3000 A.D. being green conditional on earlier
observed emeralds being green. But if she begins instead with a prior
probability function Pr′ that assigns high probabilities to emeralds observed
after 3000 A.D. being blue conditional on earlier observed emeralds being
green, then observing the green emeralds will result in increasing confidence
that emeralds observed after 3000 A.D. will be blue. Garbage in, garbage out.
The pure subjective Bayesian will say that rationality says nothing about
the choice between Pr and Pr′. If one person begins with Pr, and another
begins with Pr′, and each then receives exactly the same evidence, which
evidence includes observations before 3000 A.D. of many green emeralds
(and no blue ones), and each updates her beliefs by conditionalization, then at
the end of this process, neither has behaved more rationally than the other.
This despite the fact that the second will become increasingly certain that
emeralds observed after 3000 A.D. will be blue!
Subjective Bayesians embrace this conclusion. Fascinatingly, this
descendent of Hume’s notorious attitude toward induction is not uncommon
in contemporary formal epistemology. Is this because the field draws the
formally inclined, and the problem of constraining priors has proved formally
intractable? At any rate, any Bayesian who wants to say that the second
person has responded irrationally to the evidence must find fault with Pr′.
(Since the subject used the correct rule of updating, the fault must lie with her
prior degrees of belief.)
How might Pr′ be criticized? One (vague, in need of refinement) strategy
would invoke “simplicity”: Pr′ is worse than Pr because its description is less
simple. The probabilities of Pr that concern emerald color can be given by
simple rules that are uniform across all emeralds, such as: “for any emerald,
the probability of that emerald being green, conditional on many other
emeralds being green, is high”. But to specify Pr′ we need more complex
rules giving different conditional probabilities for an emerald’s being green,
depending on the times at which earlier green emeralds were observed.
But suppose we speak the language of ‘grue’ rather than the language of
‘green’.32 In this language, Pr′ can be specified with simpler rules, such as
“for any emerald, the probability of that emerald being grue, conditional on
many other emeralds being grue, is high”, whereas Pr will require more
complex rules, specifying different conditional probabilities for an emerald’s
being grue, depending on the times at which earlier grue emeralds were
observed. Whether a probability function is simple depends on the language
in which it is described. So a meaningful simplicity-based criticism of Pr′
requires some sort of restriction on the language in which we evaluate
simplicity.
Similar remarks apply to attempts to constrain priors using some version
of the principle of indifference. Any version of this principle says to
distribute credence equally over “symmetric” possibilities. But which
possibilities count as symmetric will depend on what language we use to
describe the possibilities. If we use the ‘grue’ language, an otherwise
reasonable principle of indifference might recommend Pr′ over Pr. Now, even
with a suitable language picked out, there are serious obstacles. To take a
familiar example, suppose a factory is making cubes of varying sizes. If I
have no information specific to a given cube produced by the factory, how
should I assign prior probabilities to propositions of the form the side length
of the cube is between l1 and l2? In direct proportion to the difference l1 − l2,
one wants to say. And how should I assign prior probabilities to propositions
of the form the face area of the cube is between a1 and a2? In direct
proportion to the difference a1 –a2, one wants to say. But these two answers
are incompatible (van Fraassen, 1989, 303–4). So selecting the right language
doesn’t, on its own, solve the problem of formulating principles of
indifference. But without an appropriate language, we cannot even get started
on a solution.
To constrain prior probability distributions, then, we need some way to
pick out appropriate languages for evaluating simplicity, symmetry, and
related notions. And—to finally get to the point—it seems reasonable to pick
them out by using the notion of structure. Now, even given the notion of
structure, there are nontrivial questions about how exactly to pick out the
appropriate languages. For example, how well do the predicates in the
appropriate languages need to carve at the joints? Again, the appeal to
structure is the beginning of a solution, not the end of one.
The argument of this section has been that structure fills a need in
epistemology. A reply would be that epistemology does not demand structure
objectively construed; a conception of “structure” tied to human history,
biology, psychology, or interests would do. My reply to such challenges
elsewhere is that subjectivity in structure would infect all notions to which
structure is connected (similarity, intrinsicality, duplication, laws of nature,
and so on). But in this case, the infected notion would be epistemic value.
And perhaps we should embrace the idea that values in general are not
objective.
But even if epistemic value is subjective along some dimensions, we
shouldn’t embrace the idea that it’s subjective along all dimensions. Let Pr be
a rational credence function we ought to adopt and Pr′ be one that we ought
not to adopt. Intuitively speaking, we might embrace subjectivity in the
“oughtiness” of the obligation while rejecting subjectivity in the distinction
between Pr and Pr′. The objective facts might not mandate that we have our
notion, or any notion, of epistemic obligation, but might nevertheless
mandate that we choose Pr over Pr′ if we do have our notion (or anything like
our notion). We will return to this issue in section 4.2.

3.4 Intrinsic structure in physical spaces

We need structure to understand talk in physics of the “intrinsic” structure of
space, time, spacetime, and other spaces.33
NonEuclidean geometries were discovered in the early nineteenth century,
proved consistent relative to Euclidean geometry later that century, and
applied in physics by Einstein in the early twentieth century, in his general
theory of relativity. Taken at face value, Einstein’s claim that physical
spacetime is curved is a substantive claim in direct conflict with the
assumption of flat spacetime implicit in classical Newtonian physics (and in
the special theory of relativity as well). But taken at face value, this claim
raises various philosophical questions: epistemic, semantic, and
metaphysical.
Let us approach the questions through the simpler case of spatial, rather
than spacetime, curvature. Imagine a series of experiments, carried out with
rigid measuring rods and the like, that apparently show that space in a certain
region is curved. The epistemic questions arise because we do not observe
space directly; what we observe is things in space, such as measuring rods.
Effects attributable to curved space could instead result from systematic
distortions to the rods. While this may at first appear to be mere Cartesian
demonry, compare two alternative hypotheses. According to the first
hypothesis, the measurements result from curvature. According to the second,
space is flat but there are “universal forces” that affect all matter, cannot be
blocked, and systematically shrink and expand the measuring rods so that
their lengths are exactly as the first hypothesis predicts.34 Unlike Cartesian
skeptical hypotheses, the second hypothesis is not scientifically absurd. The
epistemological question, then, is: What reason could we have for attributing
distortions in our measuring rods to spatial curvature rather than to universal
forces?
The semantic questions concern how the meaning of spatial language
could be fixed in such a way that it would remain an open question whether
space is curved. To simplify, pretend that all spatial facts may be expressed
using Tarski’s predicates:35
point x is between points y and z
points x and y are congruent to points z and w
One sort of semantic question arises only given an extreme empiricist
philosophy of language. If every meaningful predicate must be associated
with verification conditions for its application, then given the previous
paragraph, the predicates ‘between’ and ‘congruent’ would seem not to be
meaningful.
Other semantic worries will have force even for nonverificationists. If
Einstein is right and Newton is wrong about curvature, then the referents of
‘between’ and ‘congruent’ must satisfy nonEuclidean rather than Euclidean
axioms. But in addition to having an interpretation under which they satisfy
nonEuclidean axioms, ‘between’ and ‘congruent’ also have an interpretation
in which they satisfy Euclidean axioms. (The Euclidean axioms are true in
the abstract model whose domain is ℝ3. and in which the predicates are
interpreted in the obvious way. Since the set of physical points of space has
the same cardinality as ℝ3, the model in ℝ3 induces a model in physical
space.) So if Einstein is to be right and Newton wrong, it must be that only
one of these assignments is the correct assignment, the intended
interpretation of ‘between’ and ‘congruent’. But what determines that one of
these assignments is the correct interpretation? We cannot specify the
intended interpretation of ‘congruent’ by saying that “it is to apply to x, y, z,
and w when the distance between x and y is the same as the distance between
z and w”, for ‘distance’ is in the same boat as ‘congruent’; how is its intended
interpretation determined?
The metaphysical questions concern the same issue as the semantic ones,
only more directly. In what would the fact that spacetime is curved consist?
In the fact that the congruence and betweenness relations satisfy
nonEucledian axioms, is the obvious reply. But since there are continuum-
many spacetime points, there exist “Euclidean-congruence” and “Euclidean-
betweenness” relations that satisfy Euclidean axioms. So in what sense is
spacetime really Euclidean? What makes the “real” betweenness and
congruence relations, as opposed to their Euclidean counterparts, “physically
significant”?
We face the same questions when reading physics textbooks on special or
general relativity that speak of the “intrinsic structure” of physical space,
time, and spacetime. In classical physics, we are told, spacetime is flat, and
there is a “well-defined” relation of simultaneity, whereas in Minkowski
spacetime there is no such relation of simultaneity—there is no
“distinguished” notion of simultaneity. But of course, there are relations—
sets of ordered pairs, anyway—between space-like separated points of
Minkowski spacetime that foliate the spacetime; many such relations. What
does it mean to say that none of these relations is “distinguished”?36
Geometrical conventionalists like Henri Poincaré (1952, Part II) and Hans
Reichenbach (1958, Chapter 1) give a deflationary answer to the semantic
questions that answers the epistemological questions, and which implicitly
assumes a deflationary answer to the metaphysical questions. Return to the
question of how to specify the intended interpretation of ‘congruent’.
According to Reichenbach, a theoretical predicate like ‘congruent’ requires a
“coordinative” definition, a definition that correlates the predicate with
something that is (relatively) observable. An example of a coordinative
definition would be the definition of straight lines through spacetime as the
paths of light rays in vacuum. We might think to give a coordinative
definition of congruence in terms of measuring rods: points of space are
congruent when they can be the endpoints of a single measuring rod. But
measuring rods can be distorted by forces. Might we then define congruence
in terms of measuring rods unaffected by forces? The problem is that ‘force’
is itself a term in need of a coordinative definition, since one can no more
directly measure forces than congruence. What Reichenbach says, in essence,
is that one must simultaneously give coordinative definitions of ‘force’ and
‘congruent’ in terms of measuring rods: congruent points are those picked out
by the endpoints of a measuring rod that is not subject to forces. Since this
coordinative definition constrains two terms, there is a certain amount of
freedom in assigning meanings to those terms. One can understand ‘force’
and ‘congruent’ so that space is Euclidean but there are universal forces
acting on all objects that produce the measurements that we make, or one can
understand ‘force’ and ‘congruent’ so that space is nonEuclidean and there
are no universal forces.37 In fact, physicists have preferred the latter course,
and so we speak of space as being curved. But this is in part a matter of
arbitrary definition; physicists could have chosen the former course. They
chose the latter only because the resulting physics was simpler. So according
to conventionalists like Reichenbach, it is at best misleading to say that space
itself is curved, that space is intrinsically curved. ‘Space is curved’ is true to
say given the linguistic choices that physicists have in fact made. But those
choices were arbitrary, and moreover are inextricably tied to the conventional
choice of whether to speak of universal forces.
Given this view about the semantics of spatial language, the
epistemological questions are immediately answered. How do we know that
space is curved, rather than being flat but accompanied by compensating
universal forces? We know this simply by knowing which conventions our
linguistic community has adopted. (Better: knowledge of linguistic
conventions plus empirical observation tells us that ‘space is curved’ is the
right description. If different linguistic conventions had been adopted, the
same observations would have called for the description ‘space is flat’.)
Reichenbach (unsurprisingly) does not address the metaphysical question.
But it seems clear that he would regard talk of space as being “intrinsically”
flat or curved as misguided, given the need for coordinative definitions.
Insofar as conventionalists have a metaphysics of spatial structure at all, it is
that space is not intrinsically structured (or perhaps that talk of intrinsic
structure makes no sense).
What might an anti-conventionalist—realist—account of physical
geometry look like?38 Here is what the realist wants to say. About semantics,
spatial predicates like ‘between’ and ‘congruent’ can be understood purely
spatially; they need no coordinative definitions. And they are attached to
particular relations over physical points, which satisfy nonEuclidean axioms
(assuming that Einstein is right). As for metaphysics, space is intrinsically
structured; the genuine betweenness and congruence relations are privileged
in a way that Euclidean-betweenness and Euclidean-congruence are not. The
naïve and natural picture of physical geometry one gets from physics is
thereby vindicated. The epistemological problems then confront us head-on.
But these problems should be solved the way we realists solve all such
problems of theory being underdetermined by observation: by appealing to
criteria—“simplicity” is a common placeholder—that choose between
observationally indistinguishable theories. (If such criteria do not deliver a
verdict, we remain agnostic until some new test or consideration breaks the
stalemate.)
A realist about structure has a clear path to this realism about physical
geometry. Metaphysically, the distinction enjoyed by the genuine
betweenness and congruence relations is that they are part of reality’s
distinguished structure: they carve perfectly at the joints, unlike any relations
of Euclidean-betweenness and Euclidean-congruence. Semantically, given
any reasonable metasemantics for theoretical terms, ‘between’ and
‘congruent’ attach to betweenness and congruence rather than to any
Euclidean-betweenness and Euclidean-congruence relations precisely
because only the former are part of the world’s genuine structure.
Reichenbach was led to his position by his insistence on the need for
coordinative definitions, deriving ultimately from an internalist and highly
empiricist approach to meaning. But a more reasonable metasemantics will
allow a role for a nonob-servational and externalist determinant of meaning:
the world’s structure.39
More generally, questions about metric, affine, topological, and other
structure of space, spacetime, and other physical spaces are questions about
the distinguished structure of those spaces. There is a substantive purely
spatiotemporal fact of the matter as to whether spacetime is Galilean,
neoNewtonian, Minkowskian, or a curved Lorentzian manifold.40 The fact is
given by the joint-carving features of points of spacetime. If, for example, the
joint-carving features of points of physical spacetime are as described by
Minkowski’s theory, then physical spacetime is Minkowskian, and there is
“no physically distinguished relation of simultaneity” in the sense that there
is no joint-carving relation that foliates the spacetime (nor can any foliation
be defined from the joint-carving relations over points).
Spacetime could have had a structure that would have vindicated a kind of
geometrical conventionalism. Suppose spacetime had lacked distinguished
metrical structure—suppose there had been joint-carving topological features
but no joint-carving metrical features. Then no metric would have been
distinguished from any other, and spacetime would have been a kind of
amorphous “point soup”. Reality might at the same time have lacked
sufficient structure to define forces. In such a Reichenbachian world, we
would have been free to choose either of a pair of coordinative definitions,
simultaneously defining force and metric predicates. Neither choice would
have carved reality at its joints better than the other. Metric and force
predicates would require coordinative definitions in such a world, not
because of general semantic considerations, but rather because the world
would lack the structure needed to supply semantic determinacy. What reason
do we have to think that our world has any more structure? The fact that
physical theories with primitive metrical predicates have been so successful
(section 2.3).

1 Lewis’s reason is different: metaphysically fundamental laws would require “necessary

connections” (Lewis, 1983b, p. 366); but see section 12.5.
2 See also section 12.1.
3 I have in mind special-science generalizations that are physically contingent—perhaps because
they depend on certain physically contingent “initial conditions”.
4 It might be held instead that the point of the notion of law is to play some role in systematic
metaphysics. For example, laws are commonly taken to play a constitutive role in the analysis of
counterfactuals: we “hold constant” the laws in determining what would have happened under a
counterfactual supposition. Even so, we should still broaden our focus beyond laws, since surely we
hold constant these non-laws when evaluating counterfactuals.
5 See Hamilton (2007, section 2) for a survey.
6 Hirsch (1993, chapter 3, section 7a) argues that explanations cast in joint-carving terms can be
recast, without explanatory loss, in a priori necessarily equivalent non-joint-carving terms. But the
recast explanation will have “syntactic” demerits, such as being highly disjunctive. So let us refine our
claim about explanation: good explanations must be cast in joint-carving terms when stated in
syntactically ideal form.
7 The problem derives ultimately from Wittgenstein (1958), and in the form presented here from
Putnam (1978, part IV; 1980; 1981, chapter 2) and Kripke (1982).
8 Let us ignore the complicating factor that there will exist multiple correct interpretations, because
of benign sorts of semantic undetermination like vagueness.
9 If the sentences in S and (F) are all first-order then, by the downward Löwenheim-Skolem
theorem, it would be enough to stipulate that C is infinite. And even if S contains second-order
sentences, we can just stipulate that it does not contain the logically complex sentences necessary to
force domains larger than C.
10 The argument just given assumed an extensional conception of meaning. Williams (2005, chapter
5) shows how to rework it under richer conceptions of meaning.
11 The name was introduced by Hodes (1984, 135) for a related view (which he rejected).
12 See, for example, Chomsky (2000); Pietroski (2003); Wilson and Sperber (2004).
13 See McGee (2005).
14 A defender of this metaphysics might argue that it is unlikely that fundamental reference relates
us to the bizarre semantic values. The simplest hypothesis, it might be claimed, is that we bear the
fundamental reference relation to reasonably joint-carving semantic values; the reasoning here, it might
be claimed, is analogous to any choice of a simplest hypothesis about the behavior of fundamental
properties and relations. (Kripke (1982, pp. 38–9) complained that a simplicity response to his
Wittgensteinian skeptic ignored the fact that the skeptical problem is not primarily epistemic, but rather
is one of what constitutes semantic facts. But on the view here envisioned, semantic facts are
fundamental, and the only remaining problem is epistemic.)
15 See, however, Hawthorne (2006a, section 17).
16 Causal theories say that reference is a certain sort of causal relation, and it’s arguably built into
the nature of causation that only reasonably joint-carving relations are causal. So causal metasemantics
may not need reference magnetism. Even so, joint-carving remains crucial to metasemantics, via its
connection to causation. See the discussion of causal theories below.
17 These remarks defend only against the extreme undetermination threatened by the Putnamian
argument, not against the kind of underdetermination about which Fodor (or for that matter, Quine in
Word and Object (1960c)) is worried. The simple causal theories that Fodor rejects because they don’t
allow for misrepresentation generally employ notions that are just as joint-carving as those he himself
employs in his own theory Whether ‘rabbit’ determinately means rabbit rather than undetached rabbit
part is similarly not settled by reference magnetism. (Not that reference magnetism is irrelevant; it may
play a role in the story of why ‘identity’ means identity, which in turn is relevant to ‘rabbit’ meaning
rabbit.) Note that subsentential undetermination of Quine’s sort wouldn’t result in indeterminacy in,
e.g., whether (F) is true.
18 This view is often associated with Lewis; see especially:
… overall eligibility of referents is a matter of degree, making total theory come true is a matter of
degree, the two desiderata trade off The correct, ‘intended’ interpretations are the ones that strike the
best balance. (Lewis, 1984, pp. 227–8)
… we need some give and take between the eligibility of referents and the other factors that make for
‘intendedness’, notably satisfaction of appropriate bits of theory. (Lewis, 1983b, p. 372)
However, Lewis’s full view, as developed in (1969) and (1975), is more complex.
19 Being a maximal continuous portion of gold, to be more exact.
20 I learned of such examples from Matti Eklund, John Hawthorne, and Robbie Williams.
21 In its most simplistic form, anyway. See Devitt and Sterelny (1999, chapters 4–5) for a more
sophisticated discussion.
22 Note that the Ramsey-Lewis method for defining theoretical terms (Lewis, 1970b) has no hope of
working unless the property quantifiers are restricted to properties that carve at the joints.
23 See Lewis (1983b) on mathematical semantic skepticism, chapter 10 on logical joint-carving, and
section 3.4 on semantic determinacy for spacetime language.
24 Compare Hirsch (1993, pp. 63–5).
25 Davidsonians about causation must rephrase: my holding of the grue emerald did not cause the
sensations in virtue of its being a holding up of a grue thing. See Davidson (1970); Kim (1989).
26 Whether in its theory of events, or its theory of the causal relation, or both. See Lewis (1986a;
1973a; 1979).
27 Even the primitivist accounts of Armstrong (1983), Dretske (1977), and Tooley (1987)
presuppose sparse universals.
28 Compare Quine (1969), though Quine held a deflationary account of joint-carving (under the
rubric of “natural kinds”).
29 Indeed, it has been suggested that if one attends to the comparative notion of confirmation,
Hempel’s puzzle, anyway, evaporates. See Fitelson (2006) for a survey
30 That is, if the subject begins with probability function Pr, and has an experience described by
proposition e, then she will subsequently have a probability function that assigns to any proposition, p,
the probability that Pr assigned to p conditional on e, i.e., Pr(p/e), i.e.,
31 See Fitelson (2006); Sober (1994).
 32 This language replaces ‘green’ and ‘blue’ by ‘grue’ and ‘bleen’, where an object is bleen iff it is
blue and first observed before 3000 A.D., or green and not first observed before 3000 A.D. In a sense
this language equals the ‘green’/‘blue’ language in descriptive power; one simply says ‘grue’ instead of
‘green’ and ‘bleen’ instead of ‘blue’ for objects first observed before 3000 A.D., and ‘bleen’ instead of
‘green’ and ‘grue’ instead of ‘blue’ for other objects.
33 See also Bricker (1993); Sider (1993a, chapter 9).
34 See Reichenbach (1958, chapter 1).
35 These are the predicates from Tarski’s axiomatization of Euclidean geometry; see Tarski (1959);
Tarski and Givant (1999). Really, though, the fundamental metrical facts should probably not be taken
to be direct point-to-point distance comparisons as in Tarski’s system, but should rather be local
metrical facts, from which distances along paths may be recovered. On the other hand, the standard
development of a local metric should probably not be taken at face value, since it grounds metric
structure in the metric tensor, a mathematical object involving real numbers. Surely the fundamental
distance facts are purely about points (as they are in Tarski’s account). Thus what we really want is a
synthetic geometry from which one can prove representation theorems about the metric tensor (see
Field (1980); Mundy (1987) for two approaches to representation theorems). I do not know whether
such an account exists.
36 See North (2009) on structure in physics generally.
37 Better: we can understand ‘force’ and ‘congruent’ so that the sentence ‘space is Euclidean’
comes out true, etc.
38 Nerlich (1976), especially chapter 9, defends realism about spatiotemporal structure, and
distinguishes this realism from realism about the existence of entities (chapter 5, §6). See also Bricker
(1993).
39 Grünbaum (1973) also defended conventionalism about metric structure, but based it on
metaphysical considerations rather than on an empiricist account of meaning. An intrinsic metric,
Grünbaum argued, would have to be definable from facts intrinsic to the points, but if space is
continuous then there are no facts intrinsic to points that suffice for the definition. The realist about
structure, however, regards the distinguished structure of space as constituting facts intrinsic to points
from which a metric may be defined.
40 Or even whether there is a fundamental four-dimensional spacetime at all, as opposed to a
massively dimensional configuration space; see Albert (1996); North (2009; 2010).
4 Substantivity
Structure is the key to understanding an elusive notion of substantivity.1

4.1 Nonsubstantive questions

Is Robinson Crusoe a bachelor? Is a water glass a cup? Is a protrusion from
the floor of a deep ocean, whose tip is a tiny island, a mountain (Hawthorne,
2006b, vii.)? Is “some nonsense made out of sour green apple liqueur”,
served in a V-shaped glass, a martini (Bennett, 2009)?
Philosophers like to argue about such things at bars, but even they regard
the questions as being, in some sense, nonsubstantive (shallow, nonobjective,
conventional, terminological). By contrast, the question of whether electrons
repel one another is substantive (deep, objective, nonconventional, about the
world).
This is intuitively clear; but in what does the lack of substantivity consist?
The answer is not straightforward. It’s not a matter of mind-dependence, for
example. Suppose, for the sake of argument, that Hawthorne’s protrusion is
indeed a mountain. Then, even if no people had existed, the protrusion would
still have been a mountain; and if everyone had said ‘Let such protrusions not
be mountains’, the protrusion would still have been a mountain. To
paraphrase Lincoln, calling a tail a leg wouldn’t make it one.
(The protrusion wouldn’t have been called a ‘mountain’ if people had not
existed, or had spoken or acted in certain ways. The linguistic fact that the
sentence ‘The protrusion is a mountain’ is true is mind-dependent, since
whether the sentence is true depends on what it means, and what it means is
mind-dependent. But the same goes for every true sentence, even ‘Electrons
repel one another’. So the mind-dependence of the linguistic fact isn’t what
makes the question of whether the protrusion is a mountain nonsubstantive.)
Nor is the lack of substantivity due to the questions being about words or
concepts—not in the most straightforward sense anyway. Questions such as
“Is ‘attorney’ synonymous with ‘lawyer’?” and “Does the concept of a sport
include gymnastics?”, which name or quantify over words or concepts, are in
a straightforward sense about words or concepts; but our nonsubstantive
questions are not about words or concepts in this sense. The question of
whether martinis can be made of sour apple liqueur is about martinis, and
what they can be made of. It is no more about ‘martini’, or the concept of a
martini, than the question of whether electrons repel one another is about
‘electron’ or the concept of an electron.2
Nor is a question’s nonsubstantivity due to its having no answer—to there
being no fact of the matter as to what its answer is. For the nonsubstantive
questions considered above might well have answers. Edmund Gettier (1963)
uncovered a surprising feature of our concept of knowledge: that justified
true belief does not suffice for its application. For all we know, clever
Gettiers could convince us beyond the shadow of a doubt that our concepts of
bachelor, cup, and the rest yield answers to our questions. (As the case of
knowledge shows, this possibility isn’t ruled out by the failure of a few hours
or centuries of armchair reflection.) But the questions still seem
nonsubstantive.
What if there will be no Gettier-like successes in conceptual archaeology?
It’s arguable that our questions then have no answers. The main opposition to
this conclusion comes from the epistemicists about vagueness (Sorensen,
2001; Williamson, 1994), who say that even without the facts that Gettiers
uncover—at least in part, surprising facts of usage that nevertheless are
readily appreciated once noticed—the questions have answers.3 But if we
reject epistemicism—as I think we should—then we may well want to say
that the questions without lurking Gettiers have no answers. Where our
conceptions of bachelors, cups, and mountains run out, so to speak, there are
no “further facts” to be discovered. Failure to have an answer is thus
sometimes symptomatic of nonsubstantivity. But since it is not necessary for
nonsubstantivity, it is not the underlying cause.
The failure of the questions to be substantive, then, is hard to pin down;
it’s neither mind-dependence, nor a conceptual or linguistic subject matter,
nor the lack of an answer. This might remind one of a problematic familiar
from discussions of antirealism.4 Begin, for example, with expressivism
about morality, according to which the function of assertions about right and
wrong is to convey the speaker’s attitudes, rather than to “depict objective
facts”. But make the expressivism sophisticated, so that it applies to logically
complex sentences; and further, adopt a deflationary approach to truth and
related concepts. Then you will be willing to say things like “moral sentences
can be true” (since ‘“killing is wrong’ is true” boils down to “Killing is
wrong”, and you’re willing to say the latter because you’re down on killing),
“there are moral facts” (since “it’s a fact that it’s wrong to kill” boils down to
“Killing is wrong”), and “there are moral properties (since “killing has the
property of being wrong” boils down to “Killing is wrong”). In what, then,
does your moral antirealism consist? There are indeed analogies here with my
examples. The nonsubstantivity of my questions does not consist in discourse
about bachelors (cups, etc.) failing to be truth-apt, or there failing to be facts
about bachelors, or there failing to exist a property of being a bachelor. But
the analogy quickly breaks down, since an expressivist semantics for
discourse about bachelors, cups, and the rest is clearly a nonstarter. Nor is J.
L. Mackie’s (1977) error theory a good model: our beliefs about bachelors,
cups, and the like are surely not radically mistaken (not because of their
nonsubstantivity, anyway).

4.2 Substantivity characterized

I think that our questions are nonsubstantive because their answers turn on
which of a range of equally good available meanings we choose for the words
in those questions.5 Better (though this will need further refining): for one or
more expressions E (e.g. ‘bachelor’) in a nonsubstantive question, the
semantic candidates for E (unmarried-adult-male, unmarried-adult-male-
eligible-for-marriage, etc.) are such that i) each opposing view about the
question comes out true on some candidate; and ii) no candidate carves at the
joints better than the others. If E uniquely means one of these candidates, this
is not because that candidate is intrinsically privileged. It is only because our
linguistic community happened to select that candidate rather than one of its
mates as the meaning of E.6 The situation may be depicted thus:
Given what we happen to mean by E, the nonsubstantive question may have
an answer, but we could just as easily have meant something else by E,
something equally good, one of the other non-joint-carving candidates in the
vicinity, in which case the question would have had a different answer.
A substantive question is one that is not nonsubstantive. The answer to a
substantive question is not sensitive to a choice amongst equally joint-carving
candidates. Now, one way for this to happen is for the question to be cast in
perfectly joint-carving terms (and for none of the expressions in the question
to have multiple perfectly joint-carving candidates). Each expression E in the
question may then be depicted as follows:

But this is not the only way for a question to be substantive. I drank no
alcohol on January 1, 2011; thus, even though the question of whether I had a
martini that night is not cast in particularly joint-carving terms, it is
substantive because it has the same answer (no) under all candidates for its
terms. Being cast in joint-carving terms is thus not the sole determiner of
substantivity. Being wholly cast in perfectly joint-carving terms is normally
sufficient for substantivity (since rarely does an expression have multiple
joint-carving candidates), but it’s not necessary.
Rival conceptions locate nonsubstantivity in some defect of the
proposition in question—for example, mind-dependence, failure to have a
determinate truth-value, and so on. On my conception, even questions
without such defects can be nonsubstantive. The question of Hawthorne’s
protrusion is nonsubstantive even if ‘the protrusion is a mountain’ expresses a
true mind-independent proposition about the physical world. The
nonsubstantivity is not due to any defect of this proposition, but rather to the
proposition’s “metasemantic surroundings”—the other propositions that the
sentence could have expressed had we meant a different candidate for
‘mountain’. It is the failure of the actually expressed proposition to stand out
from its surroundings (and the fact that it differs in truth-value from some of
the surrounding propositions) that generates the nonsubstantivity. Put another
way, the nonsubstantivity results from our process of selection of the
proposition, rather than being intrinsic to the proposition itself.
What follows is a series of refinements and amplifications of this account.
This process is not intended as a conceptual analysis of substantivity,
thought of as pretheoretically given. The aim is rather to introduce a concept
that sheds light on the phenomena. Also, this concept is not intended to apply
to everything that might justly be called “nonsubstantive”. For example, it
isn’t meant to apply to equivocation between distinct lexical meanings (as in
a dispute over whether geese live by “the bank”, in which one disputant
means river bank and the other means financial bank), or disputes involving
expressivist language. Nor is it meant to capture the shallowness of inquiry
into whether the number of electrons in the entire universe is even or odd (an
inquiry that is substantive in my sense, but pointless). My goal is simply to
identify one distinctive—and neglected—type of nonsubstantivity.
On to the amplifications and refinements. First, I will speak of
substantivity for many items: sentences, questions, disputes, and so on. The
definition may be adjusted in obvious ways for these items. Examples: a
question (construed as the set of sentences that are its possible answers—this
was implicit above) is nonsubstantive iff the candidates of some expression
are equistructural and each answer comes out true under some candidate; a
sentence is nonsubstantive iff the candidates of some expression are
equistructural and both the sentence and its negation come out true under
some candidate.
Second, I said earlier that being cast in perfectly joint-carving terms
normally suffices for substantivity. But being cast in highly albeit not
perfectly joint-carving terms—a common occurrence in the special sciences
—also normally suffices for substantivity. Except for questions that strain the
boundaries of taxonomy (a relatively uncommon occurrence), special-science
questions normally fall into one of the following two categories: i) each
expression has a candidate meaning that carves far better than all other
candidates; or ii) each expression has a range of candidates that carve far
better than do other candidates not in the range, and the question’s answer is
insensitive to choices of candidates within these ranges. In either case, the
question is substantive.
Third, some disputes don’t seem substantive even though there is a unique
joint-carving candidate, if that joint-carving candidate is of the wrong sort.
Recall Hawthorne’s physically significant line in the Ural mountains (section
3.2), and consider the question of whether a certain location near the line is
part of Europe. The answer to this question turns on which of several
candidates we mean by ‘Europe’. Now, given the line, one of these
candidates carves at the joints far better than the rest, and so my account
classifies this dispute as substantive. But that seems wrong. The line through
the Urals, though physically distinguished, isn’t geographically or politically
distinguished, and therefore seems irrelevant to the boundaries of Europe.
The physically significant line is indeed a joint in nature, but it’s the wrong
sort of joint in nature, relative to a dispute over ‘Europe’.
What is the “right sort”? I don’t have a finished answer, but here is the
beginnings of one. Suppose an expression, E, is a theoretical term, in the
sense of section 3.2—a term intended to stand for a joint in nature. Suppose,
further, that there is a single joint-carving meaning, m, that satisfies enough
of the “core theory” that is collectively associated with E by the participants
in the dispute. Then m is the right sort of joint-carving meaning. For example,
in a dispute over whether any electrons are located in a certain region, R, of
space, the term ‘electron’ is a theoretical term, and the property of being an
electron will satisfy enough of the core theory associated with ‘electron’ by
the disputants—a theory saying that electrons are subatomic particles that
orbit nuclei, have negative charge, a certain mass, and so on. Thus, this
property will be the right sort of joint-carving candidate meaning, and the
dispute is substantive.
It’s important that the right sorts of joint-carving meanings needn’t satisfy
all of the core theory, since there can be substantive disputes even when the
core theory is somewhat mistaken. The dispute over whether any electrons
are located in R remains substantive even if scientists are mistaken about the
mass of an electron. The property being an electron remains the right sort of
joint-carving meaning because it satisfies enough of the core theory. Also,
there can be substantive disputes over which central principles electrons
satisfy—over what the mass of an electron is, for example. In such cases the
core theory that is “collectively” associated with ‘electron’ will be drawn
from the principles not under dispute. To have a substantive dispute about
electrons, there must remain enough common ground about what electrons
are so that all disputants can be regarded as talking about the same thing.
What is the “wrong sort” of joint-carving candidate? Hawthorne’s line is
the wrong sort to render a dispute about ‘Europe’ substantive because
‘Europe’ is not a theoretical term; ‘Europe’ isn’t intended to stand for a joint-
carving meaning at all. It might be objected that ‘Europe’ is a theoretical
term, albeit one of political science or geography rather than physics. But
then the physically significant line will still be of the wrong sort, since it
won’t satisfy the associated core theory. That core theory will be a political or
geographical theory, not a physical theory, and the physical line won’t play a
role in any distinctively political or geographical laws or explanations. (If it
did, then it wouldn’t be merely “physically significant”, and the dispute
would then be substantive.)
The revised account, then, says that a nonsubstantive question is one
containing an expression E whose candidates are such that i) each opposing
view about the question comes out true on some candidate; and ii) no
candidate carves at the joints in the right way for E better than the rest. A
candidate c1 carves better “in the right way for E” than another candidate c2,
to a first approximation anyway, iff E is a theoretical term, c1 satisfies
enough of the core theory associated with E, and c1 carves better than does
c2.
One might worry about the fact that this definition makes all differences
in joint-carving for candidates of nontheoretical terms irrelevant to
substantivity. Recall the question of whether I had a martini on January 1,
2011, which is intuitively substantive (and has the answer no) because I
drank no alcohol that night. The current definition counts this question as
nonsubstantive if its answer is sensitive to the choice of candidates for its
nontheoretical terms, regardless of distinctions of joint-carving amongst those
candidates. But won’t there be Putnamlike candidates, based on arbitrary
permutations, under which the question’s answer is yes? Perhaps it would be
enough to reply that Putnamian semantic values are not candidates because
they just couldn’t be meant by any linguistic community. Or perhaps—the
more likely case, I fear—a subtler definition of “the right sort” is needed, no
doubt based on a subtler distinction than that between theoretical and
nontheoretical terms.
Fourth, ‘candidate’ needs to be clarified. On the one hand, it shouldn’t be
taken too narrowly, to include only what supervaluationists call
“precisifications”. Precisifications of E are, intuitively, those semantic values
that our usage of E doesn’t definitely rule out. But suppose that a Gettier
could show us that our usage of ‘bachelor’ definitely excludes Crusoe. Then
semantic values under which ‘Crusoe is a bachelor’ is true aren’t
precisifications of ‘bachelor’, but they should count as candidates for
‘bachelor’ (since whether Crusoe is a bachelor is paradigmatically
nonsubstantive). So candidatehood is consistent with a certain degree of
“mismatch with usage”. On the other hand, match with usage isn’t irrelevant.
The property of being an electron fails to be a candidate for ‘bachelor’
precisely because it doesn’t come close to fitting our usage of that term. A
candidate meaning m needn’t perfectly match our usage of E; but the
mismatch can’t be too severe. If a linguistic community, roughly in our
circumstances, could have used E to mean m without seeming “semantically
alien”—could have used E to reach “the same semantic goal” as we use E to
reach, albeit perhaps by a different route—then m is a candidate for E.
This is admittedly pretty vague. Now, it’s fine if ‘substantive’ and related
notions are rough and ready, come in degrees, admit borderline cases, and so
on. Still, to firm up the notion a bit, consider another example (there will be
more later). Suppose, for the sake of argument, that i) causation does not
carve at the joints; and ii) in English, ‘cause’ has a counterfactual analysis.
Now consider a linguistic community L in which ‘cause’ has a covering-law
analysis, but in which the conceptual role of ‘cause’ is similar to its role in
English, in the following sense. Causation, in both our community and L,
stands in a complex network of conceptual relations to concepts of moral
responsibility, control, and myriad others. Although the members of L use
‘cause’ differently from how we do, there are also corresponding differences
in their usage of ‘responsible’, ‘control’, and other such terms, so that the
network of relations is preserved. I want to count this linguistic community as
not being (too) semantically alien. Given the similarity of conceptual role,
speakers of L use ‘cause’ with the same semantic goal as we do with our
word ‘cause’. Thus the covering-law meaning is a candidate for ‘cause’; and
as a result, assuming that causation doesn’t carve at the joints, the question of
whether effects counterfactually depend on their causes is nonsubstantive.
Fifth, my notion of substantivity is a metaphysical one, to be contrasted
with a notion of conceptual substantivity. Many expressions that fail to carve
at the joints are embedded in our conceptual lives in important ways; and
questions involving such expressions can have a sort of conceptual
substantivity even when my analysis counts them as (metaphysically)
nonsubstantive. Suppose again that the fact that causes counterfactually
imply their effects is nonsubstantive in my sense. Still, discovering that
causation is counterfactual dependence might reveal something important
about our conceptual scheme, in contrast with a discovery that cups are
glasses, which we would regard as an intellectual trifle. Given the network of
connections between causation and other concepts, the discovery about
causation would have far-reaching implications. And causation matters to us
in ways that the concept of a cup does not—partly because of the network.
True, in learning that causes counterfactually imply their effects, we are
primarily learning something about ourselves; that’s the metaphysically
nonsubstantive part. But we’re learning something important about ourselves.
Sixth, my notion of substantivity is essentially metalinguistic. Consider:
(1) Drinks with sour apple liqueur are not martinis.
In my opinion, (1) is true; ‘martini’ means drink made of gin or vodka and
vermouth with such-and-such proportions. (If you disagree, replace (1),
mutatis mutandis, with its negation.) But we could have used ‘martini’
differently, so as to include sour apple liqueur drinks, without being
semantically alien or carving worse at the joints. So (1) is nonsubstantive.
But now consider:
(2) Drinks with sour apple liqueur are not drinks made of gin or vodka and
vermouth with such-and-such proportions.
(2) is substantive. Although the terms in (2)—‘sour apple liqueur’, ‘drink’,
‘gin’, and so on—do not carve at the joints, they have no candidates under
which (2) comes out false. But (2) is just the result of substituting for
‘martini’ in (1) an expression that has the same meaning (though not the same
candidates). Moral: sentences that express the same proposition can differ in
substantivity. This is a feature of my account, not a bug. Notions like
substantivity, depth, objectivity, and the rest have proved so elusive partly
because philosophers have been looking in the wrong place: in what we say
with the disputed vocabulary.7 Substantivity, or lack thereof, is not intrinsic
to semantic values. We might put this by saying that we should look to
metasemantics, rather than semantics, to reveal substantivity.
Seventh, substantivity can depend on the world, and not just on the words
involved and what candidates they have. Consider the question of whether
there is lithium on Mars. It is natural to think that some expression in ‘There
is lithium on Mars’ has a range of equally joint-carving candidates
corresponding to the “fuzziness” in Mars’s spatial boundaries. For the sake of
definiteness, suppose this to involve a range of mostly overlapping candidate
referents of the name ‘Mars’, which differ from one another by including
slightly different parts near Mars’s vague boundary. The question of whether
there is lithium on Mars is nonsubstantive, then, if and only if some, but not
all, of these candidate referents contain lithium. Now, if there is, in fact,
plenty of lithium in Mars’s core, so that all the candidate referents contain
lithium, then the question is substantive, since its answer is yes under each
candidate. But if the only lithium in the vicinity of Mars is a single lithium
atom at its vague border, then the question is nonsubstantive. So the
substantivity of the question depends on a fact about the world: the location
of lithium. Still, regardless of the location of lithium, the question “admits”
nonsubstantivity because without any change in meaning or candidates, the
question could have been nonsubstantive: there might have been lithium just
atMars’s vague borderline. In contrast, “Are there charged particles?” does
not admit nonsubstantivity, and “Is it possible for a martini to be made of
sour apple liqueur?” does not admit substantivity.
Eighth, it’s natural to think of substantivity as depending on the interests
of disputants. Suppose two scientists look at an atom through a telescope and
disagree over whether it is lithium. In fact, the atom is lithium. Also, no other
lithium atom is in the vicinity of Mars, and the scientists know this. Finally,
the atom at which they are looking is in Mars’s vague border. But the
scientists don’t know this last fact; indeed, they falsely believe that the atom
is definitely on Mars’s surface. As a result, they phrase their debate thus: “Is
there any lithium on Mars?”. For they believe (falsely) that the answer to this
question is definitely yes if and only if the atom seen through the telescope is
lithium. Now, my official account says that this question is nonsubstantive
(its answer turns on which candidate for ‘Mars’ is meant). But intuitively,
what is at issue is substantive, since the scientists don’t care whether the atom
is on Mars; they’re just using the question of whether there’s lithium on Mars
to get at the question of whether the atom seen through the telescope is
lithium. We might account for this by distinguishing the question the
scientists actually ask (‘Is there lithium on Mars?’) from the question they
really care about (‘Is the atom seen through the telescope lithium?’), and
applying the official account, as-is, to these questions. Alternatively, we
might alter the official account by treating substantivity as a property of
question–context pairs, rather than a property of questions simpliciter. The
idea would be to make the set of relevant candidate meanings for the question
depend on what issues are treated as important in the context. In the context
of the scientists, different candidate meanings for ‘Mars’ are not relevant
because the scientists don’t care whether the atom is located on Mars. The
difference between these two approaches seems insignificant.
Ninth, when nonsubstantive disputes have multiple expressions that fail to
carve at the joints, sometimes the source of the nonsubstantivity can be
localized to a proper subset of those expressions. For example, suppose
‘lithium’ has two candidates, lithium1 and lithium2. (Lithium is like jade.)
And suppose that Mars is shot through with lithium1, but has no lithium2.
Then the question of whether there is lithium on Mars is again
nonsubstantive. But even though both of the terms involved, ‘lithium’ and
‘Mars’, have multiple candidates, we can place the blame for the
nonsubstantivity solely on ‘lithium’. For under any candidate for ‘Mars’, the
question’s answer remains sensitive to the choice of a candidate for ‘lithium’,
whereas it’s not the case that for every ‘lithium’-candidate, the question’s
answer is sensitive to the choice of a ‘Mars’-candidate (indeed: for every
‘lithium’-candidate, the answer is insensitive to the choice of a ‘Mars’-
candidate).
Tenth, a question can be substantive “along some dimensions” but not
along others. Contrast the following two predicates:
Fx =df x is more massive than all bachelors
Gx =df x is wittier than all bachelors
(‘=df ’ means “means by definition that”). There is, I will say, just one
dimension along which F has multiple candidates and generates
nonsubstantive questions, whereas there are two such dimensions for G. F
and G are alike in having multiple candidates and generating nonsubstantive
questions. But in the case of F, these facts come from a single source: the fact
that ‘bachelor’ has multiple candidates. (Let’s assume for the sake of
argument that ‘all’ and ‘more massive than’ carve at the joints.) F’s
candidates are generated by those of ‘bachelor’; and there’s just one way for
it to be nonsubstantive whether a given thing is F: namely, for there to be
some candidates bachelor1 and bachelor2 of ‘bachelor’ such that the thing is
more massive than all bachelors1, but not more massive than all bachelors2.
Along the dimension of ‘more massive than’ (and ‘all’ as well), we may say,
the question of whether a given thing is F is substantive. In contrast, G’s
candidates are generated both by candidates of ‘bachelor’ and by candidates
of ‘wittier than’. There are, correspondingly, two ways for the question of
whether a given object is G to be nonsubstantive—nonsubstantivity “along
two dimensions”.
Given this terminology I can clarify some cryptic remarks at the end of
section 3.3. In that section I argued that even a subjectivist about epistemic
value might regard the distinction between reasonable and unreasonable prior
probability functions as being objective. According to this position, I said,
judgments about epistemic value are subjective “along one dimension” but
not along another. What I meant can now be stated more fully. Terms of
epistemic evaluation do not carve at the joints. However, some of their
“components” do carve at the joints—just as some of F’s components (‘more
massive than’, ‘all’) carve at the joints even though F itself does not carve at
the joints. If, for example, the terms of epistemic evaluation are defined in
part by “simplicity” constraints on prior probability distributions, then the
joint-carving components in question would be those that state the simplicity
constraint in joint-carving terms. Any nonsubstantivity in the question of
whether a prior probability distribution is reasonable would not be due to
candidates for these components (since they carve at the joints—reasonably
well, anyway), but would rather be due to candidates in the rest of the
components.
Eleventh, suppose some expression E has a candidate c that carves at the
joints (in the right way) much better than all its other candidates, but that the
actual meaning of E is not c, but rather some other candidate c@ that carves
much worse at the joints. (This could happen if reference magnetism is false,
or if for some other metasemantic reason the joint-carvingness of c is
outweighed by other factors.) Now, suppose that the truth-value of some
sentence S is sensitive to the fact that E means c@, rather than c. My account
treats S as substantive (assuming that none of its other expressions have
appropriately varying candidates), since one of E’s candidates (namely, c)
carves at the joints much better than the others. Nevertheless, there is an
intuitive sense in which S is nonsubstantive since, intuitively, its actual truth-
value is a mere reflection of a linguistic choice, not the world’s structure. S,
we might say, is “actual-verdict nonsubstantive” because it could have had a
different truth-value without carving worse than it actually does—because,
that is, S has a different truth-value under some candidate for one of its
expressions that carves at the joints no worse than that expression’s actual
meaning. Something substantive is at stake here, but the actual verdict is
nonsubstantive.

4.3 Conventionality
The next two sections discuss two species of nonsubstantivity. The first is a
sort of conventionality. A sentence exhibits this sort of conventionality when
it involves, in a sense to be explained, an arbitrary conceptual choice.8
The word ‘convention’ generally signifies an arbitrary choice amongst
equally good ways to achieve a certain goal by collective action. The USA
had a goal of safely organizing its motorways; that goal could have been
achieved either by everyone driving on the right-hand side or by everyone
driving on the left; the convention to drive on the right was a more or less
arbitrary choice between these two alternative solutions.
Turning to language, there is a sort of conventionality that is both familiar
and banal: the choice of symbol or sound to represent a given content. Call
this: symbol-conventionality. A second sort of conventionality—call it
content-conventionality—is exhibited by sentences that are about
conventions. The notion of aboutness is admittedly slippery, but obvious
examples of content-conventionality include ‘there are some conventions’,
‘“Snow” refers to snow in the dominant language of North America in 2011’,
and so on.
The sort of conventionality I have in mind is different. Sometimes we
have a certain semantic goal; we need to introduce a word in order to
accomplish that goal; and there are a number of different candidate meanings,
each such that the goal would be accomplished equally well if that candidate
were chosen as the meaning of the word. The choice of one of these
candidate meanings to be the meaning of the word exhibits what I’ll call
candidate-selection conventionality (sometimes just “conventionality”, when
there’s no danger of confusion).9
To illustrate, consider the word ‘inch’. The purpose of ‘inch’ is to be a
convenient measure for smallish things, the kinds of things we can hold in
our hands. But there is a range of very similar lengths that would each have
served this purpose. We chose one of these to mean by ‘inch’, but that choice
was arbitrary; any of the others would have served our purposes equally well.
This choice was one of candidate-selection convention.
By saying that each length in the range would have served our purposes
equally well, I have in mind two things. First, the lengths in the range carve
at the joints equally well.10 And second, adopting any of the alternate lengths
would have “achieved the same semantic goal”. All length-words achieve a
general semantic goal of allowing speech of absolute and relative sizes, but
‘inch’ has a more specific goal: to be a convenient measure of smallish
things. This goal could have been achieved by many lengths within a certain
range. But if ‘inch’ had meant mile, it would not have achieved exactly this
goal, since measuring smallish lengths in miles would be inconvenient. And
if ‘inch’ had meant something other than a length—for instance, if it had
meant happiness—then it would not have achieved anything like its actual
semantic goal.
All words for units of measure are conventional in this way. The
boundaries of countries provide a further example. When countries are
formed or resized, an arbitrary choice is sometimes made about where the
new border will go. The corresponding choice of semantic values for words
about the country is candidate-selection conventional.
Call a sentence candidate-selection conventional when its truth-value
turns on a candidate-selection conventional choice. Sentence (C) is an
example:
(C) My computer screen measures exactly 15 inches.
(C) is true, but would have been false if ‘inch’ had meant a slightly different
candidate length. This is not content-conventionality: (C) is not about
conventions in any interesting sense. Nor is it mere symbol-conventionality.
Of course, (C) is symbol-conventional; all sentences are. But not all
sentences contain a word for which we could have chosen an alternate
meaning that would have equally well suited our purposes for that word, and
which would have given the sentence a different truth-value.
It might seem odd to call (C) conventional (except, of course, in the sense
of symbol-conventionality). Facts about measurable quantities like length,
after all, are as objective as can be. But remember that my account of
substantivity, and hence of candidate-selection conventionality, is
metalinguistic. It is the sentence, not the fact, that is conventional. Moreover,
recall the interest-relativity of substantivity (the eighth refinement of section
4.2). It seems odd to call (C) conventional because in typical contexts where
(C) is disputed, the focus of the disputants is not on which candidate is meant
by ‘inch’, but rather on the length of the computer screen. However, consider
a context where the disputants do focus on which length counts as being “one
inch” (and let the length of the computer screen be common ground). Then
the dispute does seem nonsubstantive, and it seems natural to call the
sentence conventional.11 And if we consider sentences about measurable
quantities where the second sort of context is more common, the label
‘conventional’ no longer seems odd at all. A dispute between an American
and someone from England over whether a certain container of milk
measures “one gallon”, where the container’s volume is not under dispute, is
quite naturally thought of as nonsubstantive; and it’s natural to call sentences
in the dispute as being conventional, since their truth turns on the
conventional decision of whether to adopt the U.S. or imperial standard for
the gallon.
(Even paradigm nonsubstantive sentences exhibit the sort of relativity just
discussed. ‘The pope is a bachelor’ could be the subject of a substantive-
seeming dispute if, say, the disputants knew little about Catholicism and were
in effect arguing about whether the Pope is married.)
Our definitions ensure that conventionality (of the candidate-selection
variety) implies nonsubstantiality; but the converse does not hold. Both
involve a sentence whose truth-value depends on which of several equally
joint-carving candidates is assigned to one of its terms. But for
conventionality, there must be a selected candidate (or a vague selected
range, in the case of vague conventionality), and that selection must be made
by arbitrary choice. If no selection is made (whether because of vagueness or
some other form of semantic indeterminacy), or a selection is made
nonarbitrarily (see section 4.4), there is no conventionality. Also,
‘conventional’ seems most apt when the arbitrary choice is made more or less
consciously, when alternative choices stare us in the face, and when those
choices accomplish exactly the same semantic goal; it seems less apt when
the choice has been made implicitly and collectively, over time, when no one
thinks much about the alternatives, and when the alternatives accomplish
slightly different semantic goals. Supposing the question of whether Crusoe
is a bachelor to have an answer, perhaps we should not call it “conventional”.
Everyone agrees that the boundaries of countries and units of measure are
in some sense conventional. But claims of “conventionality” have been made
in more controversial domains: for physical geometry (recall section 3.4),
morality, and so on.12 Candidate-selection conventionality is a useful tool for
articulating such doctrines.13 For some of these conventionalists do not want
to claim that sentences about the target domain are about conventions, nor do
they wish to merely make the trivial claim that the sentences exhibit symbol-
conventionality.
Consider, for example, the view that moral sentences are candidate-
selection conventional. According to this view, society had an interest in
introducing moral vocabulary, attached to some norms or other; but within
certain limits, various norms would have equally well served the purpose.
Thus, there were various candidate meanings available for normative words,
each of which would have achieved those words’ semantic goal. Moreover,
none of these candidate meanings carves at the joints better than the others.
One of these was selected, more or less arbitrarily—a candidate-selection
convention. Now, this view is not a mere claim of symbol-conventionality.
By claiming that the meanings we have actually selected for moral language
are on a par with alternate meanings—both metaphysically and concerning
the satisfaction of the goal of morality—this view really does downgrade
morality (a part of it, anyway) in a way that a mere claim of symbol-
conventionality would not. But nor does this view imply that moral sentences
are about conventions. Thus it allows that (for example) murder would have
been wrong even if we had chosen different norms (since the conventionality
of the choice of norms is not built into the propositions we express using
moral words).

4.4 Subjectivity
The second species of nonsubstantivity may be called subjectivity. Like
conventionality, it occurs when a linguistic community chooses one of
several candidates. But in the case of subjectivity, the choice is not arbitrary;
rather, it reflects something important about the linguistic community.
As with conventionality, this sense of subjectivity must be distinguished
from others. Consider the following toy semantic theories of aesthetic
sentences:
Expressivism By uttering ‘x is beautiful’, a speaker communicates no
proposition, but rather gives expression to a certain positive aesthetic
attitude, A, to x.
Indexicalism By uttering ‘x is beautiful’, a speaker, S, communicates the
proposition that S bears attitude A to x.
Aesthetic sentences express attitudes, given the first semantics, and
communicate propositions about attitudes, given the second. Either way,
there is a straightforward kind of subjectivity. But the kind of subjectivity I
have in mind is different. It is brought out by a third semantics:14
Projectivism By uttering ‘x is beautiful’, a speaker, S, communicates the
proposition that x is P, where the property being P is a certain physical
property that is the linguistic meaning of the predicate ‘is beautiful’ in S’s
language; being P is the linguistic meaning of ‘is beautiful’ because
members of S’s linguistic community bear attitude A to Ps.
Under this semantics, aesthetic sentences are not subjective in the
straightforward sense. Accordingly, they pass common tests for “objectivity”.
For example, sentences about beauty make mind-independent claims:
beautiful mountains would still have been beautiful even if no humans had
ever existed. Nevertheless, there remains a clear sense in which the aesthetic
is subjective. For which physical properties aesthetic predicates stand for is
determined by the attitudes of the speaker’s community and any attitudes are
as good as any other.15 Even though the properties ascribed by aesthetic
predicates are wholly mind-independent, response-independent, and so on,
the selection of these features as the features to be expressed by aesthetic
predicates is accomplished solely by our having the attitudes that we do.
Imagine a range of linguistic communities, each with different aesthetic
attitudes. The predicate ‘is beautiful’ expresses different—equally joint-
carving—properties in these different linguistic communities.16 Where we
call a mountain beautiful, speakers of another language withhold that
predicate, and instead call the mountain ‘ugly’. In such cases, everyone
speaks truly; no one is making a mistake. The mountain has the property BE
that is expressed in English by ‘beautiful’, and lacks the property BO that is
expressed in the other language by ‘beautiful’.17 The appropriateness of the
language of subjectivity in this case is manifest.
Shine a light down on a piece of paper suspended over a table. If the paper
has a geometric shape cut from it—a circle, say—the light will shine through
the hole and project that shape onto the table below. Let X be the illuminated
portion of the table. X is circular. Moreover, X—that portion of the table—
would still have been circular even if a square rather than a circle had been
cut from the paper (though X would not then have been exactly illuminated).
But there is nothing special about X. The square region that would have been
illuminated, had the cut-out hole been square, is just as good a region. The
illuminated region is a projection of the hole. An observer of the illumination
will learn something more important about the hole than about the table, even
though circularity is an intrinsic feature of the illuminated part of the table.
Aesthetic features are like the shape of the illuminated region X, the selection
of meanings for aesthetic predicates like the selection of the shape of the hole
in the paper. If projectivism is true, then aesthetic features are not “about” us,
and would persist even if there were no humans. But assuming there are no
aesthetic joints in nature, nothing beyond our aesthetic attitudes constrains
the aesthetic categories picked out by aesthetic language; an observer who
watches us label things ‘beautiful’ and ‘ugly’ will learn as much about our
attitudes as they will about the things, even if these predicates ascribe
intrinsic properties of the things. Aesthetic predicates express the properties
they do because of our attitudes; aesthetic features are projections of our
attitudes.
I do not claim that projectivism is true; it is, I suspect, an overly simplistic
model of the semantics and metasemantics of ‘beautiful’. Its point, rather, is
to establish a general fact about the nature of subjectivity: there is a kind of
subjectivity that results, not from statements in the target discourse being
about our values, but rather from our values selecting one from a range of
equally good meanings. The importance of this fact for metaethics should be
clear. To earn titles of objectivity and realism (together with the associated
imagery of externality, discovery, and so on), it is not enough that evaluative
language be assigned “objective content”. For projectivism assigns physical
contents, which are as objective as can be; yet it merits neither the names nor
the imagery. The objective content must also stand out from its metasemantic
surroundings. It must enjoy some sort of privilege over alternate candidate
contents.18
A sentence is subjective, then, in the sense illustrated by the projectivist
semantics, if and only if its truth-value depends on which of a range of
equally joint-carving candidates is meant by some term in the sentence,
where the candidate that we in fact mean was selected in a way that is not
arbitrary, but rather, reflects something important about us, such as our
values.19 As with conventionality, it is appropriate to speak of this sort of
subjectivity only if no one candidate carves at the joints (in the right way)
better than the rest. If there were aesthetic joints in reality—vindicating the
very strongest form of aesthetic realism—then one of the communities from
our earlier example might match those joints with their usage of ‘beautiful’.
This lucky community would then be uniquely right about aesthetics, and talk
of “subjectivity” would seem out of place. They might be uniquely right in
the straightforward sense of being the only community that speaks truly, if
‘beauty’ in every community has the same, joint-carving sense (this might
happen because of reference magnetism). But even if the communities mean
different things by ‘beautiful’, so that every community speaks truly, the
lucky community remains uniquely right in the sense that only their term
‘beautiful’ gets at the distinguished aesthetic structure of the world. The other
languages are metaphysically second-rate. In such a circumstance, the
language of subjectivity again seems out of place, even though the unlucky
communities all speak truly given what they mean by ‘beautiful’.
The kind of subjectivity we have been discussing results from the
projection of our values. But perhaps ‘subjective’ is also appropriate when
we project important features of ourselves other than values. Suppose
causation does not carve at the joints, and that there are a number of
candidate semantic values for ‘cause’, none of which is metaphysically
distinguished. Suppose that one of these is in fact the actual semantic value,
for some reason that reflects something important about us. Suppose, for
instance, that it’s essential to the role that ‘cause’ plays in our conceptual
scheme that it have a counterfactual analysis. Like ‘beautiful’, the semantics
of ‘cause’ reflects an important feature of ourselves, not an arbitrary semantic
decision, but unlike ‘beautiful’, the feature has nothing to do with value.
Whether or not ‘subjective’ is appropriate here, this sort of projection is
opposed to “objectivity”. If, contrary to what I think, there is a joint-carving
relation of causation, then it would be natural to describe this by saying
“there are objective facts about causation”, or “one description of the facts of
causation is objectively correct”. Objectivity is opposed to subjectivity
conventionality or any other sort of nonsubstantivity.
Philosophers often make inchoate claims that are best understood as
concerning this notion of objectivity. To take one example, consider the
criticism made by Armstrong (1983, pp. 40–59) and others that Lewisian
laws of nature are not genuinely necessary and cannot explain regularities.
Everyone agrees, after all, that mere regularities do not explain regularities,
and need not be necessary; but Lewisian laws are just glorified regularities,
regularities that are integrated into the simplest and strongest system. Lewis’s
reply always seemed elusive and unconvincing:
Some familiar complaints seem to me question-begging … If you’re prepared to agree that theorems of
the best system are rightly called laws, presumably you’ll also want to say that they underlie causal
explanations; that they support counterfactuals; that they are not mere coincidences; that they and their
consequences are in some good sense necessary; and that they may be confirmed by their instances. If
not, not. It’s a standoff—spoils to the victor. (1994, pp. 478–9)

Lewis’s response is in essence to define ‘nomic necessity’, ‘explanation’, and

so on, in terms of ‘law’, and then to claim that if you don’t think that
necessity and explanation, thus defined, are genuine, that must be because
you are just resisting his analysis of lawhood. What is really going on, I
think, is that Armstrong is bothered by the fact that on Lewis’s account, laws
(and so everything defined in terms of them) are not objective. For Lewis
there is no distinguished structure in the vicinity of laws of nature. (This is
the whole point of Lewis’s Humeanism.20) This rankles some; they believe
that laws “run deep”; they cut at the joints; they are objective. Likewise for
necessity, explanation, and the rest: Lewis provides analyses of these
concepts that to some degree fit our ordinary concepts, but are not
particularly objective. This is what really bothers Armstrong (and Lewis’s
reply is not responsive).

4.5 Epistemic value

I have been arguing for connections between joint-carving and a range of
concepts. Many of these connections can be unified by a single thesis about
epistemic value: it’s better to think and speak in joint-carving terms. We
ought not to speak the ‘grue’ language, nor think the thoughts expressed by
its simple sentences.
The goal of inquiry is not merely to believe truly (or to know). Achieving
the goal of inquiry requires that one’s belief state reflect the world, which in
addition to lack of error requires one to think of the world in its terms, to
carve the world at its joints. Wielders of non-joint-carving concepts are worse
inquirers.
Recall the community that divides the red–blue world along the diagonal
plane (section 1.1). They are missing out; they do worse than we do as
inquirers. Of course, if they explicitly form false beliefs about structure, then
there is a perfectly obvious sense in which they are doing worse. But even if
neither they nor we form such beliefs, they still do worse, simply by thinking
of the world in the wrong terms.
 Joint-carving thought does not have merely instrumental value. It is rather
a constitutive aim of the practice of forming beliefs, as constitutive as the
more commonly recognized aim of truth. Nor is joint-carving a conscious
goal, at least not of most inquirers. It is rather a standard by which beliefs and
believers may be evaluated, whether or not it is consciously acknowledged.
(Strong and weak versions of this thesis about epistemic value can be
distinguished.21 The strong version is what I have been advocating: it’s worse
to employ non-joint-carving concepts. The weak version says merely that it’s
worse to believe non-joint-carving propositions, where a proposition is joint-
carving to the extent that it can be simply expressed using joint-carving
concepts, given some appropriate notions of simplicity and proposition. (The
strong version implies the weak, more or less, given the perhaps plausible
additional assumption that anyone employing joint-carving concepts ought to
believe propositions simply expressible in terms of them.) The weak version
allows one to employ ‘grue’ and ‘bleen’, provided one does not believe
propositions like all emeralds in this room are grue, but rather believes
instead propositions like all emeralds in this room are either grue and first
observed before 3000 A.D. or bleen and not first observed before 3000 A.D.
—the idea is that this latter proposition is relatively joint-carving because it
can be expressed by “all emeralds in this room are green”. Both versions are
correct, I think, and so I stick with the strong. But one might reject the strong
version while accepting the weak; and much of what I go on to say could be
reformulated to appeal only to the weak.)
In his book Dividing Reality, Eli Hirsch argues effectively that it is
intuitively compelling that we ought to speak and think in joint-carving
terms; but he also argues effectively that this position is difficult to support.
Although the epistemic value of joint-carving inquiry is, I think, a basic one
and hence not derivable from other values, I do think it can be supported in
several ways. First, the aim of joint-carving can be seen as having the same
source as the aim of truth: beliefs aim to conform to the world. Here is a
simplified but intuitive picture. The realist about structure thinks of the world
as coming “ready-made” with distinguished carvings. By analogy with the
notion of a mathematical structure, think of The World as a structure: a set E
of entities together with a set R of relations over E (think of the relations here
simply as ’tuples of members of E). Now, ignoring partial belief, it is natural
to think of the beliefs of a subject, S, as consisting of the representation of a
structure: the subject represents there being objects, Es, together with a set Rs
of relations over E. Given this picture, it is utterly natural to think of full
conformity to The World as requiring 〈Es, RS〉 to be identical to 〈E, R〉.
Conformity requires the believer to represent the structured world exactly as
it in fact is, and thus requires the represented relations Rs to be identical to the
world’s structuring relations R. Thus if belief aims to conform to the world,
and if belief and the world are both structured, belief aims not just at truth,
but also at the right structure—truth in joint-carving terms.
Second, we think of scientific discovery as satisfying the aims of inquiry
particularly well; why? Answer: it is because scientific discoveries are
phrased in particularly joint-carving terms. Relatedly, we think of truths that
are stated in extremely non-joint-carving terms—for example, the scrambled
propositions of section 2.6—as being comparatively worthless. Relatedly,
imagine (or recall) first coming to believe that morality, beauty, justice,
knowledge, or existence is a mere projection of our conceptual scheme—that
the truth in these domains is conventional, subjective, or otherwise
nonsubstantive. Why does that feel so deflating; why does it diminish the
urgency of finding the truth; and why does it diminish the value of the truth
once found? Answer: though we might not put it exactly thus, our original
picture in these lofty domains is that of joint-carving. Morality, beauty, and
the rest are built into the nature of things, we naïvely think, rather than being
mere projections. Giving up on objectivity means giving up on joint-carving,
and hence diminishes the value of truth.
Third, consider the following series of scenarios. Scenario 1: the physical
world is pretty much the way we think it is; it includes physical objects in
addition to spacetime. Scenario 2: the physical world consists of nothing
more than propertied points and regions of spacetime. Scenario 3: the
physical world consists of a wave function in a massively dimensional
configuration space. Neither three-dimensional space nor four-dimensional
spacetime exist, fundamentally speaking. Scenario 4: our ordinary beliefs are
caused by The Matrix, a computer simulation that directly stimulates our
brains while our bodies lie in stasis.22 Scenario 5: I am a disembodied brain
floating in an utterly empty space; the changes in my brain that give rise to
my “mental states” happen purely by chance. In Scenario 1 my ordinary
beliefs about myself and my surroundings are true. The same holds in
Scenario 2, I think, though some would disagree. This is less clear in
Scenario 3, since the structures in the world that ordinary beliefs would need
to pick out in order to be true—patterns in the wave function—are so far
from the joints in reality. Matters are worse in Scenario 4: our ordinary
beliefs would be true only if interpreted as picking out aspects of the
computer program, which are (we may stipulate) quite distant from reality’s
joints in the world that houses the program. And once we get to Scenario 5, if
I can be said to have any mental states at all, nearly all of them are clearly
untrue. They would be true only if they had contents defined on the world’s
empty space (or parts of my brain); but no assignment of such content could
be regarded as being more correct than other assignments on which the
mental states would come out false. Now, in this series of scenarios, the
match between our beliefs and reality’s joint-carving structure is gradually
eroded. The erosion is severe enough to disrupt truth only late in the series.
But even earlier in the series, at stage 3, say, much of what we care about has
been lost, even if what we normally say is still true. We are a partial
intellectual failure if we live in configuration space or The Matrix, even if we
believe truly. Moreover, even if the transition from truth to falsity in our
ordinary beliefs is abrupt, what we care about in inquiry seems to be more
continuously eroded in the series. These facts suggest that what we care about
is truth in joint-carving terms, not just truth.
The thesis about the value of joint-carving inquiry unifies many of the
connections I have forged between joint-carving and other concepts. The
unification takes the form of a recurring theme, rather than a derivation from
first principles. Here are a few sweeping lines of thought.
Duplication: Lewis defined duplicates as objects whose parts have the
same natural properties and stand in the same natural relations. But why care
about duplication? A partial answer emerges from this section’s thesis: if it’s
good to classify objectively alike things together, then duplication should be
of interest to us. (This is only a partial answer, because it does not address
why duplicates in particular, rather than merely objects that share natural
properties, would be of interest. In particular, why are we interested in
objects whose parts have the same natural properties and stand in the same
natural relations? The answer here, I think, somehow concerns the distinctive
status of parthood in our thought.23)
Explanation: here we have a direct connection to the thesis of this section.
Explanation is an epistemic achievement. There are many views about
explanation, and no doubt many species of explanation; but on all views,
explanation is an attempt to improve our epistemic position in some way, to
make the world more intelligible. But the epistemic achievement will be
diminished if cast in non-joint-carving terms, given this section’s thesis.
Laws: given the Lewisian theory of laws, on which laws must be phrased
in joint-carving terms, it follows that laws are propositions in which we ought
to take an interest. Concern with what science says is not epistemically
optional.
Reference: Lewis himself defended reference magnetism by saying that it
is constitutive of reference that reference goes to the most natural candidate.
But this just invites the question of why we should care about reference.
What’s better about reference than “reference*”, which does not always go to
the most natural candidate? Answer: reference carves better at the joints than
reference*; given the thesis of this section, that is why we should care about
it.24
Induction: in Bayesian terms, the problem of induction is to characterize
the correct prior probability distributions. Now, here is a way of thinking
about this notion of correctness. In worlds like ours, human beings have
survived, and thrived, in part because of how they formed beliefs about their
environment. Taking for granted that they updated by conditionalization
(more or less), they survived because they had appropriate priors. So an
explanation of why humans survived will cite certain general features of
human prior probability functions. These general features cannot be too
specific if they are to explain why humans in general (as opposed to on this
continent or that) survived; but they cannot be too general either if they are to
robustly explain the survival. They will surely include such features as this:
humans tend to project reasonably joint-carving properties (like blue) more
than badly non-joint-carving properties (like grue). Note further that having
these features explains why humans survive in worlds like ours. If our world
contained few regularities involving joint-carving properties, then projecting
such properties wouldn’t lead to survival. So: there are general features of
human priors that explain why we survived. These features, according to this
line of thought, just are the features that make priors correct. Correctness just
is: having the features that explain humans’ survival. This is just one crude
story, at best incomplete and at worst on the wrong track; but my point could
be made under refined or alternate stories. My point is this: epistemic notions
—such as the notion of a correct prior probability distribution—have their
source in what explains certain facets of our epistemic success. Thus those
notions must respect nature’s joints in certain ways, since explanations quite
generally must be cast in reasonably joint-carving terms. Given the crude
story imagined above, for example, the features of our prior probability
distributions cited by the explanation of our survival would need to be cast in
reasonably joint-carving terms. (Similar points could be made under alternate
stories in which the explanation of our epistemic success cites features of the
present, say, rather than our evolutionary past.)
Space and time: the intrinsic structure of space and time is given by the
spatiotemporal notions that carve at the joints. But why is uncovering this
structure the goal of science? Because science—like all inquiry—aims to
correctly represent the world, and this requires representing the world “in its
own terms” it requires carving at the joints.
Objectivity: why do we prize substantivity and objectivity, and downgrade
the subjective, conventional, nonsubstantive? Because it’s better to think in
joint-carving terms; and the more one does, the more one’s meanings “stand
out from the metasemantic background”, yielding substantive, objective
claims.

4.6 Objectivity of structure

We have uncovered a web of connections between structure and various
notions. This web of connections yields the primary argument against
Goodman’s claim that structure is merely the projection of our interests or
biology: subjectivism about structure leads to subjectivism about the other
notions in the web. We could not formulate an appropriately objective form
of Bayesianism. We could not rebut conventionalists about physical
geometry. We could not believe in objective semantic determinacy.
But the most significant fallout from Goodmania, to my mind, arises from
structure’s connections with epistemic value and with objectivity.
Epistemic value: joint-carving languages and beliefs are better. If structure
is subjective, so is this betterness. This would be a disaster. Recall section
2.6. If there is no sense in which the physical truths are objectively better
than the scrambled truths, beyond the fact that they are propositions that we
have happened to have expressed, then the postmodernist forces of darkness
have won.
Objectivity: whether questions are substantive, nonconventional,
objective, and so on, depends on whether they are phrased in terms that carve
at the joints. Given subjectivism about structure, we would have subjectivism
about substantivity, depth, conventionality, objectivity. No discourses would
be objectively objective. Another disaster.
The knee-jerk realist thinks that the world is “out there”, waiting to be
discovered rather than constructed—all that good stuff. Everyone agrees that
this picture rules out views according to which all truth is mind-dependent in
the crudest ways, but it requires more. After all, under the projectivist
semantics (section 4.4) one can truly say that “the mountain would still have
been beautiful even if humans had never existed”. The realist picture requires
the “ready-made world” that Goodman (1978) ridiculed; it requires the world
to really be as physics says; it requires objectivity; it requires objective
distinguished structure. To give up on structure’s objectivity would be to
concede far too much to those who view inquiry as being merely the
investigation of our own minds.

1 Related work includes Chalmers (2011), Fine (2001) on nonfactualist discourse, and Sidelle
(2007) on verbal disputes.
2 Compare Williamson (2007, chapter 2).
3 Though they may have no “determinate” answers, given an epistemic reading of ‘determinate’.
4 See Blackburn (1993); Fine (2001); Rosen (1994); Wright (1992).
5 The account requires there to be some way of making sense of quantification over “meanings”, but
not that such quantification be fundamental.
6 How did we select the candidate? Perhaps it is the most charitable candidate given our use of E;
perhaps only it is relevantly causally related to our use of E…—the answer depends on the truth about
metasemantics.
7 Thus I deny Gideon Rosen’s (1994, p. 301) claim that “if the facts in the contested class can
simply be read off in a mechanical way from the facts in an uncontroversially objective class, then there
can be no grounds for denying the same status to facts in the contested area.” Adherence to this
principle leads Rosen to his pessimistic conclusion that “it adds nothing to the claim that a certain state
of affairs obtains to say that it obtains objectively” (p. 279).
8 Sidelle (2009) discusses a similar sort of conventionality.
9 Skow (2010, section 4) gives a similar account.
10 Or if any carves better than the others, it is of the wrong sort for ‘inch’.
11 As with substantivity in 4.2, I’m neutral on whether to relativize conventionality to context, or to
say instead that although (C) is conventional in every context, for some length, l, what the disputants
really care about in the first context is not (C), but rather the nonconventional sentence ‘the screen has
length l’. (The variable ‘l’, under a given assignment, is intended to lack alternate candidates.)
12 The geometric conventionalism of philosophers like Reichenbach should not be confused with
conventionalism about units of measure. The former concerns even unitless length predicates such as
Tarski’s ‘congruent’.
13 Not in all cases, however. Some such doctrines are best understood as claims of content-
conventionality. Conventionalism (or social constructionism) about works of art, artifacts, or gender
and race might be construed as the view that sentences about these subject matters express propositions
that are in some sense about social conventions. Were our conventions different or nonexistent, there
would exist different (or even no) sculptures, tables and chairs, men and women, and so on, since what
it is to be a work of art, etc., involves social conventions. (Compare Haslanger’s (1995, p. 98) notion of
constitutive construction.) Other such doctrines are not clearly instances of any of my three sorts—the
logical conventionalism discussed in section 6.5, for instance.
14 I am using ‘projection’ nonstandardly; it usually signifies the mistaken attribution of mental
features to the external world.
15 Compare Street (2006, section 7).
16 If predicates are individuated by the properties they express, then we cannot say that each
community uses the same predicate ‘beautiful’; what we must say instead is that each uses a predicate
that plays the role that ‘beautiful’ plays in our language. Allow me the liberty of speaking of each
community as having “aesthetic” predicates, “aesthetic” attitudes, and so on.
17 This variation in the meaning of ‘beautiful’ is not just the mundane sort of variation that is
possible because of symbol–conventionality In each language, ‘beautiful’ plays the same role; it is a
word for the things picked out by the aesthetic attitudes of that language’s speakers.
18 Moral realists sometimes implicitly support such a privilege, even if their explicit focus is on the
contents of moral sentences (see, for instance, Boyd (1988) on homeostatic property-clusters and
Railton (1986) on feedback). Note also that the appropriateness of the language and imagery of realism
and objectivity comes in degrees. Even if contents for moral language are selected by facts about us,
morality seems more realistic and objective if those facts are counterfactually robust and universal
across different societies—if they reflect the human condition rather than historical accident.
19 What if our actual values are so hard-wired into our brains that it would be difficult or even
impossible to adopt others? I still want to count other values as picking out “candidates” for the term in
question; those candidates are not “semantically alien” even if we could not adopt them. Thus it’s not
always right to say (as I have been) that “we could just as easily” have adopted alternate candidate
meanings.
20 See Lewis (1986c, introduction), Lewis (1994). In this respect, Lewis is just like the traditional
regularity theorist. Each agrees that the world is fundamentally anomic; the difference is that Lewis
does a better job of approximating in extension the ordinary notion of a law of nature. Matters are very
different for Armstrong (1983), whose relation of nomic necessitation is a universal, and so is part of
reality’s fundamental structure.
21 Thanks to Steve Steward.
22 Compare Chalmers (2005).
23 See Sider (2007b).
24 Thanks to Robbie Williams.
5 Metametaphysics
Metametaphysics is inquiry into the status of metaphysics. It is of the nature
of the beast that one is led to ask: are metaphysical disputes substantive? Are
they objective, genuine, deep? Or are they nonsubstantive in some way:
conventional, subjective, merely verbal or conceptual? Must there be a fact of
the matter about who is right?
The answers to these questions depend on which metaphysical dispute is
at issue. The crucial factor is whether the dispute is phrased in terms that
carve at the joints. This connection to metametaphysics is the final part of the
role I envisage for the notion of structure.

5. 1 The challenge of metaphysical deflationism

Metaphysics has always had critics. The most extreme base their critique on
sweeping views about language (logical positivism, ordinary language
philosophy) or knowledge (naïve empiricism). But such views are in trouble
on independent grounds. Their oversimplified conceptions of how we make
semantic and cognitive contact with the world notoriously threaten the
science of unobservables as much as they threaten metaphysics.
A more formidable challenge comes from a more modest critic, whom I’ll
call a (metaphysical) deflationist.1 A deflationist about question Q says that Q
is in some sense merely verbal or conceptual. Its answer (if it has an answer)
is determined by linguistic or conceptual rules. What is at issue is not “the
world”, but rather us—how we think and talk. (We will need to refine this
rather vague characterization, but it will do for now.) This critique, moreover,
is based, not in a sweeping conception of language or epistemology, but
rather in considerations specific to Q.
My deflationist has a favorite “go-to move”. The move is schematic, and
can be made in service of various antimetaphysical conclusions, but in each
case its form is the same:
1. The deflationist observes a certain metaphysical dispute, in which one of the contested views is
expressed by a certain sentence S.
2. He argues that there is an interpretation of the language of S—a way of assigning meanings to the
sentences of that language—under which everyone can agree that S is true.
3. And he argues for a certain parity between this and rival interpretations.

The go-to move may be summarized thus: “There’s a perfectly good way to
talk under which S is clearly true.”
Before discussing the meaning of ‘parity’ in step 3 and how the move
might lead to antimetaphysical conclusions, let me give some examples.
Confronted with a dispute over whether the relation of part to whole is
reflexive, the deflationist might point out that the following defines a
reflexive relation:
x is part* of y =df x is part of y or x = y.
Even if ‘part’ in English is not reflexive, ‘part*’ is. So under an interpretation
of English words that is just like actual English except that ‘part’ means part*,
the sentence ‘Everything is part of itself’ obviously comes out true.
Moreover, this interpretation is very similar to English since ‘part’ and ‘part*’
have exactly the same extension except perhaps for pairs of the form 〈x, x 〉—
parity.
Second example: with respect to the same dispute, the deflationist might
construct a related interpretation, also very similar to English, in which
‘Nothing is part of itself comes out true, by letting ‘part’ mean part**:
x is part** of y =df x is part of y and x ≠ y.
Third example: confronted with a dispute over whether holes exist, a
deflationist might construct an interpretation of quantificational sentences in
which ‘There is a hole in object x’ means that x is perforated. Even an
opponent of holes will agree that some such sentences are true in this
interpretation, since the opponent agrees that some things are perforated.
The deflationist can use the go-to move within various dialectical
strategies. Many of the strategies appeal to the fact that meaning is largely
determined by use. For example, a large part of why ‘bachelor’ means
(something like) unmarried adult male is that we tend to call something a
‘bachelor’ if and only if it is an unmarried adult male. This is oversimplified;
but rather than trying to refine it, let’s just employ the code: “Our use of
‘bachelor’ favors the hypothesis that it means unmarried adult male.” Here,
then, are some of the strategies:
Common-sense strategy Suppose a metaphysician argues for a noncommon-
sensical position. Argue first, using the move, that sentences expressing a
more commonsensical rival position can be truly interpreted. Argue,
second, that our use of the crucial terms favors the commonsensical
interpretation over any interpretation on which the noncommonsensical
position comes out true. Conclude that the noncommonsensical position
is not true. (For example, our use of sentences like ‘There is a hole in that
piece of cheese’ clearly favors interpretations under which the sentence
comes out true whenever the piece of cheese in question is perforated. So,
nominalism about holes is false—and this conclusion was secured simply
by reflecting on language.)
Indeterminacy strategy Argue first, using the move multiple times, that
each of a range of views about some metaphysical question comes out
true under some interpretation. Argue, second, that our use of the crucial
terms does not favor any of these interpretations over the others.
Conclude that it is indeterminate which view is true. (For example, our
use of ‘part’ favors neither part* nor part** over the other; so it’s simply
indeterminate—and therefore pointless to debate—whether parthood is
reflexive.)
Deflationary strategy As with the indeterminacy strategy, argue first that
each of a range of views about some metaphysical question comes out
true under some interpretation. But now, rather than taking a stand on
whether actual usage favors one interpretation, simply conclude that the
question is “merely conceptual”, because which view is true depends on
which interpretation is favored by our usage. (For example, both the
question of whether parthood is reflexive, and the question of whether
there are holes, are merely conceptual. Since each view about these
questions comes out true under some interpretation, the only question is
which interpretation fits English usage.)
I think of the third strategy as the core of metaphysical deflationism, since it
seems presupposed by the first two strategies, and is potent even if
unaccompanied by either of the first two.
The deflationist’s arguments are pretty unimpressive if step 3 of the go-to
move is omitted. Let p be a particle in some distant galaxy. The sentence
‘particle p is an electron’ is a paradigm of the kind of sentence for which no
form of deflationism is true. And yet, we can construct an interpretation on
which it is true and an interpretation on which it is false, each of them very
similar to English: let the first be like English except with p removed from
the extension of ‘electron’, and let the second be like English except with p
added. Step 3 is crucial; the rival interpretations must in some sense be on a
par with one another and with other competing interpretations (such as
English).
If the deflationist is a realist about structure then he can construe step 3 as
requiring at least that the constructed interpretation must carve at the joints
(in the right way—recall section 4.2) as well as its rivals. The rival
interpretations of ‘electron’ do not lead to deflationism about whether particle
p is an electron because, whereas the actual, English, meaning of ‘electron’
carves at the joints, one of the two constructed meanings does not. But when
a deflationary stance is correct, the rival interpretations carve at the joints as
well as one another and as well as English.
A deflationist who rejects realism about structure, on the other hand, will
need some other way to explain why the go-to move has a deflationary
upshot for metaphysical questions but not for the question of whether particle
p is an electron; and he will need some other way to construe step 3. It is hard
to see how this could be done.2 So I think the way to defend a targeted
deflationism is to be a realist about structure. At any rate, this is the form of
deflationism on which I’ll focus.
The deflationist should also, I think, take step 3 as requiring that the rival
interpretations not be “semantically alien”, in the sense of section 4.2. That is,
although the rival interpretations might determinately differ from English,
speakers using those interpretations should use the words in question to
accomplish the same semantic tasks as do English speakers. For the
deflationist about a question Q wants to conclude that the answer to Q
depends on how we use words, but not in the trivial sense in which all truth
depends in part on meaning. The idea should rather be that Q’s answer
depends on usage in the nontrivial sense that it depends on which nonalien
interpretation is actual.3
Thus what the deflationist is saying is in essence that Q is nonsubstantive,
in the sense of chapter 4. The deflationist’s nonalien interpretations result
from multiple candidates for the crucial terms in Q. For example, part* and
part** are candidates for ‘part’, and generate interpretations under which
different answers to the question “Is parthood reflexive?” come out true. If
each carves at the joints as well as the other (and as well as any other
candidate for ‘part’), then this question is nonsubstantive. As we saw, this is
not to say that Q is about how we use words. The question of whether
appletinis are martinis is not about words, and yet it too is nonsubstantive in
the same sense.
What are the ways for a metaphysician to respond?
Sometimes metaphysicians should embrace deflationism. And sometimes
this requires admitting that the debate is just silly, and should be
discontinued. (Even a partisan needn’t fight every battle.) But not all
nonsubstantive debates are silly. As we saw in section 4.2, questions that are
nonsubstantive in my metaphysical sense may yet be conceptually deep and
thus important if their answers reveal important facets of our conceptual
scheme. Perhaps questions about causation are like this.
But sometimes the true believer in metaphysics will want to oppose the
deflationist. This is particularly true when the believer wants to defend a
revisionary position, since such positions are hard to regard as reflecting our
ordinary conceptual scheme.4 It is also true when the believer regards herself
as doing fundamental metaphysics.
In the case of fundamental metaphysics, the most straightforward way to
resist deflationism is to claim that the crucial expressions in the debate carve
perfectly at the joints. As we saw in section 4.2, the relation between
substantivity and joint-carving is a complex one. However, a sufficient
condition for substantivity (or near enough, anyway) is that the dispute be
cast in perfectly joint-carving terms. Such a dispute concerns the nature of
fundamental reality. In such a dispute, the existence of alternate
interpretations has no more deflationary import than it had in the question of
whether particle p is an electron.
(Why is the condition only “near enough” sufficient? Because there might
be multiple joints in the vicinity, and because the joints might be the “wrong
sort”, in the sense of the third refinement of section 4.2. I’ll mostly be
ignoring these complications.)
There are other ways to resist the deflationist, but they are unappealing.
One might try to argue that the alternate interpretations simply do not exist.
For example, the deflationist about the ontology of holes claimed to produce
an interpretation under which ‘There is a hole in x’ means that x is perforated.
But the dispute over holes does not involve that sentence form alone; it
involves sentences with many different syntactic forms, for example:
There is a circular hole in x.
 There are fifteen holes in x.
The hole in x is identical to the hole in y.
In light of this, it might be argued that there is no way to give a general
antinominalist interpretation of quantification over holes. But this is a slender
reed on which to rest one’s hopes. (I argue in chapter 9 that in the case of the
ontological deflationist, the hope is in vain.)
It is also possible to grant the existence of multiple equally joint-carving
interpretations, but claim that those interpretations are semantically alien and
thus not candidates. Consider the question of whether “Magnesium is more
plentiful on Earth than carbon.” This sentence comes out true if ‘magnesium’
means oxygen, but this does not support deflationism about the question
since this alternate meaning for ‘magnesium’ is not a candidate (despite
carving at the joints). The semantic goal we are trying to achieve with
‘magnesium’ is not so unspecific that it could just as well have been achieved
by letting ‘magnesium’ mean oxygen. However, this kind of response to
deflationism seems inapplicable in cases of philosophical interest. For what is
distinctively puzzling about philosophical questions is closely connected with
the fact that we can imagine ourselves speaking in any of a number of
different ways using the disputed term, without thereby being semantically
alien.
One might instead make a big deal out of the fact that the candidate
meanings for the crucial term differ from its actual meaning:
“Who cares whether there are ways to interpret ‘There is a hole in x’ so that it comes out true? What I
care about is whether there is a hole in x; and if you define ‘There is a hole in x’ to mean that x is
perforated, it no longer says that there is a hole in x!”

This performance is oblique, since it rejects none of the deflationist’s claims.

It does not challenge the claim that the interpretations exist, nor does it deny
that they carve at the joints equally well, nor does it claim that they’re
semantically alien. And the deflationist never claimed that ‘There is a hole in
x’ still means that there is a hole in x when it means that x is perforated. What
is conveyed by the agitated speech is that the “deflationist’s” claims do not
deserve the name ‘deflationism’, that the status quo in metaphysics is
unthreatened by the existence of the interpretations.
This is a hard attitude to maintain. Most metaphysicians at least
sometimes think of themselves as not being engaged in conceptual
archaeology. (From this point of view, Peter Strawson’s (1959) “descriptive
metaphysics” is a near oxymoron.) Instead, they think of their project as
being rather like speculative science. This self-conception isn’t always
articulated, but it is often subconsciously present. It reflects itself in the
willingness to take noncommonsensical positions seriously, and to be guided
by theoretical virtues that are prized in science (such as “simplicity”). This
self-conception cannot survive the admission that rival answers to one’s
question come out true under equally joint-carving candidate interpretations.
Imagine we live on the shore of a gigantic lake, around which there are
numerous other isolated linguistic communities. None of these is
semantically alien with respect to the others, but the different communities do
use ontological language differently: in one, ‘There are holes’ is true, in
another it is false; in another ‘There are numbers’ is true; in another it is
false; and so on. And suppose further that none of these languages carves
better at the joints than any of the others. If we metaphysicians learned of all
this, it would seem perverse for us to continue to regard our ontological
questions with the original, quasi-scientific attitude. Granted, in some of
these other languages, ‘There are holes’ does not mean that there are holes.
But so what? Speakers of those languages could make parallel true statements
about us in their language: “Your sentence ‘There are holes’ does not mean
that there are holes.” Look out over the lake—the parochiality of our
conception of there being holes is staring us in the face. The metaphysical
attitude requires a more transcendental view.

5.2 Personal identity, causation

Let’s make all of this a little more concrete. Consider a group of proponents
of the doctrine of temporal parts who disagree over the criterion of personal
identity. Their dispute is over which relation between person stages is the
“unity relation” for persons—that is, over which relation holds between all
and only the person stages that are part of some one continuing person. Some
of them think that the unity relation is that of psychological continuity; others
think that it is the relation of bodily continuity (i.e., spatiotemporal continuity
under the sortal: human body).
Never mind which group has the right answer. Let us instead ask the
metameta-physical question: Is the dispute substantive?
A deflationist about personal identity could argue as follows. First, there is
no perfectly fundamental unity relation over person stages. There are
numerous relations over person stages in the vicinity: the relation of
psychological continuity, the relation of bodily continuity, and so on. But
none of these carves at the joints perfectly, and none carves better than the
others. Moreover, these are candidates for our talk of personal identity.
Neither a community who spoke of personal identity as being governed by
psychological continuity, nor a community who spoke of it being governed
by bodily continuity, would be semantically alien. So the question of personal
identity is nonsubstantive. There is no objective, substantive, deep answer as
to whether I would survive certain transformations—those in which my
psyche is transferred to a new body, for example, or in which I abruptly lose
all my memories. Which answers are correct is largely a question about our
conceptual scheme, not a question of reality’s fundamental structure (and our
conceptual scheme might be silent about some cases).5
For another example, return to causation. If there is a perfectly
fundamental relation of causation, then there must be objective, deep answers
to questions like: Is causation two-place or four-place? Does it relate facts or
events? But otherwise, deflationism about some questions about causation
will presumably be true, since linguistic communities differing over whether
‘cause’ expresses a two-place or four-place relation, for example, do not
seem linguistically alien. The questions about causation may have no answers
at all; and insofar as they do have answers, those answers will be a mere
reflection of our concept of causation.
The distinction between conceptual and metaphysical substantivity
(section 4.2) is particularly important in metametaphysics. Even if questions
about personal identity and causation are not metaphysically substantive, they
clearly are conceptually substantive, given how deeply the concepts of
sameness of person and causation are embedded in our conceptual scheme.
These concepts play a role in many central aspects of our thought: thought
about ourselves, moral responsibility, deliberation, control, prediction,
explanation, and myriad others. Even if this entire edifice rests on
metaphysical sand, understanding its inner workings is a crucial part of
understanding ourselves.
Disputes over causation, personal identity, and the like are conceptually
deep, even if metaphysically shallow; this is a significant part of their
philosophical interest. Conversely, and ironically, some of the metaphysically
deepest disputes—certain ontological disputes, for example (see chapter 9)—
are conceptually shallow in that they have few implications outside rarified
metaphysics. This, I suspect, contributes to the common distrust of those
disputes.6

5.3 The metaphysics room

Sometimes fundamental metaphysics can be conducted in ordinary language.
But not always. Metaphysicians need a plan B.
Suppose we attempt to ask a question of fundamental metaphysics using
some ordinary, natural-language expression E. Suppose further that there is
some joint-carving meaning m “in the vicinity” (see below). The problem is
that E might not mean m. Various metasemantic scenarios discussed in
section 3.2 could have this result; here are three. 1. Reference magnetism is
false in all its forms, and E is not a theoretical term—like ‘amulet’ it is not
“trying” to stand for a joint-carving meaning. 2. Some form of reference
magnetism other than simple charity-based descriptivism is true, and E is not
a theoretical term. 3. Simple charity-based descriptivism is true, but m fits
badly with the use of E. (According to simple charity-based descriptivism,
the correct interpretation must maximize the combination of charity and
eligibility; thus a highly joint-carving interpretation might nevertheless be
incorrect if it is too uncharitable. Not all words carve at the joints, after all!)
If E does not stand for m, then it might instead stand for some non-joint-
carving meaning that reflects a more-or-less arbitrary choice of usage that our
linguistic community has made—a choice of one amongst a range of equally
non-joint-carving candidate meanings. Thus the question would be
nonsubstantive in that its answer would turn on linguistic usage, not reality’s
structure. (More accurately, the answer would be actual-verdict
nonsubstantive in the sense of section 4.2.)
That is, the ordinary, natural language question, phrased in terms of the
ordinary, natural-language expression E, would be nonsubstantive. But we
could discard E, and enter the metaphysics room, so to speak. We could
replace the ordinary expression E with an improved expression E* that we
stipulate is to stand for the joint-carving meaning in the vicinity. The question
we ask in the metaphysics room, cast in terms of E* rather than E, is
substantive. Indeed, it is superior to the original question, for it concerns
reality’s fundamental structure, rather than its merely conventional or
projected aspects. This is plan B.
Early on in philosophy we are taught not to abandon ordinary language on
the battlefield. If a novel language had to be stipulated in order to carry on a
debate, we’re warned, there could not be open questions about what is true in
that language, since the answers would need to be settled by stipulation.
Arguing about what is true in the novel language would be like arguing about
how the queen ought to move in a new version of “chess” whose rules are
unconstrained by the existing rules. The traditional debate over whether
freedom is compatible with determinism, for example, would be trivialized if
we had to stipulate a meaning for ‘free’. If we stipulated that ‘free’ means
‘undetermined by the laws and past’ then there would be nothing worth
debating: “freedom” thus understood is obviously incompatible with
determinism. And if we instead stipulated that ‘free’ means ‘not in chains’,
then again we would have nothing worth debating; “freedom” thus
understood is obviously compatible with determinism. The only way to have
a meaningful debate, so we are taught, is to abandon such stipulations, and
mean by ‘free’ … freedom!—freedom in the ordinary sense.7
Abandoning ordinary language is indeed often a bad idea, but when it is,
that is because there is no other way to anchor the debate, no other way to
explain the meanings of the crucial words without trivializing the debate. But
joint-carving meanings give us another anchor. We can then introduce new
words with a minimum of semantic pressure, with only minimal stipulations
on their behavior. Of course the stipulations must be strong enough to pick
out unique meanings (by picking out the relevant “vicinity”); but this aside,
we stipulate nothing that settles disputed claims. We nevertheless succeed in
securing unique meanings because joint-carving meanings are sparse—there
is usually at most one joint-carving meaning in a given “vicinity”. (To
suppose otherwise would be to needlessly attribute complexity to the world.)
Picture introduction of terms with minimal semantic pressure as something
like definition by ostension. We are saying “Let us introduce words for these
meanings, so that we can disagree about how they behave.” If there are
indeed (unique) joint-carving meanings in the vicinity—the anchor—then the
words in question will have determinate meanings even though the
stipulations were so minimal. The answers to the disputed questions will not
be settled by the stipulations, and inquiry into those questions—questions
about reality’s fundamental structure—will be worthwhile.
Let’s consider an example. Suppose there is a fundamental relation C that
is a lot like causation, except that it holds only between events at the
subatomic level. Macro events, such as the throwings of rocks and the
breakings of windows, never stand in C. Now, the ordinary English term
‘cause’ may well not mean C. For i) C fits terribly with ordinary usage of
‘cause’ (or at least with usage of simple causal sentences such as ‘the
throwing of the rock caused the window to break’); and ii) ‘cause’ may well
be a nontheoretical term in English. Rather than standing for C, ‘cause’ may
instead stand for that non-joint-carving relation that best fits our usage of
‘cause’. A debate involving ‘cause’ would then not be (actual-verdict)
substantive. But we could enter the metaphysics room, and coin a new term,
‘cause*’, for the joint-carving relation in the vicinity of causation. ‘Cause*’
will stand for C—fundamental causation, we might call it—and our new
debate about causation* will be substantive.
How, exactly, will we fix the meaning of ‘cause*’? It is to stand for the
joint-carving meaning “in the vicinity of” causation. Thus its metasemantics
should be like that of ‘cause’ except that joint-carving is paramount. Any
general metasemantic presumption that non-joint-carving candidates may be
assigned is suspended in the case of ‘cause*’; carving perfectly at the joints is
an absolute requirement. ‘Cause*’ should stand for that meaning which i)
carves at the joints (perfectly); ii) fits our use of ‘cause’ better than any other
joint-carving meaning; and iii) fits our use of ‘cause’ well enough. If nothing
satisfies all three of these conditions—if, that is, there is no fundamental
causation, or if there is more than one sort—then ‘cause*’ stands for nothing,
and debates about causation* are ill-posed. Conditions ii) and iii) are
admittedly vague, but harmlessly so. What might we actually do to coin a
new term with this metasemantics? I see no reason to deny that the following
performance would do the trick: “Let ‘cause*’ be a theoretical term for that
perfectly joint-carving relation (assuming there is such a thing) that is in the
vicinity of the ordinary notion of causation, but which (since it carves at the
joints) may differ somewhat from that ordinary notion.”
The metaphysics room gambit requires that successful stipulations of the
envisaged sort be possible. It may be objected that new languages or terms
can be introduced only by translation into existing natural languages or
terms.8 But that can’t be right. Natural languages themselves had to
bootstrap; they had to somehow latch onto the world in the first place. So
why can’t we bootstrap now? It may be thought that the stipulation could not
succeed if reference magnetism is false, or worse, if some radically
nonreferential conception of meaning—some form of deflationism or
inferentialism, perhaps—is true. But even such views must allow for the
introduction of new theoretical terms in science. If they don’t, then they are
thereby refuted. And if they do, then given realism about joint-carving they
surely also allow the introduction of theoretical terms in the metaphysics
room. The cases really are parallel: in each case we introduce new terms with
minimal semantic pressure, we intend to mean “whatever is in the vicinity”,
and semantic determinacy is achieved (if it’s achieved) primarily because the
world contains an appropriate meaning, not primarily because of facts about
us.
The gambit also requires that semantic interpretation—at least for terms
stipulated to carve at the joints—be under our control. But it does not require
that grammar be under our control in the same way. Consider, for example,
the fact that natural language quantifiers are grammatically binary, whereas
first-order logic’s quantifiers are monadic. It might be argued that for this
reason, the first-order quantifiers are in some sense grammatically impossible
for us. Even if we try to pronounce “∃x” in a monadic-looking English form,
“For some x …”, we nevertheless continue, despite ourselves, at some
psychologically basic level, to represent the quantification binarily: “For
some object x…”. Binary quantificational grammar is built into our minds in
a way that’s difficult or even impossible to change. But even if this view is
correct, it is no obstacle to introducing the language of first-order logic in the
metaphysics room and giving it a monadic semantics. The formal sentences
of this language would describe fundamentally monadic facts; it’s just that
we couldn’t think about those facts “natively”. There would be a mismatch
between the structure of the facts and the structure of our thoughts (though
not between the facts and the sentences of the formal language).

5.4 Substantivity in nonfundamental disputes

The simplest way to regard a dispute as substantive is to regard it as being
phrased in terms that carve perfectly at the joints. But sometimes one wants,
instead, to regard a dispute as substantive because it concerns reality’s less-
than-perfectly fundamental joints.
To illustrate, it is natural to wonder whether anything substantive is at
stake in the dispute over the location of the semantics/pragmatics border.
Everyone agrees that context supplies an enormous amount of information
that is relevant to communication. But is that information part of the
semantics or the pragmatics? For there to be any interesting issue here, we
must fix the notions of semantic and pragmatic without begging any disputed
questions. For example, if it is to be an open question whether the semantic
contents of sentences are (complete, truth-conditional) propositions, we must
not simply define the semantic content of a sentence as a proposition that is
conventionally associated with that sentence.
To secure substantivity, the debate should be regarded as being about the
way to carve the subject matter of communication at its natural joints—about
the form that a good theory of communication ought to take. Two participants
in this debate might advocate their positions as follows. Participant one: “A
good theory of communication divides the processing of communicated
information into two stages. At stage 1, a fairly simple level of content is
associated with each word. This level of content is memorized by the
competent speaker. Stage 2 combines the output of stage 1 with all the
information provided by the context, resulting in truth-conditions of the
utterance. In computing stage 2 content, the speaker appeals to her general
world-knowledge. Further, stage 1 associates a complete proposition—
something with truth-conditions—with each grammatical sentence.”9
Participant two: “I agree about the two stages, but I disagree with your final
claim. Stage 1 does not associate a complete proposition with each
grammatical sentence; sometimes it instead associates what Kent Bach
(1994) calls a ‘proposition radical’.” We can think of this dispute as being
over whether semantic contents are always complete propositions, since we
can think of stage 1 as semantics and stage 2 as pragmatics. And the dispute
isn’t terminological since ‘semantic’ and ‘pragmatic’ have been picked out by
their role in a good theory of communication, and not by a definition that
begs the question of whether semantic contents are propositions.
The story I just told in effect picks out the semantics/pragmatics divide via
its role in the Ramsey sentence for a theory of communication. This pattern—
a term of philosophical interest is picked out by its role—recurs all over
philosophy.10 Think of debates over perception. How much is contained in
“the content of visual experience”, it is asked.11 Information about real-world
objects, or just their appearances? Causal information? This debate can seem
puzzling to outsiders; how is the meaning of ‘the content of visual
experience’ to be fixed so that the questions remain open? The answer must
be that the participants in this debate are attempting to carve the subject
matter of perception at its joints. A good theory of this domain, they believe,
will appeal to a certain notion of the content of visual experience; and their
debate is about that notion—the notion that appears in a good theory of
perception. (The role might be expansive. Some, for example, pin down the
notion of the content of experience by its role in epistemology)
Construing a dispute in this way doesn’t on its own secure substantivity,
because it might be that no joint-carving theory takes the shape that the
disputants envisage. In the dispute over the border between semantics and
pragmatics, a third participant might deny the claim shared by participants
one and two, that a good theory of communication has anything like the two
stages. If she is right, then in a sense there would be no substantive questions
about semantics and pragmatics, since the distinction is based on a false
theory. Either our uses of ‘semantic’ and ‘pragmatic’ are empty of meaning,
or they are massively indeterminate, or their meanings are determined simply
by how they are casually used in ordinary speech.12 (Even so, a substantive
three-way debate over the shape of a good theory of communication would
remain.)
So: substantivity, in these questions of perception and meaning, turns on
the nature of reality’s joints. (Likewise for questions of substantivity in the
special sciences, although this is less often in dispute.) But reality presumably
contains no perfectly fundamental perceptual or meaning structure. So we
must make sense of joint-carving in domains that are not perfectly
fundamental. We will revisit this issue of imperfect fundamentality in section
7.11.

5.5 A test case: extended simples

To illustrate my approach to metametaphysics, let us look in detail at the
recent dispute over “extended simples”.13 My guess is that many readers
suspect that this dispute is not substantive (as I myself did initially). Is this
suspicion justified?
“Extended simples” are defined as spatially extended objects without
proper parts. For example, if I am an extended simple then I have no proper
parts. Not only do I lack “philosophers’ parts” such as a “right half”, I also
lack such ordinary parts as a head, arms and legs, internal organs, molecules,
atoms, and subatomic particles. There is, in my vicinity, but a single thing—
me—which occupies an extended, person-shaped region of space.14 A
limiting case would be existence monism, in Jonathan Schaffer’s (2007;
2010c) terminology, according to which the entire world is a single extended
simple.15 The dispute is over whether there are, or could be, such things.
The status of this dispute is tied up with general questions about space and
ontology. Are ontological questions in general substantive? Do there exist
such entities as points and regions of space? If so, are objects in space to be
identified with points or regions? If not, how are objects in space related to
space? The answers to these questions affect whether disputes over extended
simples are substantive.
To illustrate, suppose for the sake of argument that the following answers
to the general questions—call them, collectively, the occupation picture—are
correct:
1. The notions of logic, mereology, and physical geometry (quantification, identity, parthood, sum,
point, region, …) carve perfectly at the joints.
2. Substantivalism is true: there exist points and regions of space.
3. Supersubstantivalism is false: objects in space (“inhabitants”) are not identical to points or
regions of space.
4. Spatial facts about inhabitants emerge from the holding of a perfectly joint-carving relation of
occupation between inhabitants and space.

Now, the question of whether there are any extended simples, given the
occupation picture, becomes a question about the pattern of instantiation of
the occupation relation:
Does there exist something that lacks proper parts but occupies more
than one point of space (i.e., occupies multiple points or a sum or
region of points)?
And given my metametaphysics, this question is substantive, since it is
phrased in terms that carve perfectly at the joints.
I have seen philosophers roll their eyes when extended simples come up.
They regard that topic as “spooky metaphysics” (in a bad sense, it would
seem)—as being somehow misguided. But what is the complaint?
In fact, there are a number of complaints that one could be making. The
list of possible complaints, I hope, challenges both the extended-simple
enthusiasts and the eye-rollers. The enthusiasts must check that they avoid
each complaint. The eye-rollers must check they can genuinely embrace one
of them, and they must specify which one that is (the eye-rolling tends to be
scattershot and uncritical).
One complaint is purely epistemic. It admits that the dispute is
substantive, but claims that the considerations offered by the enthusiasts are
inadequate to resolve it. We do not know whether there exist extended
simples, it is alleged, and the enthusiasts’ arguments aren’t helping. Such
allegations may or may not be right, but one cannot make them without
actually entering into the fray. (Eye-rolling tends to come from the sidelines.)
Also, this complaint does not demand that the enthusiasts cease and desist.
The enthusiasts might instead take it as an exhortation to do better.
A second complaint concedes that the question of whether extended
simples actually exist is substantive, but notes that some enthusiasts address
primarily the question of whether they possibly exist. This shift makes a big
difference to the tenability of deflationism about the debate. Given my
metametaphysics, a deflationist about an issue must locate a nonfundamental
term essentially involved in its statement. Once the issue becomes the
possibility of extended simples, it is open to claim that the nonfundamental
term is ‘possibly’. In particular, suppose that there are, in fact, no extended
simples. One could then claim that the question “Is it possibly the case that
there exist extended simples?” is nonsubstantive, on the grounds that its
answer is yes under some candidate meanings of ‘possibly’ and no under
other equally joint-carving candidates.
The theory of possibility to be defended in chapter 12 could sustain such a
claim. Very roughly, the theory says that for a proposition to be necessarily
true is for that proposition to be i) true; and ii) of the right sort. What is the
“right sort”? This is given by a list of properties of propositions, a list
including the property logical truth, the property mathematical truth, and
certain others. Moreover, the list is determined more or less arbitrarily, rather
than by some deep criterion. Our meaning of ‘necessity’ is tied to our list; but
other linguistic communities could choose different lists and thus different
meanings for ‘necessarily’. They wouldn’t thereby carve at the joints worse
than we do; they would just be different—like a linguistic community that
counted the pope as a “bachelor”. Given this theory of modality, different
candidate meanings for ‘necessary’ arise from different choices of lists. If the
true proposition that there are no extended simples falls under some of these
lists but not others, then ‘necessarily there are no extended simples’ counts as
true under some candidates and not others; and so, ‘possibly, there are
extended simples’ counts as true under some candidates and not others.16
Intuitively: extended simples would depart from actuality in certain ways; the
question of whether they are possible is the question of whether these
departures are too drastic; whether the departures are too drastic turns on the
metaphysically shallow question of what we have decided to mean by
‘possible’.
Our first two complaints interact. Given the second complaint, the only
substantive issue can be the question of the actuality of extended simples. But
much of the literature on extended simples has discussed only their
possibility. If this portion is omitted, the first complaint becomes more
powerful.
A third class of complaints results from rejecting one facet or another of
the occupation picture. An ontological deflationist, for example, who thinks
that quantifiers do not carve at the joints, could claim that the sentence ‘there
are extended simples’ is true under some candidate meanings for ‘there are’
and false under others, and that all such candidates carve at the joints equally
well. Thus the sentence is nonsubstantive.
There are many questions here. There is first the question of whether
ontological deflationism is true; I will argue in chapter 9 that it is not. There
is also the question of whether ontological deflationism is behind the eye-
rolling; I think that in most cases it is not. Most of the people I have caught
rolling their eyes are not skeptical of ontology in general (though perhaps
their eye-rolling is a sign of an underlying malady that would lead to
ontological deflationism, or worse, if not treated). And there is, finally, the
question of whether ontological deflationism is enough to deflate all of the
disputes in the vicinity. It’s hard to tell, because ontological deflationists tend
not to say what they think the fundamental structure of reality is; they merely
say what they think it isn’t (namely, quantificational). Once the deflationist
has developed his fundamental language—which must be able to completely
describe the world, including its scientific aspects—who knows? Perhaps we
will be able to raise, in its alien terms, a substantive question that is akin to
the question of whether there are extended simples.17
A fourth and final sort of complaint amounts to simply rejecting extended
simples—perhaps all extended simples, or perhaps only certain kinds. This
may not seem like a metametaphysical complaint, but it feels more like one if
the reasons for the rejection are sufficiently general. Consider, for example,
what Kris McDaniel (2003) calls “spanners”: extended simples that occupy
an extended region, rather than occupying each of the many points in that
region. One might object to spanners because they violate a sort of micro-
reductionism, namely, the view that all fundamental properties and relations
relate mereologically simple entities. (One might, in turn, base this on the
view that only mereologically simple entities exist, in the fundamental sense
of ‘exist’; see chapter 13.)
We have seen four different sorts of complaint about extended simples.18
Notice that, with the exception of the first, each presupposes a substantive
thesis of metaphysics. The second depends on the claim that there are no
modal joints to reality; the third depends on the claim that there are no
quantificational joints to reality; and the fourth assumes micro-reductionism.
No complaint wholly sustains the common attitude of the eye-rollers that the
extended-simple enthusiasts are making a methodological error. The
complaints mark a difference of opinion about metaphysics. As we will see in
the next section, this is a general feature of metametaphysical critiques.

5.6 Metametaphysics as just more metaphysics

One worry about metaphysics is that we have no way to answer metaphysical
questions. But is metaphysics so much worse off than the rest of philosophy?
19 Many scattershot critics move from the difficulties of metaphysical
epistemology to the conclusion that metaphysics is something like
meaningless, without realizing how close this comes to assuming a crude
form of verificationism. Often these critics have a blind spot: they are
verificationist when thinking about metaphysics, but not when thinking about
other matters (especially about their own bit of philosophy!). A sensible
attitude is that metaphysics, like much of philosophy, is just hard. Its
epistemology is hard too, but this is no cause for panic.
A different (though related) worry is the feeling that metaphysicians see
substantive issues where there are really just different equally good ways to
talk. It is on this strand—metaphysical deflationism—that I have focused.
Details aside, my crucial claim has been that a sufficient condition for
substantivity is being cast in joint-carving terms. An important consequence
is that metametaphysical critiques are distinctively metaphysical in nature.
Whether they are correct is a function of the facts—a function of what joints
reality in fact has. One cannot do metametaphysics simply by examining
metaphysical language and reasoning. For given the sufficient condition, in
order to claim that a question is nonsubstantive, one must claim that it is not
cast purely in joint-carving terms, and such a claim cannot be supported
solely by reflecting on language and reasoning. For example, we saw in the
previous section that a sufficient condition for the extended-simples debate
being substantive is that a certain metaphysical thesis be true: the occupation
picture. Thus, in order to decide whether that debate is substantive, one must
directly engage in metaphysics.
It may be objected that this conclusion—that metametaphysical critiques
are metaphysical in nature—is simply an artifact of my way of conceiving of
metametaphysics. The objection is partly correct. An austere
metametaphysician who rejected realism about structure could perhaps
introduce some sensible notion of substantivity that doesn’t presuppose the
notion of carving at the joints. But even so, his metametaphysical critiques
could not be wholly ametaphysical. For imagine a metaphysician who shrugs
off a purely methodological or linguistic critique by saying that she is a
realist about carving at the joints, and that she believes her ideology to carve
at the joints. Surely the austere critic couldn’t simply concede these claims.
For, I hope, this book provides a model of metaphysics; a model of how, in a
world with objective structure, language could be attached to that structure
and metaphysicians could reasonably speculate on its nature. So even the
austere critic should agree that given the realism about structure, metaphysics
would make sense, both linguistically and methodologically. So the austere
critic must oppose the metaphysician’s claims. Granted, he could oppose
them by rejecting the realism about joint-carving, rather than by accepting
this realism while denying that the metaphysician’s ideology carves at the
joints. But this rejection of joint-carving is just more metaphysics.
Many are drawn to metaphysical deflationism because they want the
epistemic high ground. They want to rid the world of difficult-to-answer
substantive questions. But their very metaphysical deflationism costs them
the high ground. For since metametaphysical critiques are just more
metaphysics, they raise all the old epistemic questions. This is certainly true
if the critic embraces realism about structure and claims that the crucial
notions in the targeted debate do not carve at the joints—the epistemology of
such a claim is as hard as can be. But it’s even true if the critic is austere,
since rejecting realism about joint-carving raises the same old epistemic
issues.
This does not mean that metaphysical deflationism is never reasonable.
It’s often attractive to avoid some bits of metaphysics, some difficult
questions. But it does mean that no one should adopt metaphysical
deflationism with the goal of avoiding metaphysics, or difficult questions, in
their entirety.
Nor should one adopt deflationism about a metaphysical question merely
because the ideology needed to state that question differs from one’s own.
This should go without saying, but in fact, many critics have a blind spot here
too. They notice that the crucial term T in a metaphysical question can be
understood in multiple ways in terms of their own ideology, which they
regard as fundamental, never stopping to think that the participants in the
debate regard T itself as being fundamental. In the dispute over holes, certain
critics keep complaining that they don’t see the question. “Is it whether holes
are causally efficacious? Is it whether some of their properties fail to
supervene on the presence of matter and its arrangement?” No, the question is
simply: Do holes exist?!

1 Not to be confused with a deflationist about truth such as Paul Horwich (1990). Although my
metaphysical deflationist is not any specific person, Eli Hirsch’s work on metaontology—certain facets
of it anyway—may be kept in mind as a model.
2 Hirsch tries to target his deflationism by applying it only when the disputed metaphysical
proposition is regarded by one disputant as being a priori necessarily equivalent to one undisputed
proposition, and by another disputant as being a priori necessarily equivalent to another undisputed
proposition. But the disputants need make no such claims about a priority and necessity. Most
metaphysicians nowadays regard their views as theoretically supported conjectures, not as propositions
knowable by anyone who understands them. And some of Hirsch’s targets regard their subject matter as
contingent (Cameron, 2007). Moreover, a priori necessary equivalence seems relevant only if one
thinks of it as a kind of sameness of meaning, whereas surely it is no such thing. For a more
comprehensive (and very insightful) discussion of Hirsch, see Hawthorne (2009).
3 Note also that if the interpretations are alien, then there is no hope of carrying out the common-
sense and indeterminacy strategies.
4 Hard but not impossible; perhaps certain internal tensions in our conceptual scheme are best
resolved by preserving some aspects at the expense of others.
5 See Sider (2001a).
6 Thanks to Eric Funkhouser here.
7 Qualification: some stipulation might be allowed, if it is ultimately grounded in ordinary terms
(‘free’ could be stipulatively tied to ‘moral responsibility’ in its ordinary sense).
8 Hofweber’s (2009) critique of “esoteric” metaphysics—metaphysics done using distinctively
metaphysical rather than ordinary concepts—is related.
9 The two-stage process is oversimplified; a more realistic account would divide linguistic
processing into multiple stages (phonological, syntactic, logical form…), and contextual information
might intrude at multiple points. The point still stands: there can be substantive dispute about the
semantics/pragmatics border if the disputants can agree on a sufficiently detailed skeleton of a theory of
linguistic processing so that they can disagree over where, in that skeleton, truth-conditions figure in.
 10 See also Chalmers (2011).
11 See Siegel (2006).
12 The following can happen. (a) A philosophical debate began when there was no decent scientific
theory of a certain domain. (b) Subsequently a good scientific theory was developed. (c) No good rival
philosophical theory was developed; the concepts of the original philosophical debate were simply
badly chosen; they carved nature at its joints badly Nevertheless, (d) the philosophical debate continues
uncritically, in the old terms.
13 See Hudson (2006, chapter 4); Markosian (1998; 2004a; b); McDaniel (2003; 2007a; b); Parsons
(2003).
14 In this example we have a qualitatively heterogeneous extended simple, which goes beyond the
bare definition of an extended simple. See Sider (2007b, section 1).
15 See also Horgan and Potrč (2000; 2002).
16 As is customary, I define “possibly ϕ” as “not necessarily not-ϕ”.
17 See also section 9.6.2.
18 A fifth complaint might target the dispute only when extended simples are construed as having
“pseudo-parts”—further extended simples that are not genuine parts (i.e., parts in the joint-carving
sense) but are located where genuine parts would be located. This dispute might be claimed to be
nonsubstantive because the truth of ‘there are extended simples’ turns on whether ‘part’ means genuine
parthood or pseudo-parthood. The claim might be correct if ‘part’ is not a theoretical term; for then,
even though exactly one candidate meaning for ‘part’ carves at the joints, that candidate might be the
“wrong sort”—recall the third qualification of section 4.2. But if ‘part’ is a theoretical term then the
dispute remains substantive, for it is then cast in purely joint-carving terms of the right sort. Thanks to a
referee. A sixth complaint might be that the question just isn’t significant. (The old joke about angels
dancing on the head of a pin combines this and the first complaint.) And a seventh complaint might be
that the answer is obvious in some sort of Moorean way. Neither the sixth nor seventh complaint
challenges the substantivity (in my sense) of the question. Moreover, each seems wrong. Why wouldn’t
a question about the fundamental relationship between objects and space be significant? And why
should the nature of this relationship be obvious to “common sense”? (See also the critique of
Mooreanism in Sider (2011).)
19 Timothy Williamson (2007) gives a compelling account of the epistemology of philosophy (and
of the sociology of pessimism about philosophy) with which I mostly agree, except that for the
epistemology of metaphysics I would stress continuity with scientific explanation rather than
counterfactual reasoning.
6 Beyond the predicate
Structure is more general than its kin, Armstrong’s universals and Lewis’s
natural properties and relations, along two axes.
The first axis is ontological. Call a predicate “sparse” when it marks a
joint in nature. For Armstrong, a predicate is sparse when there exists a
corresponding universal; for Lewis, a predicate is sparse when there exists a
corresponding natural property or relation.1 Each assumes the existence of
abstracta. But the idea that the world has a distinguished structure—that
electrons go together and not together with cows, that it is better to think in
terms of electrons than in terms of electron-or-cows, and so on—does not
require this assumption. (Nominalists could surely embrace the idea.2) The
notion of structure is to be free of commitment to abstract entities.
The second—more important—axis concerns the scope of structure.
Armstrong and Lewis’s accounts are confined to properties and relations.
Linguistically speaking, their focus is on the predicate. Structure, on the other
hand, is not to be restricted to any particular grammatical category. Just as
Lewis and Armstrong ask which predicates get at the world’s structure, we
can also ask which function symbols, predicate modifiers, sentence operators,
variable binders, and so on, get at the world’s structure.
One might force these expressions into the Armstrong/Lewis mold, by
analyzing them using predicates and then assessing the predicates for
sparseness as before. Sometimes this is harmless; one can treat function
symbols as standing for relations, for example. But in other cases, it foists
unwanted commitments on us. For example, if sentence operators for
negation and disjunction are to be predicates, they must presumably be
predicates of propositions. Thus there must exist propositions; and further,
even friends of propositions might doubt that the most fundamental negative
and disjunctive facts are facts about propositions. (We will return to this last
point.) And in still other cases the predicate strategy is unavailable. Variable
binding expressions, for example, seem impossible to treat as predicates.
(One can take quantifiers to be predicates rather than variable binders, but
this just moves the bulge in the carpet: new variable binders must then be
introduced. Stalnaker (1977), for example, treats quantifiers as not binding
variables, but his variable-binding operation of complex predicate formation
cannot be thought of as predication. Similar remarks apply to Montague’s
treatment of quantification.3)
The two axes are connected. For Armstrong and Lewis, a predicate is
sparse when it stands for an appropriate sort of abstract entity. When we
move beyond the predicate, for instance to sentential connectives or
quantifiers, it becomes increasingly strained to think in terms of abstract
entities. So, insofar as one is wedded to abstract entities (first axis), it is hard
to move beyond the predicate (second axis).

6. 1 The reason to generalize

Why move beyond the predicate? Because the connections with the notion of
structure that we have been exploring are not confined to the predicate.
For example, I have argued that a dispute’s substantivity—whether it
concerns reality or just how we conceptualize reality—depends on how well
its crucial expressions carve at the joints. Sometimes those crucial
expressions are indeed predicates; disputes over causation, for example, are
usually cast in terms of predicates of facts or events. But other times the
crucial expressions are not predicates, and the question of substantivity can
be posed nevertheless. In certain disputes over ontology, time, modality, and
classical logic, for example, the crucial expressions are quantifiers (or
perhaps names), tense operators, modal operators, and negation and other
sentential connectives, respectively. (These disputes will be discussed in
subsequent chapters.) As before, one can attempt to force-fit the predicate
mold. But often this can be done only given the existence of disputed entities
(such as propositions); and with disputes involving quantification, predicate
reinterpretation seems unavailable.
Advocates of restrictive ontologies will be particularly loathe to formulate
their distinctive claims using predicates. Rejectors of both events and facts,
for example, cannot construe causal claims as involving a predicate of events
or facts; they might instead use a sentence operator: “that ϕ causes it to be the
case that ψ”. Rejectors of propositions cannot regard modal and temporal
claims as involving predicates of propositions; they might instead use
sentential modal and tense operators. Rejectors of sets cannot define
generalized quantifiers such as ‘finitely many’ and ‘most’ in terms of set
theory; they might instead take them as primitive.
And even defenders of more permissive ontologies often require
nonpredicates in subtle ways. For example, someone who regarded modal
and tense operators as really being predicates of semantic entities might still
require primitive nonpredicates such as ‘the proposition that ϕ’ and ‘the
property of F-ing’ to characterize the space of semantic entities. (These are
not predicates; their grammar is to turn sentences and predicates,
respectively, into terms.)
To assess the substantivity of disputes whose crucial terms are
nonpredicates, then, we need to go beyond the predicate. Such disputes also
introduce a further need for doing so, within epistemology. As we saw in
section 2.3, ideological simplicity is part of what makes a theory
choiceworthy. Further, in choosing a fundamental theory we often face the
choice of whether to adopt increased ideology or ontology. But theories
which forgo ontology at the cost of increased ideology often do so precisely
by introducing distinctive non-predicate ideology. The function of predicates,
after all, is to combine with singular terms—which denote entities—to make
statements; renouncing certain entities can thus call for distinctive modes of
expression other than predicates. So to evaluate proposed trades of ontology
for ideology—which we must do in order to choose which fundamental
theory to believe—we need to speak of joint-carving for expressions other
than predicates.
We have seen two ways in which the proposed “applications” for the
notion of structure are not confined to the predicate. There are many more.
(To mention just two: the doctrine of reference magnetism should be
extended to nonpredicates; and in the Lewisian theory of laws, the logical
expressions in the lawmaking language must be required to be joint-carving;
otherwise cheap simplicity can be obtained with rigged logical expressions
just as through rigged predicates.) The domains in which we need to speak of
structure are not confined to the predicate. We need a broader conception
than that of Armstrong and Lewis.
But this broad conception of structure often meets resistance. The
following sections examine various sources of this resistance, and argue in
each case that the resistance is misguided.
6.2 Inapplicability of the similarity test
One common source of resistance is the apparent failure of a similarity
conception of structure to apply to expressions other than predicates.
Armstrong and Lewis tie their accounts to similarity. Lewis, for example,
says of his perfectly natural properties:
Sharing of them makes for qualitative similarity, they carve at the joints, they are intrinsic, they are
highly specific, the sets of their instances are ipso facto not highly miscellaneous … (1986b, p. 60)
For many, this “similarity criterion” is the only handle they have on Lewisian
naturalness. But it seems inapplicable to quantifiers and sentential
connectives, for example. Quantifiers and sentential connectives aren’t
“shared”, nor do they have instances, so we can’t assess whether their sharing
makes for qualitative similarity or whether the sets of their instances are
highly miscellaneous. As for “carving at the joints”, that metaphor suggests a
similarity-theoretic reading: i) do carve where there’s a joint, i.e., assign
different natural properties to dissimilar objects; and ii) don’t carve where
there’s no joint, i.e., assign similar things the same natural properties. And
neither i) nor ii) seems to apply to quantifiers or sentential connectives. So
it’s easy to see how someone whose only entrée to naturalness is similarity
would be baffled by the more general notion of structure.
Worse, insofar as the similarity criterion can be applied in these cases, it
seems to deliver the wrong results. Does the existential quantifier carve at the
joints? I will argue in chapter 9 that it does. But every two things, no matter
how dissimilar, share the feature of existing; so doesn’t that mean that the
existential quantifier in some sense fails the similarity criterion?
The worry is not that since natural properties are defined as similarity-
makers, the notion of naturalness cannot be generalized to quantifiers. Lewis
doesn’t define ‘natural’ at all.4 The worry is rather the following. As an
undefined theoretical term, ‘natural’ is understood through its theoretical role
(section 2.1). The theoretical role consists of principles specifying how
naturalness relates to certain other notions, such as similarity. My term
‘structure’ has its theoretical role fixed by many of these same principles, and
more besides. If the theoretical role of ‘structure’ is exhausted by its
connection to similarity, and if this connection is restricted to predicates, then
we have no understanding of how ‘structure’ could apply to expressions in
other grammatical categories.
But the theoretical role that Lewis offers for naturalness is not exhausted
by the similarity criterion. Still less is the theoretical role for the more general
notion of structure exhausted by the similarity criterion. The connection to
similarity is just one of a network of theoretical connections that give the
notion its life.5 These are the connections we have been exploring, the
connections to laws, explanation, metasemantics, epistemology, physical
geometry, substantivity, objectivity, and epistemic value. None of these
further aspects of structure’s theoretical role relies on similarity. So even if
we set aside similarity, structure has a rich enough remaining theoretical role
to be intelligibly applied beyond the predicate.
But in fact, we needn’t set aside similarity. The connection to similarity
can be maintained, even for expressions other than predicates, if that
connection is properly understood.
First we must expose a confusion. A putative reason for doubting that the
existential quantifier carves at the joints was the observation that every two
objects, no matter how dissimilar, share the feature of existence. But this
observation is irrelevant. It concerns the predicate ‘exists’, whereas our
question concerns the existential quantifier.
Next we must change our focus, from object-similarity to fact-similarity
The connection between similarity and structure, in the case of the existential
quantifier, should be understood thus: if existential quantification carves at
the joints, then whenever two facts are existential facts, that is a genuine
similarity between them. The fact that there is a donkey and the fact that
there is an electron are genuinely similar in that each is an existential fact.
Similarly, to evaluate whether ‘or’ ensures similarity, we must ask whether
any two disjunctive facts are ipso facto similar; to evaluate whether modal
operators insure similarity, we must ask whether all facts of the form
possibly, ϕ are ipso facto similar; and so on.
Language is no more a guide to fact-similarity than it is to similarity in
general. Just as the applicability of ‘grue’ to each of a pair of particulars does
not guarantee the similarity of those particulars, the recurrence of ‘there is’ in
the sentences used to state a pair of facts does not on its own guarantee the
similarity of those facts. If all existential facts are indeed similar, this is
because of something about quantification: being existential is a genuine, not
merely nominal, feature of facts. (We can invent words with the same
grammar as quantifiers for merely nominal features of facts. “There
schmexists an F”, recall, means that the property of being an F is expressed
by some predicate in some sentence of this book. It is a fact that there
schmexists a donkey, and also a fact that there schmexists a brilliant
Republican president born in New Haven, but these facts share nothing in
common.)
Why this move to fact-similarity? The idea of a genuine similarity is that
of a real commonality. Here we have a negatively charged thing; there we
have another negatively charged thing; has something in nature recurred? Is
there a real commonality between the negatively charged things? We think
so. The recurrence was within objects in this case, but that is inessential to
the legitimacy of the question. There exists a donkey; there exists an electron
—has anything recurred? Is there something in common between there
existing a donkey and there existing an electron? If I am right that
quantification carves at the joints, then the answer is yes; these facts share a
real commonality. But if I am wrong (as defenders of “quantifier variance”
think—see chapter 9) then the answer is no; quantificational facts do not
particularly “go together”.
It has been convenient to speak of facts in order to introduce this broad
conception of similarity, but the similarity judgments in question don’t really
require reifying facts.6 The idea of a genuine commonality, of recurrence,
requires no things in which the recurrence occurs. Such nonontic similarity
judgments can be regimented using a sentence operator: “When ϕ, it’s like
when ψ”7 Thus we can say, for example, that when there exists a donkey, it’s
like when there exists an electron. I don’t claim that we already speak this
way (though it’s not clear to me that we don’t; consider “When it sleets, it’s
like when it snows”). What I do claim is that there are coherent judgments
that can be thus expressed, and that they’re similar enough to ordinary ontic
similarity judgments to be thought of as being in the same species.

6.3 No entities
We have already touched on a second source of resistance to going beyond
the predicate. Armstrong/Lewis sparseness—the most familiar game in town
—is “entity-based”. For Lewis, in order to evaluate whether a predicate
carves at the joints we must look at a certain entity—the set of the predicate’s
actual and possible instances—and ask whether the entity is natural. And for
Armstrong, we must ask whether a certain entity exists—a universal
corresponding to that predicate. To extend their strategy to expressions like
quantifiers or sentential operators, we would need to identify corresponding
entities. But there seem to be no such entities.
One might reply that the entities do exist after all. In the usual model
theory of first-order languages, quantifiers and sentential connectives are
taken to be syncategorematic: the definition of truth in a model fixes their
contributions to truth-conditions without appealing to entities associated with
them. But one can instead associate entities with these expressions as their
semantic values. One can treat quantifiers as denoting second-order
properties of (or relations between) properties (Montague, 1973), and one can
treat sentential connectives as denoting truth functions or relations between
propositions.8 And then joint-carving for quantifiers and sentential
connectives can be treated in Armstrong and Lewis’s way, as turning on the
existence or nature of denoted entities.
But this does not get to the heart of the issue—at least, not if we are
looking for an account of what structure is (please read with the appropriate
cadence), as opposed to merely seeking a systematic way to talk about
structure. For the treatment of quantifiers as expressing second-order
properties or relations, however appropriate in linguistic theory, does not ring
true at a metaphysical level. Let ∃2 be the second-order property of having
at least one instance. That there exist cows does not seem to be a second-
order fact. It surely concerns the concrete world directly, rather than through
abstract intermediary entities like 32. But if quantificational facts are not
about ∃2, then surely facts about quantificational structure are not about ∃2
either.
Similarly, it may be appropriate in linguistics to treat ‘and’ as a relation
between propositions; but metaphysically speaking, the fact that I am human
and I am typing surely concerns neither propositions nor relations between
them. I’d even like to say the same about predicates. Even if linguistics is
right to associate ‘is human’ with a semantic value, the ultimate metaphysics
of my being human has nothing to do with this semantic value.9 And if
conjunctive and predicational facts don’t involve the semantic values of ‘and’
and predicates, then surely the corresponding facts about structure don’t
involve these semantic values either.
The view that structure facts do not concern semantic values is certainly
intuitively compelling; but it’s best not to rely solely on intuitive
compulsions. Fortunately, there is a systematic consideration as well:
semantics is, like other special sciences, not fundamental. Our most
fundamental level of theorizing should no more recognize distinctively
semantic entities and ideology than it should recognize distinctively
economic or psychological entities and ideology.10 This is not to say that the
statements of semantics are untrue, only that they are not fundamental (see
section 7.8). But if semantics isn’t fundamental, the facts about carving at the
joints can’t fundamentally involve semantic entities.11
It’s a mistake, then, to think of structure as concerning semantic values.
For similar reasons, it would be a mistake to think of structure as concerning
linguistic items. The fact (if it is a fact) that ‘is negatively charged’ carves at
the joints isn’t in the first instance a fact about the predicate ‘is negatively
charged’. The fact is simply a fact about the concrete, nonlinguistic world—
about its “charge aspect”, so to speak. Likewise, the fact (if it is a fact) that
the first-order quantifiers carve at the joints isn’t a fact about the linguistic
items ‘there is’ and ‘for all’. It’s a fact about the world—specifically, its
quantificational aspect. (Not to reify aspects.)
The no-entities worry can seem inescapable given a certain regimentation
of structure-talk, which takes the core locution to be a predicate. In Lewis, for
example, the core locution is the predicate ‘is natural’. But predicates must be
ascribed to entities; for Lewis, the entities were the semantic values of
predicates; this then leads to taking the facts about structure to involve
semantic values.12 (And it’s no better to take the core locution to be a
predicate of linguistic items.) This suggests a way to answer the no-entities
worry: introduce a regimentation in which the core locution is not a predicate,
so that we can talk about structure without bringing in entities of any sort.13
We want a locution, call it “ ”, with which to make statements about
structure. What should its grammar be? Here we face an obstacle. Since we
are going beyond the predicate, must somehow combine with expressions
α of arbitrary grammatical category—with quantifiers, sentential operators,
and so on, as well as predicates—to form sentences. But what kind of
expression has this sort of grammar? (The attraction of the predicate
regimentation is that it avoids this problem: we first convert each such α into
a singular term, tα, to which a predicate for structure may be applied. But
what would tα be? The only possibility seems to be a quotation name of α
itself, or a name of α’s semantic value; but then we’d be back to treating the
facts about structure as involving linguistic items or semantic values.)
We might overcome this obstacle by taking to have a very flexible
grammar, so that it attaches directly to α, regardless of α’s grammatical
category. I don’t know of any natural language expressions with this sort of
grammar, but I don’t see that as a problem. Some philosophers think that we
can understand only what can be defined using the pre-existing resources of
natural language; but this stultifying doctrine is inadequate to the evident fact
of linguistic innovation within science, as well as to the initial emergence of
natural language itself. If the inferential role of a novel expression has been
made tolerably clear, and if the world contains structure corresponding to the
new expression, then surely the introduction of the novel expression has been
successful.
So on this way of overcoming the obstacle, the core locution for talking
about structure is an “operator” , which can attach to an expression of any
grammatical category α to form a grammatical sentence (α). Thus, we can
say “ (is negatively charged)”, “ (ϕ)” (or “ (there is)”), “ (∧)” (or “
(and)”), and so on. Since nothing in English really matches this
regimentation, I’ll tend to revert informally to a predicate of linguistic items
or abstract entities; I’ll speak of ‘is negatively charged’, ‘and’, and ‘there
exists’—or negative charge, conjunction, and existential quantification—as
“carving at the joints” or “being fundamental”. But the facts of structure are
more faithfully represented using .14
To say (and) is not to say something about an alleged object
Conjunction. It is not to say anything about any thing at all. It is nevertheless
to say something true, something objective, something about reality.
Nowhere is it written in stone that all facts must be entity-involving. In
Graham Nerlich’s phrase, “realism need not be ontic”.15 To be sure, the
entity-based ideology of predicate logic is simple, beautiful, and well-
behaved, and it’s best to stick to it whenever possible. But the realist about
structure, it would seem, cannot live by predicate logic alone.
There are hard questions about thus taken. I said that it can attach to
expressions “of any grammatical category”. What, exactly, does that mean?
must at least be able to attach to all primitive expressions of the language
in question; but what about complex expressions? (As we will see in section
7.13, we need to be able to query complex expressions for joint-carving.)
We might say that can attach to any “grammatical unit”—intuitively,
any string that is either a primitive symbol or a complex symbol that is
generated at some point by the language’s recursive formation rules. This has
certain limitations, depending on the grammar of the fundamental language
we’re using to talk about structure. Suppose we want to ask whether a certain
conjunctive predicate, the conjunction of predicates F and G, carves at the
joints. If our language has a predicate functor c for predicate conjunction then
this is straightforward. Our formation rules will include a clause for complex-
predicate formation; one of the expressions formed via that clause will be the
conjunctive predicate F c G; so we can form the sentence (F c G). But
suppose (as I suspect is more likely) that our fundamental language is
grammatically simpler. Suppose, for example, that its grammar is that of
predicate logic (without function symbols), with the addition of the operator
In that case, there simply are no complex predicates, since the only rules
of formation are for the grammatical category of sentence. So the only
complex strings we can query for structure are sentences (including those
with free variables). But then, we can achieve our goal of evaluating the
conjunction of F and G for structure only indirectly, by forming the open
sentence (Fx∧Gx).16 And the querying becomes even more indirect for,
say, complex operators. Suppose, for instance, that the only primitive
sentential connectives in the language are ∼ and ∨, and we want to query for
structure a complex expression with the truth table of the material
conditional. We can query the various sentences of the form ∼ϕ ∨ ψ (for
example, ∼Fx ∨ Gy, ∼Ha ∨ ∀x∀y Rxy, and so on); but each of these
sentences also queries certain other expressions; namely, the expressions
occurring inside ϕ and ψ.
We might just live with the fact that we can’t query complexes directly.
We could say that the question of whether “the material conditional carves at
the joints” is not a matter of the truth of any one sentence in our fundamental
language; rather, it emerges from the totality of sentences of the form (∼ϕ
∨ ψ).17 Alternatively, we might pursue a different method for overcoming
the obstacle to constructing a grammar for .18 According to this method,
no longer has a flexible grammar; now it is a one-place sentence operator.
However, we include in our fundamental language a collection of dummy
variables. There are dummy variables of every grammatical category:
individual dummy variables, sentential dummy variables, predicate dummy
variables, sentence-operator dummy variables, and so on. Dummy variables
are not bindable. Their purpose is to combine with other expressions to form
complete sentences, so that the sentence operator may then be applied.
The expressions other than dummy variables in such sentences are those that
are queried for carving at the joints. So we no longer query sub-sentential
primitive expressions by directly attaching to them; instead, we attach
those primitive expressions to appropriate dummy variables to obtain a
sentence, and then attach to that sentence. For example, we query the
predicate G, the name a, the quantifier ∃, and the operator □, with the
following sentences, respectively (dummy variables are in sans serif):

Complex expressions are now straightforward to query. We can query the

material conditional, for example, using the sentence (∼P ∨ Q).
I have argued that we should not think of judgments about structure as
concerning entities. Judgments about structure concern ideology, not
ontology. Let us close with a discussion of an opposing, ontic approach.19
This opposing approach is modeled on Armstrong, rather than Lewis.
Lewis’s approach to regimenting talk of structure was to posit an abundant
group of entities (for him, sets of ’tuples of possibilia) and regard only some
of them as being structural (natural). The Armstrongian approach, by
contrast, does away with the abundant entities, and posits only a sparse group
of entities. Given a sparse entity, there is no further question of whether it
carves at the joints; sparse entities automatically carve at the joints, so to
speak. But given a linguistic entity there is a further question of whether it
stands for a sparse entity. The sparse entities, for Armstrong, were universals;
thus on his view, a predicate carves at the joints iff it stands for some
universal.
The sparse entities of this approach are not to be thought of as semantic
values. (After all, most meaningful expressions do not stand for sparse
entities.) Thus the approach is immune to the objection that semantics is
nonfundamental.
Call this approach ontologism, since it insists that fundamental
metaphysical commitments be ontic. Distinguish two forms: methodological
and metaphysical. The defender of methodological ontologism says that his
opponents are making some sort of methodological or conceptual mistake.
It’s somehow conceptually confused to think of a fundamental metaphysics
as being given by anything other than a list of entities. The defender of
metaphysical ontologism, on the other hand, puts forward her position as a
substantive hypothesis about the nature of fundamental reality.
Methodological ontologism seems hopeless. Earlier in this section I put
forward a conception, using the operator , of what non-ontic claims of
fundamental metaphysics might look like. It’s hard to see how this position is
conceptually or methodologically confused. Further, it’s tempting to view the
defender of methodological ontologism as arbitrarily privileging his own
ideology. His ideology includes the quantifiers and a special predicate that
singles out the sparse entities; my ideology includes and various other
expressions; what makes his ideology the sole intelligible vehicle for giving
fundamental descriptions of reality?
Quine’s terminology—to return to my rant—has perhaps contributed to
methodological ontologism appearing more plausible than it really is.
‘Ideology’ suggests a purely arbitrary, conceptual matter; ideology and
ontology are supposed to be exhaustive; thus, the only nonconceptual
question is that of one’s ontology. Quine’s terminology is so ingrained that
this reasoning can seem built into the very concept of metaphysics.20 Simply
appreciating the possibility of an alternative, of ontology-free but
nevertheless worldly metaphysics, should break this spell.
Metaphysical ontologism is a much more likely proposition. The attitude
here is that the alternative is inferior, not unthinkable; the best metaphysics of
fundamentality is entity-based.
Ontologism could, in principle, sustain much of the project of this book.
For in many cases, talk of joint-carving ideology can be replaced with talk of
sparse entities. In a discussion of whether causal disputes are substantive, for
example, instead of asking whether (causes), the defender of ontologism
could ask whether there exists a causal universal. But problems begin when
the defender of metaphysical ontologism tries to go beyond the predicate.
Whenever she wants to speak of joint-carving she faces a choice: either posit
a corresponding sparse entity, or else do not speak of joint-carving at all.
For example, if she wanted to say that modal operators carve at the joints,
she would have to posit an appropriate entity and claim that the modal facts
ultimately boil down to the facts about this entity. Candidates for the entity
might include i) a modal universal that is instantiated by other universals;21
ii) a modal universal that is instantiated by propositions (which would
themselves need to be reified—a very unArmstrongian move); or iii) a modal
entity of a different sort, call it a “modal monad”, that plays a role in
instantiation: the necessary possession of a universal U amounts to the
instantiation of U by a particular with respect to the modal monad, whereas
the merely contingent possession of U is the instantiation simpliciter of U by
the particular. If she is unwilling to say any of these things—to underwrite
the claim of modal structure by positing a sparse modal entity—then she
cannot speak of modal structure at all.
I myself do not believe in modal structure (chapter 12), so I don’t regard
this limitation as unwelcome. But consider, next, the case of quantification.
As we saw, it is artificial to take a quantifier, in one’s ultimate metaphysics,
as standing for an entity. So the defender of ontologism must choose between
an artificial metaphysics of quantification and forgoing talk of joint-carving
for quantifiers. And the latter would be a real limitation: as we will see in
chapter 9, the thesis that quantifiers carve at the joints is the best way to
defend the substantivity of ontological questions. This limitation is the chief
problem with metaphysical ontologism.
The previous paragraph’s line of thought is especially dialectically
effective because the defender of ontologism has a particular reason to regard
quantificational questions as being substantive. After all, questions about the
nature of fundamentality, for her, turn on whether there are appropriate
sparse entities. She, of all people, cannot acquiesce to the Carnapian thought
that talking as if sparse entities do not exist is just as good as talking as if
they do. Thus metaphysical ontologism is “unreflexive”: given its strictures
on what can be evaluated for fundamentality, its own apparatus cannot be
ratified as fundamental.
Armstrong’s own theory is similarly unreflexive. With good reason,
Armstrong refrains from positing a universal of instantiation (1978a, chapter
11, section 1). Thus, although ‘instantiates’ is an ineliminable part of his
ideology, he cannot recognize ‘instantiates’ as fundamental, in the way that
he recognizes fundamental physical predicates as fundamental. But claims
specifying which particulars instantiate which universals are clearly part of
Armstrong’s fundamental theory of the world. Merely to list the universals
and particulars, without specifying which particulars instantiate which
universals, would be a woefully partial description of what, according to him,
the world is fundamentally like. There is, therefore, overwhelming pressure
on Armstrong to recognize, somehow, that talk of instantiation is
fundamental; but he is barred from doing so by his ontologism.22
The defender of ontologism might, in Wittgensteinian fashion, reply that
one cannot say the whole truth about fundamentality. The whole truth can
only be shown, by quantifying over entities, saying that particulars instantiate
universals, and so on. If this position is uncomfortable, that is a reason to
reject ontologism, and to adopt a broader conception of metaphysical
commitment.

6.4 Unclear epistemology?

A third reason to fear going beyond the predicate is epistemic: how could we
ever tell when attributions of structure to nonpredicates are justified? But in
fact, the generalized conception of structure raises absolutely nothing new,
epistemically. Questions about how much nonpredicate structure the world
contains are substantive metaphysical questions, just like the most
substantive questions of first-order metaphysics, and can be addressed in the
same way.
Section 2.3 presented a broadly Quinean approach to the epistemology of
metaphysics, and to the epistemology of joint-carving in particular. It is
reasonable to regard the ideology of our best theory—“best” by the usual
criteria for theory choice, such as simplicity—as carving at the joints. This
approach is not bound to the predicate. Successful theories justify belief in all
of their ideological posits. I will, for example, argue in chapter 9 that since
our most successful theories employ quantification, we have reason to believe
that quantifiers carve at the joints—that quantificational structure is
fundamental. This argument is exactly parallel to the argument for
fundamental spatiotemporal structure: quantification is no less part of the
ideology of fundamental physics than are spatiotemporal notions.
When a conceptual decision has become wholly familiar, it is easy to
forget that it is nevertheless a decision. Such are the overwhelmingly
successful conceptual decisions of modern logic. Thinking in terms of and,
or, not, all, some, and identical has led to great strides in the foundations of
logic and mathematics, and so, less directly, in the foundations of all other
disciplines as well. The success of these conceptual choices justifies belief in
the existence of corresponding structures in the world. Once “ideology” is
purged of its psychological connotations, there is no barrier to recognizing a
theory’s logical ideology as a coequal part of that theory’s portrayal of the
world, and thus as being as good a candidate for carving at the joints as the
theory’s nonlogical vocabulary.
6.5 Logical conventionalism
A fourth source of resistance to going beyond the predicate, and in particular
to speaking of joint-carving logical notions, is vaguer, more primordial, and
(I think) widespread. It is the thought that it is appropriate to evaluate
expressions for carving at the joints only when they are “contentful”.
Predicates are paradigmatically contentful. But logical expressions, on the
other hand, are purely “formal”, so the thought goes. They do not describe
features of the world, but rather are mere conventional devices. Since logical
expressions are not “worldly”, it is inappropriate to speak of the world as
containing structure corresponding to those expressions.
This is picture thinking. But behind the picture, I suspect, there lies an
identifiable—and mistaken—philosophical doctrine: the doctrine of logical
conventionalism.
Actually, what I really think is widespread is not so much an acceptance
of logical conventionalism as a failure to fully repudiate it. The status in
contemporary philosophy of logical conventionalism and the related doctrine
of “truth by convention” is curious. On the one hand, few people self-identify
as logical conventionalists. If pressed on why not, I suppose most would
gesture at Quine’s famous critique in “Truth by Convention”. But on the
other hand, the picture thinking described above really does have staying
power, which would be explained by latent logical conventionalism.
Moreover, the language of truth by convention persists: one still hears the
phrase “true purely by virtue of meaning”, logical truths are still described as
being “trivial” or “empty” (and are thus thought of as being epistemically
unproblematic), and so on.23
Against logical conventionalism, I uphold Russell’s (1919, p. 169)
diametrically opposed position: “logic is concerned with the real world just as
truly as zoology, though with its more abstract and general features”.
Evaluating logical expressions for joint-carving is therefore not different in
kind from evaluating any other expressions for joint-carving.
All I have to offer in support of Russellian realism about logic is a critique
of conventionalism; discussion of intermediate positions remains a lacuna.
Logical conventionalism originated in the “linguistic theory of the a
priori”, popularized by A. J. Ayer in Language, Truth and Logic (1936):
Like Hume, I divide all genuine propositions into two classes: those which, in his terminology, concern
“relations of ideas,” and those which concern “matters of fact.” The former class comprises the a priori
propositions of logic and pure mathematics, and these I allow to be necessary and certain only because
they are analytic. That is, I maintain that the reason why these propositions cannot be confuted in
experience is that they do not make any assertion about the empirical world, but simply record our
determination to use symbols in a certain fashion. (p. 31)

A proposition (sentence) is analytic, Ayer goes on to say, “when its validity

depends solely on the definitions of the symbols it contains …”. Analytic
propositions can be known a priori because they are “devoid of factual
content” (p. 78), because they merely “record our determination to use words
in a certain fashion”. Something like this view was once widely held, by
logical positivists, Wittgensteinians, and ordinary language philosophers.24
The core of the view is that an analytic truth, for instance the truth that all
horses are horses, is true purely by virtue of linguistic conventions. By
adopting certain rules governing the use of logical words like ‘all’, language
users somehow make ‘all horses are horses’ true.
Quine famously objected to the doctrine of truth by convention as follows.
(B) is allegedly true by convention:
(B) An object is a bachelor iff it is an unmarried man.
Pretend that, as a matter of convention, ‘bachelor’ means the same as
‘unmarried man’. Thus, (B) means the same as the following logical truth:
(A) An object is an unmarried man iff it is an unmarried man.
The introduction of the convention governing ‘bachelor’ therefore makes (B)
have the same truth-value as (A); but this on its own, Quine pointed out, does
not make (B) true. The truth of (B) requires the “prior” truth of (A). As Quine
says, “… definitions are available only for transforming truths, not for
founding them” (1936, p. 81).
(B) would be rendered true by convention if the logical truth (A) itself
were in some sense true by convention. But as Quine went on to argue,
logical truths do not in any interesting sense owe their truth to conventions.
In particular, he considers the thought that we can legislate a logical truth T
by proclaiming “Let T be true.” He points out a problem (one that arises even
before the pressing question of how such proclamations are supposed to do
the trick). A would-be legislator of logical truth cannot individually legislate
each logical truth, for there are infinitely many of them. He might think to
make general legislations to the effect that every sentence of a certain form is
to be true, such as:
(I) Let every instance of the following schema be true: If ϕ then ϕ .
But any such legislation results in truths of a single fixed logical form,
whereas logical truths can take infinitely many forms. So, he might think, a
second type of general legislation is required, which specifies that if certain
statements are true, then others are to be true as well. Here is an example:
(II) If a statement ϕ and a statement If ϕ then ψ are true, then let ψ be true
as well.

The conventionalist’s hope, as Quine imagines, is to make true all the truths
of propositional logic using legislations of types (I) and (II). For there exist
complete axiomatizations of propositional logic with finitely many axiom
schemas in which the only rule of inference is modus ponens; each axiom
schema could be legislated in style (I), and modus ponens could be legislated
into effect by proclaiming (II). One could then go on to legislate the truths of
(first-order) predicate logic in a similar fashion.
According to Quine, the problem for conventionalism thus understood is
that logic is needed to apply the conventions, and cannot therefore be
grounded in the conventions. Suppose that statements ϕ and if ϕ then ψ
have been legislated to be true by legislations of sort (I). (II) now says that if
these sentences are true, then ψ is to be true as well. To derive from this the
result that ψ is indeed true, we must perform modus ponens—we must use
logic. But logic is exactly what the legislations are supposed to ground.
For various reasons, it seems to me that Quine’s objection—that logic will
be needed to legislate the infinity of logical truths—does not get to the heart
of what is metaphysically problematic about conventionalism.25 Imagine a
finitary conventionalist, who tries to introduce conventional truth in a
language whose set of well-formed formulas is finite. Or imagine a
conventionalist with an infinitary mind, who can legislate each of the
infinitely many logical truths individually. Logic would surely not be true by
convention even in these cases, but in neither case does Quine’s objection
apply.
Moreover, the conventionalist might reply to Quine that legislations of
form (II) are conditional legislations rather than legislations of conditionals.
(Compare the distinction between conditional probability and the probability
of a conditional, or between conditional obligation and an obligation to see to
the truth of a conditional.) Quine’s objection is actually pretty elusive, but
one way of taking it is as follows:
After legislations of type (I) are made, for certain sentences ϕ and ψ, both sentence ϕ and the sentence
 if ϕ then ψ are true by convention; and after legislation (II) is made, the following conditional
sentence is also true by convention: If ϕ and if ϕ then ψ are both true, then ψ is also true . But we
cannot pass from the fact that these three sentences are true by convention to the conclusion that ψ is
true by convention unless we make the further assumption that the truths by convention are closed
under modus ponens (and also under conjunction-introduction, truth-introduction, and truth-
elimination). Since that further assumption is precisely what (II) was supposed to accomplish, (II) is
ineffective.
This objection, notice, assumes that the effect of legislation (II) is that a
certain conditional sentence (namely If ϕ and if ϕ then ψ are both true,
then ψ is also true ) is true by convention. But the conventionalist might
reply that its effect is instead that it be the case that if ϕ and If ϕ then ψ are
both true by convention, then ψ is also true by convention. The
conventionalist might reply, that is, that his claim all along was that (II)
results directly in the set of truths by convention being closed under modus
ponens, and not that it results in a conditional sentence corresponding to
modus ponens being true by convention. (II) is not an unconditional
legislation that a certain sentence (a conditional sentence) be true by
convention. It is rather an irreducibly conditional legislation, which results in
its being the case that if certain sentences are true by convention, then so is a
certain further sentence. The latter sentence is legislated conditionally on the
former sentences being legislated. Quine in effect grants the conventionalist
for the sake of argument that the words pronounced in (I), “Let any statement
of the following form be true …” have their desired effect. But the words are
not magic: something about the pattern of beliefs and dispositions in the
linguistic community that results from the pronouncement of those words is
what allegedly does the trick: certain sentences individually become true by
convention. If Quine is willing to grant that the words in (I) have this effect,
why is he not also willing to grant that different words, the words in (II),
which result in a different pattern of beliefs and dispositions, have a different
effect, that of a conditional legislation? Quine’s argument does not go far
enough. An adequate critique must challenge the very idea of something’s
being “true by convention”. Even an infinite mind, or a conventionalist with
only finite aspirations, or a conditional legislator, could not make the logical
truths, or any other sentence for that matter, true by convention (unless the
sentence is about conventions). The components of this critique are not new,
but are nevertheless worth repeating.
Part of the critique consists in pointing out that it is no easy task for the
conventionalist to specify an appropriate sense in which logical truths are
“true by convention”. There is a mundane sense in which all true sentences
are partly true because of convention, since all sentences exhibit symbol-
conventionality (section 4.3). Even a synthetic sentence like ‘Snow is white’
is true partly because of its meaning; it would not have been true if it had
meant that grass is green. Of course, in order for ‘Snow is white’ to be true,
the world must also cooperate: the world must really be as the sentence says.
Snow must really be white. So, it might be thought, the conventionalist could
claim that the requirement of worldly cooperation is not present for logical
truths. But on the face of it, this is wrong. By convention we make it the case
that the sentence ‘If it is raining then it is raining’ means that if it is raining
then it is raining; but in order for the sentence to be true, the world must also
cooperate; the world must really be as the sentence says. It must really be that
if it is raining then it is raining. It is easy to overlook the requirement of
cooperation in this case because it is so obvious that if it is raining then it is
raining. But no sense has yet been given to the idea that the requirement is
not present.
A conventionalist might reply that ‘If it is raining then it is raining’
“automatically” becomes true upon being endowed with its meaning; that is
the sense in which there is no further requirement that the world cooperate.
But what does ‘automatically’ mean here? It could be understood in terms of
necessity: it is necessarily true that if ‘if’ and ‘then’ mean what they do then
‘If it is raining then it is raining’ is true. Thus understood the claim is correct,
but it does not secure a truth-making role for the convention. The mere fact
that it is necessarily true that it is raining if it is raining ensures that ‘If it is
raining then it is raining’ is “automatically”—in the current sense of
‘automatically’—true once it has been given its meaning. Conventionalism
thus understood says little more than that logical and other analytic truths are
necessary; nothing is left of the intuitive idea of their truth being grounded in
conventions.
We are still in search of an appropriate sense in which logic is true by
convention. Ayer’s claim that analytic truths “simply record our
determination to use symbols in a certain fashion” is unhelpful. It suggests
that analytic truths make statements about linguistic conventions. But this is a
nonstarter; statements about linguistic conventions are contingent, whereas
the statements made by typical analytic sentences are necessary (Broad
(1936, p. 107), Lewy (1976, p. 9)). Ayer’s claim that analytic truths “say
nothing about the empirical world” is similarly unhelpful: it is hard to attach
any sense to it that advances his cause. ‘If it is raining then it is raining’
seems to say something about the empirical world: that the empirical world
contains rain if it contains rain. Of course, the thing it says is a logical truth.
We might define “about the empirical world” to exclude logical truths, but
what would be the point?—the claim that logical truths “say nothing about
the empirical world” could then play no role in explaining the epistemology
or metaphysics of logical truth.
To further reinforce the difficulty of finding an appropriate sense of ‘true
by convention’, consider that the phrase is intended to indicate an intimate
sort of dependence of truth on convention. But what sort, exactly? The
conventionalist will surely deny counterfactual or temporal dependence, at
least of the sort that would imply absurd statements like the following:
Before we introduced our conventions, not all green things were
green.
If we had introduced no conventions, not all green things would have
been green.
Of course, metalinguistic counterfactual and temporal statements such as the
following are unproblematic:
Before we introduced our conventions, the sentence ‘all green things
are green’ was not used to express a truth.
If we had introduced no conventions, the sentence ‘all green things are
green’ would not have been used to express a truth.
But all truths depend on conventions in this metalinguistic way; before we
introduced our conventions, the sentence ‘Snow is white’ was not used to
express a truth. It remains unclear just what sort of dependence of truth upon
conventions is supposed to be distinctive of conventionalism.
Here are two further failed attempts to understand what the defender of
truth by convention has in mind. Return to the would-be truth-legislator, who
says “Let every sentence of the form ‘If ϕ then ϕ’ be true.” What is this
performance intended to accomplish? On the one hand, the legislator could
be resolving to use the word ‘true’ in a new way; he could be listing the
sentences to which this new term ‘true’ applies. But this obviously isn’t what
the conventionalist wants. On the other hand, the legislator could be
constraining the intended meaning for conditional constructions. He could be
placing a necessary condition on the function from sentences to the
propositions that they mean: this function must assign a true proposition to
each sentence of the form ‘If ϕ then ϕ’. Any function that violates this
constraint, the legislator is saying, is not the means-in-English function. This,
too, is not what the conventionalist wants, for it assumes an antecedent notion
of propositional truth that has not been shown to depend in any way on
convention.
This last point bears emphasis. We should all agree that one way to
constrain the meaning of an expression, E, is to stipulate that E be interpreted
so that certain sentences containing E turn out true, or that certain inferences
involving E be truth-preserving. It can seem that such stipulations create
truth, or truth-preservation, on their own. But this is not the case, as was
illustrated by Arthur Prior (1960) in dramatic fashion. Prior imagined the
introduction of a new sentential connective ‘tonk’, stipulated to obey a
disjunction-like introduction rule “From ϕ infer ϕ-tonk-ψ”, as well as a
conjunction-like elimination rule “From (ϕ-tonk-ψ infer ψ.” The stipulations
do not result in the two rules being truth-preserving, for the rules would allow
us to infer any statement ψ from any other statement ϕ (first infer ϕ-tonk-ψ
from ϕ using the introduction rule, and then infer ψ using the elimination
rule). ‘Tonk’ is stipulated to stand for a meaning that obeys the two rules; but
there simply is no such meaning; ‘tonk’ cannot be interpreted so as to obey
the rules. Now, we do not believe the usual logical connectives to be like
‘tonk’. When we stipulate that conditional sentences are to be so interpreted
that every sentence of the form ‘If ϕ then ϕ’ is true, or when we stipulate that
‘and’ and ‘or’ obey their usual introduction and elimination rules, we believe
that these expressions can be understood so as to obey the stipulations. But
the case of ‘tonk’ shows that the stipulations do not, on their own, create the
truth, or truth-preservation.
The critique so far has not produced an argument against conventionalism;
it has merely cleared away obstacles to understanding, by enumerating
various things that conventionalists cannot mean by ‘true by convention’.
Now, this sort of clarification can be effective. For some, conventionalism
will lose whatever appeal it had, once the scales fall from their eyes.
Nevertheless, direct arguments against conventionalism would be welcome.
It is difficult to argue against a doctrine that has not been clearly
articulated. But what we can do is formulate the doctrine in schematic terms,
and then argue that so long as those schematic terms behave in a certain way,
the doctrine must be false. I will give two arguments of this form, assuming
the following schematic understanding of the doctrine of truth by convention:
“We can legislate-true the truths of logic.”
The first argument assumes that sentences that are about certain parts of
the world cannot be legislated-true. These are the parts of the world that I
cannot affect simply by wishing or pronouncing or legislating. I cannot, for
example, make true the sentence ‘My computer monitor has been thrown out
the window’ by wishing or pronouncing or legislating; I must defenestrate the
monitor myself, or pay or incite someone else to do it. Indeed, given my lack
of magical powers, the only statements that I can affect by mere wishing or
pronouncing or legislating would seem to be sentences about conventions or
related matters, such as which noises I make. We nonmagical humans can
legislate-true such sentences because they are about us. The first argument,
then, is this. Sentence (O) is not about us:26
(O) Either it is raining or it is not raining.
Since the only statements that we can legislate-true are those that are about
us, we cannot legislate-true the logical truth (O).
Talk of “aboutness” is admittedly slippery. Now, all the first argument
needs is that there is a sense of ‘about’ on which (O) is not about us, and on
which only sentences about us can be legislated-true. Still, an argument that
makes no appeal at all to aboutness may be welcome.
The second argument fits the bill. What it assumes about the schematic
notion of legislating-true is that i) I cannot legislate-true ‘It is raining’; and ii)
I cannot legislate-true ‘It is not raining’; and iii) if I cannot legislate-true ϕ,
nor can I legislate-true ψ, then I cannot legislate-true the disjunction ϕ or ψ
. In defense of iii): a disjunction states simply that one or the other of its
disjuncts holds; to legislate-true a disjunction one would need to legislate-
true one of its disjuncts. (To know (believe, promise, …) a disjunction, one
needn’t know (believe …) one of its disjuncts; but this needn’t undermine
iii), which is specific to legislating-true.) Given premises i)—iii), I cannot
legislate-true (O).
It is open, of course, for the defender of truth by convention to supply a
notion of legislating-true on which the argument’s premises are false. The
challenge, though, is that the premises seem correct given an intuitive
understanding of “legislate-true”.

1 Lewis (1983b, 347–8) considers replacing his primitive predicate of natural sets with a primitive
multigrade predicate of similarity amongst possible individuals. But this proposal does not generalize
along the second axis, and it arguably requires Lewis’s modal realism.
 2 Not that Armstrong and Lewis claim otherwise. They simply formulate the idea in their own
terms.
3 See Montague (1973) (Dowty et al. (1981) makes it go down easier).
4 Nor could he have defined natural properties as similarity-makers, at least not without some
serious chisholming (along the lines of Hirsch (1993, chapter 3, section 3, and appendix 2)). He doesn’t
count extremely specific intrinsic properties—for example, the property shared by all and only perfect
duplicates of a certain porcupine—as being perfectly natural, but their sharing certainly makes for
similarity. Indeed, their sharing makes for more similarity than the sharing of the properties that Lewis
does count as being perfectly natural: the fundamental properties of particle physics.
5 See also Sider (2004, p. 682).
6 Better: it does not require that there are facts in the fundamental sense of ‘there are’; see section
9.3.
7 Jason Turner suggested this regimentation, pointing out that we can think of the ‘it’ as being like
the ‘it’ in ‘It is raining.’ (The suggestion is not that this sentence operator is fundamental.)
8 Indeed, one can treat all expressions as standing for entities, as in categorial grammars (Gamut,
1991, chapter 4), although then there is a meaningful syntactic operation—concatenation—that stands
for no entity.
9 See Melia (1995; 2000).
10 By “distinctively semantic ideology” I do not mean metasemantic ideology like ‘refers’, which
relates semantic entities to human populations (though such ideology is surely not fundamental either).
I mean, rather, ideology that gives the intrinsic structure of the domain of semantic entities (for
instance: notions of conjunction, entailment, and the like, over propositions; or a fundamental predicate
functor ‘the property of ϕ-ing’).
11 More needs to be said about this argument; see chapter 13 for a fuller discussion.
12 Caveat: Lewis takes properties to be sets of possible individuals; and set-theoretic ontology and
ideology can be viewed as earning their keep in physics, not semantics, which answers my argument
against taking the facts of structure to involve semantic values. See the fuller discussion of this
argument in chapter 13.
13 Thanks to Robbie Williams for discussion of the following issues.
14 In a fundamental language, all and only primitive expressions carve at the joints. Thus (α)
will be true iff α is a primitive expression. One might worry that this somehow makes the operator
metalinguistic or trivial. But that would be a mistake. First, it doesn’t follow that (α) says that α
is a primitive expression. Second, you can’t learn that (α) is true simply by observing that α is a
primitive expression; you would also need to know that your language is a fundamental one.
15 Nerlich (1982, p. 274). See also Putnam (1975d, p. 70); McGinn (1981, 169–70); Yablo (2000,
section IX).
16 And even this does not fully isolate the complex predicate, since one is also querying the
variables.
17 We might introduce a nonfundamental language with grammatical resources to directly query the
material conditional. A metaphysical semantics (see section 7.4) for such a language might assign to the
query the truth-condition that each sentence of the fundamental language with the form (∼ϕ ∨ ψ)
be true.
 18 Compare Sider (2009, section 8).
19 Thanks to Jonathan Schaffer for discussion here, though the opposing approach should not be
attributed to him.
20 In fact, one can discern a usage of ‘ontology’—particularly prevalent in the philosophy of
physics literature—as a synonym for ‘metaphysics’.
21 Compare Armstrong’s (1983) own approach to laws of nature.
22 To be fair, Armstrong is not really asking the same set of questions as I am, so it is somewhat
misleading to describe him as accepting ontologism.
23 Ironically, one source of the lingering conventionalist strain may be a backlash against Quine’s
critique of analyticity. Quine’s overarching critique contained, as a part, the empirical assertion that
there are no facts of meaning (1951b). The manifest inadequacy of this view may have led to a failure
to appreciate the most powerful part of the critique, namely, his attack on truth by convention (1936;
1960a). See also Boghossian (1997).
24 Conventionalists include Ayer, Britton (1947), Carnap (1937, §69; 1950), and Malcolm (1940).
Pap (1958, chapter 7) contains a thorough (critical) discussion of conventionalism. See also Lewy
(1976, especially chapter 5), and Boghossian (1997).
25 It does a better job of showing that conventionalism cannot epistemically justify logic: we
already need to be justified in using logic before we can gain the justification that the conventionalist is
trying to supply
26 This could be regarded as a stand-alone premise; or it could be supported thus: ‘It is raining’ is
not about us; ‘It is not raining’ is not about us; the statements that are not about us are closed under
disjunction. Supporting the premise in this way would draw this first argument closer to the second.
7 Questions
Friends of fundamentality face some abstract questions about its nature. My
way of thinking about fundamentality—in terms of structure—is distinctive
in large part because of how I answer the questions. My answers: the
fundamental is complete, pure, subpropositional, absolute, determinate, and
fundamental.

7.1 Complete?
It is natural to assume that the fundamental must be “complete”, that the
fundamental must in some sense be responsible for everything.
Completeness seems definitive of fundamentality. It would be a nonstarter
to say that the fundamental consists solely of one electron: thus conceived the
fundamental could not account for the vast complexity of the world we
experience.
A preliminary formulation of completeness might run as follows: every
non-fundamental truth holds in virtue of some fundamental truth.1 But the
exact content of this formulation is far from clear. What do ‘in virtue of and
‘fundamental truth’ mean here? There are subtle issues about how to
understand these notions in terms of my official notion of structure. I want to
postpone discussion of those subtleties, however; so for now let us leave
completeness stated in this intuitive way. A fundamental truth (or fact),
intuitively, is a metaphysically basic or rock-bottom truth (fact). Facts about
the positions of subatomic particles would be, on most views, fundamental
facts, whereas the fact that some people smile when they eat candy would
presumably not be. ‘In virtue of’, intuitively, stands for the relationship
whereby the fundamental facts underwrite or give rise to all other facts. The
fact that some people smile when they eat candy holds in virtue, perhaps, of
certain facts about the states of subatomic particles (or, given a less
materialistic outlook: in virtue of these subatomic facts plus certain
fundamental mental facts).
Though I will be leaving ‘in virtue of’ at an intuitive level for now, I
should say up front that it is not to be understood in terms of modality,
truthmaking, or fact-identity. Thus I reject these conceptions of
completeness:
“All truths are necessitated by (or, supervene on) a fundamental description
of the world.”
“Every truth has a fundamental truthmaker.”
“Every fact is identical to a fundamental fact.”
The modal gloss imposes no meaningful requirement of completeness for
necessary truths, the truthmaking gloss requires a commitment to
truthmaking, and the fact-identity gloss requires a commitment to facts,
individuated in an appropriate way. (I’ll have more to say about truthmaking
and facts in chapter 8.) A less objectionable way to cash out “in virtue of”
would appeal to the—currently very popular—notion of ground: “all truths
are grounded in fundamental truths”.2 To say that the existence of a city is
grounded in certain facts about subatomic particles is to say that the latter
facts produce or account for or explain the existence of a city in a
distinctively metaphysical way. Although ground implies necessitation,
necessitation is insufficient for ground. (Examples like the following are
often given: snow’s being white does not ground its being the case that either
grass is green or grass is not green, even though it’s necessary that if snow is
white then grass is either green or not green; what grounds the truth of grass’s
being either green or not green is its true disjunct: grass’s being green.) Thus
the grounding approach lays down a meaningful requirement of completeness
for necessary truths.3 Further, properly understood, speaking in terms of
ground requires no commitment to truthmaking, or to facts, propositions, or
any other abstract entities. I will indeed sometimes speak of the grounding of
propositions or facts (and will likewise speak of in-virtue-of relations
amongst facts and propositions), but such talk is dispensable: one can always
construe ‘ground’ (and related locutions) as a sentence operator: “That ϕ
grounds its being the case that ψ” (Fine, 2001). Later on I will criticize the
ground-theoretic interpretation of completeness, and propose a different gloss
of “in virtue of”. But the ground-theoretic gloss is much closer to my own
than are those in terms of modality, truthmaking, or fact-identity, so if it is
familiar, it can serve as a working heuristic: “Every nonfundamental truth is
grounded by some fundamental truth.”

7.2 Pure?
There is a second assumption about structure that I think we ought to make—
what I call “purity”: fundamental truths involve only fundamental notions.
When God was creating the world, she was not required to think in terms of
nonfundamental notions like city smile, or candy.
As with completeness, there are subtleties about how exactly to
understand purity in my preferred terms. “Fundamental notion” is easy (it
means “carves at the joints”) but “fundamental truth” remains to be
explained. Let us postpone discussion of these subtleties just a little longer.
(But do notice that purity concerns two distinct concepts of fundamentality:
the concept of a fundamental notion, and the concept of a fundamental truth.)
Suppose someone claimed that even though cityhood is a nonfundamental
notion, in order to tell the complete story of the world there is no way to
avoid bringing in the notion of a city—certain facts involving cityhood are
rock-bottom. This is the sort of view that purity says we should reject. This
might seem obvious and uncontroversial. But in fact, purity has some very
striking consequences.
Here is a truth: there exists a city. Since the notion of a city is not
fundamental, purity says that this truth is not fundamental. No surprises so
far. Completeness then says that this truth holds in virtue of some
fundamental truth T—perhaps some truth of microphysics. So we have:
(1) There is a city in virtue of the fact that T.
Still no surprises.
But now consider (1) itself. Just like ‘There are cities’, (1) is a truth
involving the notion of a city. And so, given purity, it cannot be a
fundamental truth. And so, given completeness, it must itself hold in virtue of
some fundamental truth.
Now, I accept this consequence (given the way I will eventually
understand “in virtue of”). (1) is not itself fundamental (nor is any other in-
virtue-of truth, in my view). So (1) must itself hold in virtue of other truths.
But this is a nontrivial claim; and it is a claim that some people are going to
want to resist. A certain sort of primitivist about in-virtue-of, for example,
will refuse to explain truths like (1) in other terms. Purity stands in the way of
this sort of primitivism; it requires facts about the relationship between the
fundamental and the nonfundamental to be themselves nonfundamental. Thus
purity brings a heavy explanatory burden: it requires there to be facts in
virtue of which in-virtue-of-facts hold. But this is a burden we ought to
shoulder. The rock-bottom story of the world ought not to mention cityhood
at all, not even in sentences like (1). The primitivist about in-virtue-of who
opposes this is in an awkward position. On the one hand she must surely
acknowledge that most truths involving cityhood—truths such as “There is a
city”, “Philadelphia is a city”, “Candy can be purchased in most cities”, and
so on—are not fundamental; and she must surely feel the force of the thought
that this is in some sense because such truths involve the nonfundamental
notion of being a city. But then why the special exception for truths like (1)?
Admitting that (1) is a fundamental truth would drag the notion of cityhood
itself into the realm of the fundamental, since the admission concedes that the
fundamental story of the world cannot be told without bringing in cityhood.
Let’s think a little more about purity, and in particular, how it relates to
forms of primitivism. Consider the doctrine of modalism, which I understand
as the claim that necessity is a fundamental notion (in my terms, carves at the
joints). Now, many modalists would, I think, take this a step further, and say
also that modal truths are fundamental truths. But given purity, it cannot be
that all modal truths are fundamental. The modal truth that it is necessary that
all cities are cities, for example, must be nonfundamental given purity, since
it involves the nonfundamental notion of cityhood. But then given
completeness, “Necessarily, every city is a city” must hold in virtue of some
further fundamental truth N; and N cannot, given purity, involve the notion of
cityhood. Notice, though, that N can involve the notion of necessity, if
modalism is true. N might, for example, have the form “Necessarily, all Cs
are Cs”, where C involves only fundamental notions. (Think of C as a
“metaphysical definition” of the notion of being a city.)
So what we have learned is this: even if the modal notion of necessity is
fundamental, purity prohibits modal truths involving nonfundamental notions
from being fundamental. The only fundamental modal truths are an array of
“austere” or “pure” modal truths that give the necessary connections amongst
fundamental notions. (The array will include necessitations of logical truths
that contain only fundamental notions—such as “Necessarily, all Cs are
Cs”—but it will presumably include further truths; perhaps: “Necessarily,
nothing is both negatively and positively charged”, “Necessarily, if x is more
massive than y and y is more massive than z then x is more massive than z”,
and the like.)
I myself reject modalism (chapter 12); but I accept other sorts of
primitivism for which purity has analogous consequences. Negation, on my
view, is a fundamental notion; but since ‘eats’, ‘candy’, and ‘smile’ are
nonfundamental notions, purity implies that the truth “It is not the case that
something eats candy without smiling” is not a fundamental one. It holds,
perhaps, in virtue of some fundamental truth of the form “It is not the case
that something Es some Y but does not S”, where E, Y, and S are
“metaphysical definitions” of ‘eats’, ‘candy’, and ‘smile’, respectively.
Likewise, I think that quantifiers are fundamental (they carve at the joints);
but given purity, such truths as “There exists a city” are nonfundamental, and
hold in virtue of quantificational truths (perhaps of the form “There exists a
C”) that involve only fundamental notions. As with modality, even if
negation and quantification are fundamental notions, the only fundamental
facts involving those notions are pure—they involve those notions in
combination only with other fundamental notions.
The issue of purity can be further explored by discussing a particular
example—one that, I hope, an opponent of purity will agree is a sort of
crucial experiment, on which the issue turns. First some setup. Let C0 be a
predicate that describes New York City at the subatomic level in complete
detail (relationally as well as intrinsically). The following is presumably true:
(2) Necessarily, every C0 is a city.
(“City-zombies” are impossible.4) Finally, assume for the sake of argument
(what I do not myself believe) that modalism is true—necessity is a
fundamental notion.
Purity implies that (2) is not a fundamental truth, since it contains ‘city’.
An opponent of purity, I think, will take this as a good place to draw a line in
the sand. She will say that (2) is a fundamental truth; the modal connection
between C0 and cityhood is incapable of further explanation. (This is a good
place to draw the line in the sand because the modal connection between
macro-predicates like ‘city’ and their micro-realizations is particularly
resistant to reduction.5) I, on the other hand, accept purity’s consequence that
(2) is not fundamental (even under the pretense of modalism). My argument
has been simply that the fundamental story of the world ought not to mention
cityhood at all, not even within sentences like (1) and (2). (“When God
created the world, she did not need to use ‘city’.”) But this can be brought out
more vividly.
Think of a sentential operator like ‘necessarily’ as a machine. You feed it
a statement (proposition, interpreted sentence), and it spits out a truth-value.
The output true means that the input statement is necessarily true; the output
false means that the input statement could have been false. Think of the
fundamental facts of the form “Necessarily, ϕ ” and “It’s not the case that:
necessarily, ϕ ” as corresponding to the inputs ϕ that the machine is equipped
to handle—the inputs for which the machine “knows what to do”. If it’s a
fundamental fact that it’s necessary that all electrons are electrons, the
machine “knows what to do” with the input ‘all electrons are electrons’; it
spits back true. The thought in favor of purity is then this: the machine should
not know what to do with the input ‘Every C0 is a city. If the machine did
know what to do with this input, it would “know” how to inspect the notion
of a city, and figure out what its microphysical modally sufficient conditions
are. And this is an inappropriate capacity for the machine to have. When God
created the fundamental notion of necessity, she needed to endow it with the
fundamental capacity to interact with other fundamental notions (perhaps: all,
and, not, electron, and the like), but not with notions like city, smile, and
candy. (After all, fundamentally speaking there are no such notions.) This is
not to deny that (2) is true; it is only to deny that (2) is fundamentally true.
Demanding that (2) be fundamentally true demands more of a fundamental
notion of necessity than it has to give.

7.3 Purity and connection

In the previous two sections I formulated completeness and purity in intuitive
terms—“fundamental fact”, “in-virtue-of”—rather than in my own official
terms. This was not only for ease of digestion: it was also because the issues
are not particular to my own metaphysics of fundamentality. The question of
whether facts about the relation between the fundamental and the
nonfundamental (like (1)) are themselves fundamental, for example,
confronts everyone who takes the notion of fundamentality seriously. But
ultimately I want a formulation of completeness and purity in my own terms.
The next few sections head in that direction.
What is the relationship between the fundamental and the
nonfundamental? In what sense do all fundamental matters “boil down to” or
“derive from” or “hold in virtue of” fundamental matters, as completeness
says they do? How does the nonfundamental connect to the fundamental?
We learned something important about the nature of the connection in
section 7.2. As we saw, purity implies that claims like (1) and (2), which
describe in-virtue-of and modal connections, respectively, express merely
nonfundamental truths. But purity lets us draw a more general conclusion. On
any conception of the nature of the connection, connecting truths—true
statements expressing the distinctive connection between fundamental and
nonfundamental matters—are going to, like (1) and (2), involve
nonfundamental notions. (This is so by definition; connecting statements
relate nonfundamental notions to fundamental ones.) So on any conception of
the nature of the connection, the connecting truths are going to have to be
nonfundamental.
This in turn tells us something about how to attack the question of the
relationship between fundamental and nonfundamental: we should not attack
it using the method of metaphysical posit—by positing a metaphysically
fundamental notion (such as in-virtue-of, necessity, or truthmaking) for the
connection. The point of using the method of posit would be to claim that the
connecting truths involving the posited notion are fundamental truths, thereby
obviating the need to say in virtue of what such connecting truths hold. But
purity implies that the connecting truths could not be fundamental.
(Although purity tells us that it’s pointless to posit fundamental notions
like in-virtue-of, necessity, or truthmaking in order to attack the question of
the connection, it does not prohibit such posits per se.6 One could, for
example, posit a fundamental notion of in-virtue-of, and claim that in-virtue-
of truths involving only fundamental notions—for example, “There exist
things that are either negatively or positively charged in virtue of the fact that
there exist things that are negatively charged”—are fundamental truths. But
such truths are not connecting truths, and do not help with the problem of
how the fundamental relates to the nonfundamental. And absent some
independent motivation, there would be no reason for the posit.)
We have been focusing on the relationship between the fundamental and
the nonfundamental, worrying about whether facts about this relationship are
fundamental, and so on. It’s a little tempting (though only a little) to dismiss
such concerns by saying something like this: “Given the fundamental facts,
nonfundamental facts follow automatically; they’re not extra facts, somehow
over and above the fundamental facts; so there’s no need for facts connecting
the fundamental to the nonfundamental.” But the problem with this thought is
pretty transparent. What do the italicized phrases mean here? It’s natural to
construe them all in terms of in-virtue-of.7 To say that nonfundamental facts
“follow automatically” from, and are neither “extra” relative to nor “above
and beyond” the fundamental facts is simply to say that they hold in virtue of
fundamental facts. But then it becomes clear that the concerns have not been
dismissed at all, only relabeled. The fact that there being cities follows
automatically from, etc., a certain fundamental fact is exactly the sort of fact
whose status we have been questioning.
A related attempt to dismiss the concerns is a little more tempting: “The
fact that there is a city just is the fact that there is a C; so there is no need for
any fact connecting the ‘two’. The relationship between so-called
fundamental facts and nonfundamental facts is simply identity (so really we
shouldn’t call some of them fundamental and others nonfundamental); and
the status of identity facts is unproblematic.” It’s worth spending a little time
discussing this, since it illustrates a way of hiding metaphysical commitments
that will be important later (section 8.5). The objection’s crucial claim is that
the relationship between fundamental and nonfundamental is unproblematic
because it consists simply of identities like these:

(*) The fact that there is a city = the fact that there is a C.

Now, I agree that identities are in a sense unproblematic. But notice that the
singular terms flanking ‘=’ in (*) are not names, but are rather complex
singular terms, formed using the locution ‘the fact that’. This is important.
Grammatically, this locution combines with a sentence ϕ to form a singular
term, ‘the fact that ϕ’. Metaphysically, the locution functions as a connector
between whatever ideology is allowed to occur within ϕ and certain entities
(the facts). The use of this locution marks a serious metaphysical
commitment; intuitively, it is this locution, not the identity sign, that both
does the work in connecting the fundamental to the nonfundamental and also
creates problems with purity. This is clearest if we break (*) into the
following three components (where a and b are proper names of facts):
(i) the fact that there exists a city = a;
(ii) the fact that there exists a C = b;
(iii) a = b.
When the objector says that identities between facts raise no problems, this is
partly right: purity does not conflict with (iii)’s being a fundamental truth.8
Nor does it conflict with (ii)’s being fundamental (provided our fan of facts
regards ‘the fact that’ as a fundamental notion). But purity does rule out (i)’s
being a fundamental truth, since (i) contains ‘city’—even if ‘the fact that’ is
taken to be a fundamental notion. Return to the metaphor of the machine: ‘the
fact that ϕ’ is a machine that takes ϕ as input and picks out the corresponding
fact. If (i) expresses a fundamental fact, then the machine knows how to pick
out the appropriate fact when ϕ contains ‘city’; this is an ability that one of
reality’s basic building blocks should not have.

7.4 Metaphysical semantics

As we saw in the previous section, we cannot attack the problem of the
relationship between fundamental and nonfundamental by the method of
posit—by positing a fundamental notion for the connection. We should
instead take facts about the connection to be nonfundamental facts, which
may ultimately be explained in terms of fundamental facts that do not involve
a fundamental notion of connection.
This is not to say that we should require ourselves to actually specify what
those fundamental facts are, in terms of which facts about connection may be
explained, before we are willing to speak of the connection. That would
require us to possess a metaphysical reduction of the connection; and
metaphysical reductions are quite generally very hard to come by. (More on
this methodological issue in section 7.6.) Instead, we should introduce a
suitable undefined but nonfundamental notion for the connection. We should
explain and clarify the suitable notion as best we can, and there should be
reason to believe that it could in principle be metaphysically reduced; but we
need not ourselves possess a reduction.
What is this suitable notion to be? (“In virtue of” has been a mere
placeholder.) Earlier I mentioned reasons not to construe it in terms of
necessity, truthmaking, or fact-identity. Might the suitable notion be that of
ground, regarded as a non-fundamental notion? This is a close relative of my
own proposal; but for reasons to be given in section 7.9, I prefer a slightly
different, linguistic approach. The suitable notion, in my view, is that of a
metaphysical semantics—a certain sort of semantic theory. As we will see,
completeness may then be understood roughly as the claim that every
language has a metaphysical semantics.
A metaphysical semantics is a semantic theory with two distinctive
features. First, meanings are to be given in purely joint-carving terms. For
example, if the semantic theory takes the form of a truth-theory, then the
truth-conditions must be stated in perfectly joint-carving terms. More on this
below.
Second, the explanatory goals differ from those of linguistic semantics. In
one way they are more ambitious, and in another, more modest. Metaphysical
semantics is more ambitious in that by giving meanings in fundamental
terms, it seeks to achieve something not sought by linguistic semantics: to
show how what we say fits into fundamental reality. Metaphysical semantics
is more modest in that it tries to explain a narrower range of phenomena. The
semantic theories of philosophers of language and linguists attempt to explain
a broad range of phenomena, psychological and social, as well as narrowly
linguistic. According to a traditional conception (largely associated with
Frege), meaning plays a broad theoretical role: the meaning of a sentence is
conventionally encoded by that sentence, grasped by anyone who
understands the sentence, is communicated when the sentence is used;
sentence-meanings are the objects of thought and other propositional
attitudes, and so on. Successor theories have in some cases bifurcated this
role, but in other cases have expanded it, by taking semantics to interface
with adjoining theories of cognitive science, both linguistic (such as syntax)
and psychological. The metaphysical semanticist seeks to explain some of the
same phenomena as does the linguistic semanticist. For example, just like the
linguistic semanticist, she wants to help explain why English speakers will
point to the salient horse, rather than the salient car, when they hear the
sounds “Point to the horse!”; and like the linguistic semanticist, she will
invoke concepts like truth and reference to do so. But she is not concerned to
integrate her semantics with other linguistic or psychological theories. Thus
she is not trying to integrate her semantics with syntactic theory, for example.
And she is free to assign semantic values that competent speakers would be
incapable of recognizing as such, for she is not trying to explain what a
competent speaker knows when she understands her language. She might, for
example, assign to an ordinary sentence about ordinary macroscopic objects a
meaning that makes reference to the fundamental physical states of subatomic
particles. And she might simply ignore Frege’s (1952/1892) puzzle of the
cognitive nonequivalence of co-referring proper names, since she is not
trying to integrate her semantics with theories of action and rationality.
Returning to the first distinctive feature of metaphysical semantics:
suppose a metaphysical semantics for a language L takes the form of a truth
theory—a theory issuing in theorems of the form:9
Sentence S of L is true in L iff ϕ.
The requirement that meanings be “given” in purely joint-carving terms
amounts to the requirement that ϕ be phrased in purely joint-carving terms.
(‘Sentence of L’ and ‘is true in L’ are, of course, not purely joint-carving.
Remember that the notion of metaphysical semantics is not intended to be a
fundamental one.)
A truth theory is just one form that a metaphysical semantics might take.
A metaphysical semanticist might conclude that, rather than assigning truth-
conditions, a more explanatory approach would be to assign expressivist
assertion-conditions to normative discourse, say, or proof-theoretic assertion-
conditions to mathematical sentences. In the former case, the metaphysical
semantics would issue in theorems of the form:
Normative sentence S of L, as uttered by speaker x, is expressively
appropriate for x in L iff ϕ(x)
where ϕ(x) is a condition on speaker x’s attitudes. Here the requirement that
meanings be given in purely joint-carving terms requires ϕ(x) to be stated in
purely joint-carving terms. If psychological language is fundamental then
ϕ(x) may contain such language; otherwise (the more likely case, to my mind)
ϕ(x) will need to be phrased in the terms one would use for giving
metaphysical truth-conditions for factualist discourse about psychology.
Expressivism thus construed is immune to a common dilemma. Suppose
we extend the expressivist semantics to encompass a disquotational truth
predicate (and related vocabulary). Expressivism is supposed to be a form of
nonfactualism; but the claims one can assert in L (such as “The sentence ‘It is
wrong to kill’ is true”) now sound just like the claims that factualists make.
What happened to the nonfactualism?10
The answer is that the metaphysical semantics for factual discourse (say,
discourse about chemical or biological phenomena) has a different shape
from the metaphysical semantics for discourse about value. To a speaker’s
sentence, the former assigns a condition on the world—a closed sentence—
whereas the latter assigns a condition on the speaker—an open formula
applied to a variable to which the speaker is assigned; and this condition,
moreover, concerns the speaker’s mental state.11 True, this difference
disappears if one states a semantics for L from within L (provided L contains,
or is enhanced to contain, the appropriate semantic vocabulary). Such a
(nonmetaphysical) semantics might then take the form of a truth theory,
issuing in theorems like:
‘It is wrong to kill’ is true in L iff it is wrong to kill.
But the difference remains at the level of metaphysical semantics. A
metaphysical semantics could not take this form, assuming that ‘wrong’ and
‘kill’ fail to carve at the joints. And the expressivist who is opposed to
reductive naturalism will argue that no metaphysical semantics issuing in
truth-conditions of the form:
‘It is wrong to kill’ is true in L iff ϕ
where ϕ is a condition on the world (rather than the speaker) phrased in
fundamental terms, is adequate to our use of normative language.
Returning to metaphysical semantics of the truth-theoretic form: what
exactly is required of a fundamental truth-condition ϕ for a sentence S in such
a theory? We know from the literature on Donald Davidson’s (1967b)
approach to semantics that sentences ϕ and ϕ′ can have the same truth-value,
even necessarily so, despite the fact that ϕ is an appropriate truth-condition
for S while ϕ′ is not.12 ‘Snow is white’ is an appropriate truth-condition in a
(nonmetaphysical) semantics for ‘Snow is white’; neither ‘Grass is green’ nor
‘Snow is white and 2 + 2 = 4’ is appropriate. A metaphysical semantics must
successfully explain the linguistic behavior of the population in question, and
a truth-theory with necessarily true conjuncts tacked onto each of its truth-
conditions is presumably not explanatory, though it’s a hard question why
not. I have no particular answer to the question, though I suspect that the
approach of section 3.2 is applicable. And if the question proves intractable,
the metaphysical semanticist could abandon the Davidsonian approach. I
have chosen that approach largely because it’s simple, not because
metaphysical semantics is wedded to it.
What I have said about metaphysical semantics falls far short of a full
characterization, and it could surely be improved in various ways. However,
it is unlikely that there is any single best way to improve it. The notion of a
metaphysical semantics for a language, after all, is a high-level notion—a
notion posited in the course of explaining a high-level phenomenon, the
phenomenon of language-use by flesh and blood people; and there are often
different, equally good ways to explain such phenomena, as well as equally
good ways to carve the world into groups of phenomena to be explained.
Given our discussion of purity, this is a virtue. We should take a
nonfundamental, no doubt vague, and perhaps not even uniquely correct,
approach to the question of the relationship between fundamental and
nonfundamental.

7.5 Completeness and purity reformulated

We can now, at last, take up the question of how to formulate completeness
and purity in my preferred terms. Those theses were, recall, initially
formulated as follows:
Completeness Every nonfundamental truth holds in virtue of some
fundamental truth.
Purity Fundamental truths involve only fundamental notions.
The theses involve three crucial notions: fundamental truth, in-virtue-of, and
fundamental notion. A fundamental notion is just one that carves at the joints;
and in-virtue-of I am going to explain in terms of metaphysical semantics.
But how to understand the notion of a fundamental truth?
Here we encounter something interesting. There are two natural
definitions of fundamental truth in terms of the other two notions; what is
interesting is that on each definition, one thesis comes out trivial and the
other comes out nontrivial. The first natural definition of a fundamental truth
is that of a truth involving only fundamental notions. On this definition,
purity comes out trivial while completeness comes out nontrivial. The other
natural definition of a fundamental truth is that of a truth that does not hold in
virtue of any truth. Now purity becomes nontrivial while completeness
becomes—somewhat—trivial. I say “somewhat” because completeness does
not follow solely from the definition; but it does follow from the definition
together with a natural assumption about how in-virtue-of behaves, namely
that it is transitive and “well-founded” in the sense that if a truth holds in
virtue of any truth at all, then it is connected by an in-virtue-of chain to some
truth that does not hold in virtue of any truth (“no unbounded descending
chains of in-virtue-of”).13 So it would seem that completeness and purity are
in some sense not wholly independent, at least when the notion of a
fundamental truth is regarded as being definable in terms of the other two
notions.
My—somewhat arbitrary14—decision for how to define ‘fundamental
truth’ will be the first: a fundamental truth is a truth involving only
fundamental terms. Thus understood, purity becomes trivial. But notice that
its upshot has not disappeared. Completeness now says the following: “Every
truth that involves at least one nonfundamental notion holds in virtue of some
truth that involves only fundamental notions” and this has implications that
are analogous to the distinctive implications of purity discussed in section
7.2. Completeness, as it’s now understood, implies that even if a notion is
fundamental, any fact involving that notion together with some
nonfundamental notions must hold in virtue of pure or austere facts about that
notion, which do not involve any nonfundamental notions. (It implies, for
example, that even if necessity is a fundamental notion, the truth that every
C0 is a city must hold in virtue of truths that involve only fundamental
notions.)
So: completeness, as we’re now understanding it, says that every truth that
involves at least one nonfundamental notion holds in virtue of some truth that
involves only fundamental notions. It remains to eliminate “in virtue of” in
favor of metaphysical semantics. Actually, I wish simultaneously to
generalize completeness a little, to handle the case of expressivist and other
such language:
Completeness (new version) Every sentence that contains expressions that
do not carve at the joints has a metaphysical semantics.
By a metaphysical semantics for a sentence, I mean either a truth-condition,
an expression-condition, a proof-condition, or perhaps some other sort of
semantic condition, that is assigned to that sentence by some metaphysical
semantics for its language. This principle constrains the notion of carving at
the joints in accordance with the issues we have been discussing in this
chapter. There is no need for a further principle of purity, since the distinctive
implications of that principle discussed in section 7.2 are implied by this new
version of completeness. (But when I wish to emphasize these implications, I
will speak of “purity”.)
7.6 Metaphysics after conceptual analysis
How much should we expect from ourselves, if we attempt to actually give
metaphysical truth-conditions? Not much, I think.
We certainly should not expect to be able to give truth-conditions that are
intuitively correct in every possible world or conceivable circumstance.
Judging from the history of conceptual analysis, that is unattainable. In
retrospect, we should never have expected that project to succeed. Why
should there be any simple definitions, preserving intuitive or cognitive
significance, of any of our words in any other terms? Words aren’t generally
introduced as definitional equivalents of pre-existing phrases, and even then
they subsequently take on semantic lives of their own. Current meaning
derives from a long, complex history of use, which would seem unlikely to
result in neat equivalences. Our failure to come up with counterexample-free
definitions of ‘cause’, ‘knows’, and ‘good’ is not due to the philosophical
depth of these concepts. We’d have no better luck with ‘city’, ‘smile’, or
‘candy’. Words just aren’t neatly equivalent to other words, and there’s no
reason to expect them to be.
And it only gets harder if the truth-conditions must be stated in absolutely
fundamental terms, as metaphysical truth-conditions must be. Our ignorance
of the facts of physics then compounds our ignorance of the facts of meaning.
We have no chance of actually giving a metaphysical semantics for any
significant fragment of a natural language.
A more reasonable goal is the construction of “toy” metaphysical truth-
conditions. These will be toy in at least two ways. First, they needn’t match
with intuitively correct usage in absolutely all possible worlds or conceivable
circumstances. The mesh need only be approximate (the more mesh, the
better). Second, they needn’t be stated in perfectly fundamental terms.
Rather, they must be stated in terms that are fundamental enough for the
purpose at hand (and again, the more fundamental the better).
What is the point of toy metaphysical truth-conditions? One point is to
convince us that real, non-toy metaphysical truth-conditions exist.15
Consider, for example, the controversy over whether causation is
fundamental. To help resolve this controversy, we might try to produce
reductive (i.e., not involving ‘cause’) toy metaphysical truth-conditions for
‘cause’. If all attempts fail, then the case for fundamental causation will
receive a nice boost, especially if we discern in-principle reasons for the
failures. For if reductive metaphysical truth-conditions for ‘cause’ do not
exist (and if ‘cause’ lacks an expressivist or otherwise non-truth-conditional
metaphysical semantics), then completeness forces us to say that causation is
fundamental. But if a toy semantics can be produced, we needn’t recognize
fundamental causation (not because of completeness, anyway). The toy
needn’t be perfect. For its purpose is not to be a real metaphysical semantics,
but rather to convince us that there is a real metaphysical semantics, even if
that metaphysical semantics is too complex for us to discover. The purpose of
the toy is to convince us that the unfathomable workings of history and usage
can do their thing with ‘cause’, as well as with ‘city’, ‘smile’, and ‘candy’.
The reason philosophers obsess over the definition of ‘cause’ is not that they
think that a word must be defined before it’s legit (otherwise no philosophers
would speak of cities, smiles, or candy). It’s rather that it’s a live issue
whether causation is part of the fundamental furniture of the universe. We
don’t obsess over the definition of ‘candy’, not because we could easily
define it if we wanted to (we couldn’t), but rather because no one seriously
contemplates fundamental candy.
Conceptual analysis is out of fashion in metaphysics, but there is
uncertainty about what its replacement should be. Reduction?
Supervenience? Realization? The proposed replacements have tended to be
either inadequate or unilluminating (or both). The recent trend is to think in
terms of a kind of purely metaphysical analysis. There’s a bad idea in here
mixed in with a good one. The bad idea is that we should posit a fundamental
gizmo for the relationship between analysans and analysandum (truthmaking,
necessity, fact identity, ground). The good idea is Armstrongian:
metaphysical analysis is not linguistic analysis.16 But that leaves a big gap: if
not linguistic analysis then what? I say: metaphysical truth-conditions—toy
models of them, at any rate.

7.7 Metaphysical semantics for quantifiers

Metaphysical semantics are not required by definition to take any particular
form. They must presumably be compositional in some sense (since they
must be explanatory and hence cast in reasonably joint-carving terms, and
must contend with infinitely many sentences). But this still allows
considerable variation.
In particular, the form of metaphysical truth-conditions can depend on
what is fundamental. After all, metaphysical truth-conditions must be stated
in perfectly fundamental terms; thus which terms are in fact fundamental, and
what is true at the fundamental level, will affect what metaphysical truth-
conditions can look like.
An illustration comes from ontology. If fundamental ontology is
abundant, then the metaphysical truth-conditions for existential claims in
nonfundamental languages can be existential in form; but if fundamental
ontology is sparse, then these metaphysical truth-conditions need to take
some other form.17 To bring this out I will consider an example at length. I
will give (toy) metaphysical truth-conditions for certain statements of
chemistry, first assuming classical mereology and then assuming
mereological nihilism.
Assume classical mereology. More fully, assume for the sake of argument
that the joint-carving notions are those of logic: ∃, ∀, ∼, ∧, =, etc.; a
predicate < for mereological parthood; and the following physical predicates:
E (“is an electron”), P (“is a proton”), N (“is a neutron”), R (“orbits”), and U
(“is a nucleus”). (U is to apply to fusions of protons and neutrons that are
bound together into a nucleus by the strong nuclear force.18) Our fundamental
language is thus one in which only these notions are primitive. One can then
define, in the fundamental language, various mereological notions using < in
well-known ways. For example, one can define Oxy (“x overlaps y”) as
meaning ∃z(z < x ∧ z < y). And for any fixed positive integer n, one can
define an n + 1-place fusion predicate xFuny1 … yn (“x is a fusion of y1 …
yn”) thus:

And assume further that classical mereology’s principle of “unrestricted

composition” is true. For present purposes we may take this as the
assumption that for each n, ‘∀y1 … ∀yn∃x xFuny1 … yn’ is true.
Now consider a very simple language of chemistry, which is just like the
fundamental language except for containing new predicates: H (“is an atom
of hydrogen”) and L (“is an atom of helium”).19 Thus, this language contains
sentences like:

(1) ∃xHx
“There exists an atom of hydrogen.“
(2) ∃xLx
“There exists an atom of helium.”

We can give metaphysical truth-conditions for these sentences by, in essence,

defining atoms as fusions of their subatomic particles. Say that a sentence ϕ
in the fundamental language “translates” a sentence X in the language of
chemistry iff ϕ results from X by replacing occurrences of H and L according
to the following definitions:

“x is a hydrogen atom iff x is a fusion of an electron and proton where the

electron orbits the proton.”

“x is a helium atom iff x is a fusion of two electrons and a nucleus, where the
nucleus is a fusion of two protons and two neutrons, and the electrons orbit
the nucleus.”

We could say, then, that the metaphysical truth-condition of any sentence in

the language of chemistry is its translation. Thus, the metaphysical truth-
conditions of (1) and (2) are, respectively:

This was a particularly simple example of metaphysical truth-conditions.

Given classical mereology, our fundamental ontology already contained
helium and hydrogen atoms, so to speak; all we lacked was primitive
predicates classifying them as such. So the definition of translation was quite
easy; all we needed to do was define the predicates L and H in fundamental
terms. Things get more complex if a sparser fundamental ontology is true, as
we’ll now see.
Assume next that mereological nihilism is true—no mereologically
composite things exist. More fully, assume that ∀x(Ex ∨ Px ∨ Nx) is true
in our fundamental language—nothing exists other than electrons, protons,
and neutrons (these lack proper parts, let us pretend). Drop < from the
fundamental language (it is unneeded since nothing has proper parts). The
fundamental language must no longer speak of nuclei (there aren’t any); let
us replace, therefore, the predicate U with a two-place predicate B (“bonded”)
holding between the protons and neutrons that we formerly called the parts of
nuclei. Similarly, instead of speaking of electrons orbiting nuclei, let us now
speak of electrons as orbiting protons and neutrons. Let us assume that B is
transitive and symmetric, and reflexive over protons and neutrons, and that if
Rxy and Byz then Rxz—if an electron orbits one subatomic particle “in a
nucleus” then it orbits each subatomic particle “in that nucleus”.
What form must metaphysical truth-conditions for sentences of chemistry
now take? We can no longer translate claims of chemistry into fundamental
claims simply by giving definitions, in fundamental terms, of the predicates
H and L. If we proceeded that way, then the translation of sentence (1), i.e.,
∃xHx, would have the form ∃xψ(x), where ψ(x) is the proposed definition,
in fundamental terms, of Hx. Since only subatomic particles exist, and since
we want (1) to come out true in the language of chemistry, ψ(x) must apply to
subatomic particles. But ψ(x) was supposed to be the definition of ‘x is a
hydrogen atom.’20
We need a different strategy of translation. A natural approach is to
translate (1) and (2) into:

On this approach, the translation of ‘There exists a hydrogen atom’ is: ‘There
exist an electron and a proton, the first of which orbits the second’. The
translation omits reference to the hydrogen atom itself; it states the nihilistic
basis for the entire sentence ‘There exists a hydrogen atom.’21
In such a metaphysical semantics, the truth-conditions for sentences of the
form ∃xFx do not have the form ∃xψ(x), with ψ a translation of the
predicate F. Quantification over Fs disappears when we move from the
chemical sentence to its metaphysical truth-condition. Given this, it is natural
to say that the existential quantifier in the language of chemistry does not
mean what it means in the fundamental language used to give metaphysical
truth-conditions. Quantification in the language of chemistry is
nonfundamental quantification. We might make this explicit by using ‘there
is’ in the language of chemistry, reserving ‘∃’ for fundamental
quantification.22
Quantifiers in many languages—for example, ordinary languages in which
we quantify over tables and chairs—might in this way express
nonfundamental quantification, if fundamental ontology is sparse. Granted,
the metaphysical semantics for a more complex language will need to be
more complex than the toy semantics just mentioned. And particularly
austere views about fundamental ontology or ideology might make it
impossible to give metaphysical truth-conditions for some high-level
language—which might be a reason for abandoning such austere views. The
point here is just to demonstrate some of the resources available for giving
metaphysical truth-conditions, and to show how a sentence’s metaphysical
truth-conditions might look quite unlike that sentence, as with the truth-
conditions (1N) and (2N).

7.8 Metaphysics and the study of language

Suppose that mereological nihilism is true. Should linguistic semanticists—
not metaphysical semanticists, but rather, real live semanticists in linguistics
departments—then follow the lead of last section’s metaphysical semanticist,
and assign truth-conditions like (1N) and (2N) to sentences of the language of
chemistry? It might be thought that they should; otherwise, they would count
the chemist’s sentences as being false.
There is an alternative. If the language of chemistry can have a
metaphysical semantics that allows its sentences to be true despite
mereological nihilism, then why not the metalanguage used by linguists?
Linguists could then use sentences quantifying over atoms of hydrogen,
helium, and the like, rather than sentences like (1N) and (2N), to give truth-
conditions for the sentences of chemistry. The chemist’s ‘There exists a
hydrogen atom’ would then have, as a linguistic truth-condition, the linguist’s
homophonic sentence ‘There exists a hydrogen atom.’ This linguistic truth-
condition must be distinguished from the metaphysical truth-condition shared
by both sentences: ‘∃x∃y(Ex ∧ Py ∧ Rxy)’ (i.e., (1N)). Similarly, linguists
might assign to ordinary sentences about tables and chairs, truth-conditions
that themselves quantify over tables and chairs, provided the sentences of
their metalanguage have appropriate metaphysical truth-conditions.
There is a tradition in the philosophy of language according to which
linguistic and metaphysical inquiry should tightly constrain each other.
Davidson (1977) is a representative example. According to Davidson, a good
semantic theory must count ordinary sentences as being, for the most part,
true. Suppose the best semantic theory for a discourse assigns truth-
conditions to ordinary sentences that quantify over entities of a certain sort.
Then, since the ordinary sentences must be counted as true, the assigned
truth-conditions must be true; and so, the entities in question must exist. For
example, Davidson argued, the best linguistic theory of adverbial
modification assigns truth-conditions quantifying over events (1967a); thus
we must embrace an ontology of events. This is an example of linguistic
theory constraining metaphysics, but there is no reason in principle, given
this tradition, to resist the reverse direction of influence: that of metaphysics
constraining linguistic theory. Powerful metaphysical arguments that events
do not exist, for example, would give us reason to reject Davidson’s approach
to adverbial modification.
An advantage of the metaphysical semantics approach, as against the
Davidson-ian tradition, is that it allows for a looser relationship between
metaphysics and linguistics. We can agree with Davidson that linguistic
semantics ought to count ordinary sentences as being mostly true, without
needing to embrace an ontology of events, tables and chairs, or atoms of
chemistry, because ontology concerns fundamental existence, whereas
linguistic semantics is given in the metalanguage of linguistics, whose
quantifiers need not be fundamental. The linguist’s sentences about events,
atoms, and tables and chairs can be true even if such entities do not
fundamentally exist, given an appropriate metaphysical semantics for those
sentences.
The approach allows, more generally, for a looser relationship between
metaphysics and the special sciences. That relationship may be pictured thus:
Linguistics, psychology, economics, and other special sciences may be
carried out in their own languages—largely natural languages, enhanced here
and there with special-purpose vocabulary. Sentences of special-science
languages have metaphysical truth-conditions, but these are of no more
concern to the special scientist than the underpinnings of her discipline in
fundamental physics. Explanations of high-level data are given in the
language of these special sciences, not in the underpinning languages of
physics or metaphysics.
The advantage of this approach is that it allows linguists, psychologists,
and economists to be guided by considerations internal to linguistics,
psychology, and economics. It would be inappropriate to complain to an
economist that economies don’t really exist, or to insist that an engineer
rewrite her book on repairing potholes to reflect the fact that holes do not
really exist. Likewise, it would be inappropriate to require linguists to warp
semantics around metaphysical scruples about molecules of helium, or tables
and chairs, or events.
Conversely, it allows metaphysicians to be guided by considerations
internal to metaphysics. It has always seemed odd that insight into the
fundamental workings of the universe could be gained by reflection on how
we think and speak. Of course, such reflection can provide some constraint
on metaphysics. Human thought and speech are real phenomena, and so must
fit somehow into any adequate metaphysics. But this is a far cry from reading
off one’s fundamental metaphysics directly from the structure of thought and
talk.
Thus we have a limited Carnapian (1950) spirit of tolerance. Special
sciences can conduct their business without interference from metaphysics, if
their languages can be given a metaphysical semantics. (The same point from
another angle: metaphysics can be relatively free from interference from the
special sciences. Note how structure opens up breathing room for
metaphysics.) The tolerance is limited for two reasons. First, the
metaphysical semantics must not be too complex; otherwise one might
question whether the science is genuinely explanatory. Whether, and if so
when, an alleged science is unexplanatory for this sort of reason is a difficult
question about special-science explanation.23 Second, it cannot be just
assumed that a metaphysical semantics can be given. In particular, Carnap’s
insouciance about the ontological status of mathematical entities is not
justified, since it is particularly hard to see how mathematical language can
be given a metaphysical semantics if, fundamentally speaking, mathematical
entities do not exist. It is comparatively easy to see how chemistry and
biology could rest on top of physics, since the world of physics is fine-
grained enough to supply sufficiently many facts to underly chemistry and
biology. But the infinitary nature of mathematics presents a special challenge.
If one’s conception of the fundamental is overly sparse—if it contains for
example, neither mathematical entities nor a correspondingly rich structure of
modal or higher-order logical facts—then there may simply be no way to
give a metaphysical semantics for mathematical language.

7.9 Nonfundamental ground

I have given a linguistic account of connecting facts—facts connecting
nonfundamental to fundamental. It might be urged, against this, that the
matter of how chemistry, biology, economics, and so forth, relate to the
fundamental does not seem to concern language, and hence that we ought to
regard connecting facts as involving some nonlinguistic notion such as
ground. As we saw, purity requires connecting facts to be nonfundamental;
but a friend of ground might embrace this.
The friend of ground that I have in mind shares my main approach to
funda-mentality: his basic notion is that of structure, and he embraces purity;
it’s just that he accounts for the connection between fundamental and
nonfundamental in terms of ground, not metaphysical semantics.24 Given his
embrace of purity, he must say that ground-theoretic connecting facts—facts
like: there being C0s grounds there being cities—are themselves grounded in
facts that do not involve nonfundamental notions; and he presumably cannot
produce reductive definitions of ‘ground’, ‘city’, and other relevant terms that
demonstrate compliance with purity. However, I am in the same boat: I am
committed to the existence of a metaphysical semantics for sentences
containing ‘metaphysical semantics’, but I cannot produce that metaphysical
semantics.
It would be easy to overstate the difference between this approach and my
own. After all, a metaphysical semantics is supposed to explain linguistic
phenomena in purely fundamental terms, and the sort of explanation required
is distinctively metaphysical in nature since the meanings must be given in
fundamental terms. Still, there is a reason to prefer my approach: it handles
nonfactual discourse more smoothly.
Given my approach, there is a simple and natural way to distinguish
factual from nonfactual discourse: the difference is one of the “shape” of the
metaphysical semantics. The shape is truth-conditional in the former case,
and some other shape in the latter. For example, an expressivist metaphysical
semantics for evaluative discourse might take the form of an assignment of
expressively appropriate assertion conditions; and a formalist metaphysical
semantics for mathematical discourse might take the form of assertion
conditions that are sensitive to the proofs that the speaker possesses.
How will the friend of ground distinguish factual from nonfactual
discourse? I see two main possibilities. The first would be to exclude
nonfactual discourse from the scope of grounding. One way to implement
this would be to say that grounding only concerns facts; and since nonfactual
sentences do not express facts, grounding does not apply to them. But the
notion of fact thus invoked is in need of explanation (it cannot be
disquotational), and surely the explanation ought to have something to do
with grounding. So let us understand talk of facts disquotationally from now
on, and implement this first possibility differently: nonfactual sentences do
express facts, alright—call these “nonfactual facts” (sorry)—it’s just that
those facts are ungrounded. Moreover, nonfactual facts are not fundamental
(that is, they are not cast purely in joint-carving terms), which is what
distinguishes them from, say, physical facts, which are also ungrounded.
The problem here, however, is that the approach has nothing to say about
how nonfactual facts relate to the fundamental. Various sorts of nonfactual
facts (moral and mathematical, perhaps) are all lumped together as being
ungrounded and nonfundamental. My approach, in contrast, makes
distinctions within the class of nonfactual sentences, depending on the
“shape” of their metaphysical semantics.
To be fair, the friend of grounding can make distinctions within the class
of nonfactual facts by an indirect method, following Fine (2001). Unlike
mathematical facts themselves, facts about our beliefs about mathematics,
and metalinguistic facts about mathematical language, are grounded in the
fundamental, and may be thus grounded in a different fashion from how facts
about our beliefs in value, or metalinguistic facts about evaluative language,
are grounded in the fundamental. Perhaps the former facts are grounded in
facts involving proof whereas the latter are grounded in facts involving our
attitudes. So the friend of grounding can draw the distinctions that need to be
drawn. But they must be drawn so indirectly and so differently from how
analogous distinctions are drawn in the case of factual discourse.
A second possibility for treating nonfactual discourse would be to allow
nonfactual facts to be grounded. This is Fine’s approach (2001). Some
nonfactual facts are grounded in others: the fact that either murder or
snorkeling is wrong is grounded in the fact that murder is wrong. But not all
nonfactual facts are grounded: the fact that murder is wrong is perhaps an
example. This fact is ungrounded, but is distinguished from the physical facts
in that it is not fundamental.25 This second possibility is the more attractive
one, I believe, but it faces the same problem as the first. It has nothing to say
about how ungrounded nonfactual facts are related to the fundamental; it
lumps all ungrounded nonfactual facts—moral, mathematical, say—together.
As with the first possibility, distinctions can indeed be made, by examining
how moral beliefs and metalinguistic facts about moral language are
grounded; but these distinctions would be drawn indirectly.
My objection, then, is not that the grounding approach cannot draw the
distinctions that need to be drawn. It is that it does not draw them in the most
perspicuous way.
The root of the problem is the connection between the grounding approach
and the disquotational conception of fact. Think of the grounding approach as
follows. We begin in natural language, a language that is highly
heterogeneous in that it contains both factual and nonfactual discourse. We
then introduce a disquotational notion of fact, which applies to all asserted
sentences, whether factual or no. Finally, we apply the notion of grounding to
facts thus understood.26 Now, disquotational notions of fact, truth, property,
and the like are prized because of their ability to obliterate metaphysical
differences. When I lack information about what my neighbor has said—
including information about its status as factual or nonfactual—it helps to
have catch-all notions of fact, truth, and so on, by which I may express
agreement, disagreement, or otherwise make cognitive contact with my
neighbor. But however valuable this metaphysical neutrality of disquotational
notions is for ordinary purposes, it is a liability in the present, metaphysical,
context, for here we are trying to highlight differences in how our thought
and talk connect with fundamental reality. In the present context, it’s best not
to adopt such a catch-all conception of fact, for doing so already obscures
many of the differences we wish to capture—even if those differences can be
accounted for, down the line, in some indirect way.
There are other—related—reasons to prefer metaphysical semantics over
grounding. I’ll mention two. First, consider the approach to the liar paradox
according to which both the liar sentence and its negation must be rejected
(where rejection is not the same thing as assertion of the negation). On a
grounding approach, we seem not to be able to say anything at all here. Since
neither the liar sentence nor its negation is assertable, we can say nothing of
the form “ϕ grounds L”, where L is either the liar sentence or its negation
(assuming ‘grounds’ to be factive). But a natural approach can be taken if we
speak in terms of metaphysical semantics rather than ground; see section
10.6.
Second, consider nonfundamental natural language quantification.
Suppose that in the fundamental sense of ‘there is’, there are no such things
as statues or lumps of clay, but that natural language is governed by a
metaphysical semantics specifying that if some clay is appropriately shaped,
then the following sentence is true in English: “There exists a lump made
from that clay with modal properties mL, and there also exists a distinct statue
made from that clay, which has modal properties mS.” If we do not
semantically ascend, and ask simply after the grounds of facts construed
disquotationally, we will be led to an awkward place, as follows. Since there
exist a lump and statue as described, there must exist a pair of singular facts,
the fact that L has mL and the fact that S has mS, where L and S are the lump
and statue in question. Further, if these are distinct facts, there must surely be
some fundamental ground of one that is not a ground of the other—
something fundamental that grounds their distinctness. But no such
differential ground can be located.27 Now, it’s not as if the friend of ground
has no response. He might claim that although no ground differentiates the
pair of facts, the complex fact that L and S are distinct and instantiate mL and
mS, respectively does have a ground: namely, the fact that the clay exists and
is appropriately shaped.28 But it remains awkward that the facts in question
lack differential grounds. Intuitively, one wants to say, there really is no such
fact as that L has mL, or that S has mS, because there really are no such things
as L or S. A more satisfying picture of the situation is achieved by semantic
ascent. The metaphysical semantics for English provides metaphysical truth-
conditions for various statements about statues and lumps of clay, but it does
not do so by associating referents to singular terms like ‘S’ and ‘ L’. It rather
does so by associating complex truth-conditions for whole sentences
containing quantifiers over, or singular terms for, both statues and lumps.
These truth-conditions render ‘Lump L has mL and the distinct statue S has
mS’ true, despite the absence of distinct metaphysical truth-conditions for the
sentences ‘L has mL and ‘S has mS.’29
These examples—of evaluative discourse, the liar paradox, and
nonfundamental quantification—illustrate how adopting the disquotational
notion of fact and asking after the grounds of facts thus understood can
obscure what is important about the metaphysics of nonfundamental matters.
We can speak of facts and ground in these contexts, but a clearer view is
attained if we semantically ascend and describe how our discourse about
values, truth, and nonfundamental entities relates to fundamental reality.

7.10 Subpropositional?
Conceptions of fundamentality may be propositional or subpropositional—
they may be notions that apply to entire propositions, or to constituents of
propositions. To avoid reifying propositions and their constituents, we can
put it linguistically: a locution for talking about fundamentality might be
sentential—applying to entire sentences—or subsentential—applying to parts
of sentences.
Lewisian naturalness is subpropositional: it is properties and relations,
rather than entire propositions, that are evaluated for naturalness. The notion
of a fundamental truth is propositional: truths are entire propositions, not
proposition-parts. The notion of ground is propositional: entire propositions
ground one another.
 Structure is subpropositional. In my official regimentation, judgments of
structure take the form (α), where α may be a subsentential expression
(such as ‘is an electron’ or ‘and’). Thus the ultimate locus of fundamentality
is for me subpropositional. (I have no objection to propositional notions of
fundamentality— such as various notions of fundamental truth—so long as
they are defined in terms of the subpropositional notion of structure.)
There are both systematic and intuitive reasons for taking structure to be
subpropositional. The systematic reasons will emerge in section 8.3: a
subpropositional notion is explanatorily more powerful. The intuitive reason
is that subpropositionality is tied to the following attractive picture: there are
some fundamental “building blocks”—the “ultimate constituents of
reality”—and the nature of reality is given by the arrangement of those
building blocks.

7.11 Absolute?
Conceptions of fundamentality may be comparative or absolute. Lewisian
perfect naturalness, for example, is absolute: one says of a property or
relation that it is perfectly natural (or not) simpliciter. Lewis also spoke of
properties and relations being more or less natural; this is an example of
comparative fundamentality. (Lewis defined comparative naturalness in terms
of perfect naturalness and length of definitions; but an alternate approach
would be to take the former as basic and define the latter in terms of it.30)
Structure is absolute: I say ‘is structural’ rather than ‘is more structural’.
(In my official regimentation, the structure operator attaches to a single
expression rather than to a pair of expressions: “ (is an electron)”, “
(and)”, and so on.) (I have no objection to comparative notions of
fundamentality so long as they are defined in terms of the absolute notion of
structure. More on this below.)
The main reason for taking structure to be absolute is that facts about
structure (in interesting cases, anyway) are fundamental, whereas facts about
comparative fundamentality are nonfundamental. Why regard facts about
structure as being fundamental? Because structure is itself structural (section
7.13). While this allows some facts about structure to be nonfundamental, in
interesting cases, claims about structure cannot be further explained. Why
regard facts about comparative fundamentality as being nonfundamental?
Because of purity. The point of a comparative conception of fundamentality
would largely be to connect fundamental to nonfundamental matters; but
given purity, such comparisons could not be fundamental facts.
There is a further reason in favor of absolutism. As before, in interesting
cases the facts of structure are fundamental; and fundamental facts are always
determinate (section 7.12) and objective (chapter 4). But in many interesting
cases, it’s hard to believe that the facts of comparative fundamentality are
determinate and objective—consider, for example, the question of whether
geological notions are more fundamental than biological ones.

7.11.1 Absolutism and comparative structure

I have not been practicing what the previous section preaches. Throughout
this book I have spoken of comparative structure: of carving “reasonably
well” at the joints, carving “equally well” (though not perfectly) at the joints,
carving “badly” at the joints, and so on. Examples: a nonsubstantive question
was characterized as one whose answer depends on which of various
candidate meanings we adopt, where the candidates are equally joint-carving,
and where no other candidate is more joint-carving (section 4.2). The
doctrine of reference magnetism appealed to imperfect joint-carving (section
3.2). Explanations were required to be cast in joint-carving terms, and in the
case of special-science explanation, “joint-carving” cannot mean perfectly
joint-carving (section 3.1). Relatedly, think of the traditional “levels” picture
of the sciences, with physics at the bottom, chemistry next, and the other
sciences arranged in some order or other on top. This ordering can be thought
of as corresponding to the comparative fundamentality of the notions of those
sciences.
Thus we need a comparative notion of structure in many of the
applications. This may be reconciled with absolutism by distinguishing the
fundamental notion of structure from the notion of structure in those
applications; it is only the fundamental notion which is absolute. (Or we may
rephrase thus, to avoid putting weight on notion-identity: the fundamental
facts about structure involve only absolute structure; all facts about
comparative structure are nonfundamental.) Talk of comparative structure
must have metaphysical truth-conditions in terms of absolute structure (and
other fundamental notions).
How to give such metaphysical truth-conditions? How to define
comparative structure? I do not know. But I can suggest several elements to
employ in a definition.
One element comes from Lewis. Lewis’s notion of a perfectly natural
property or relation is absolute; but he went on to define a comparative
notion. A property or relation is more or less natural, he said, depending how
short a definition it can be given in a perfectly natural language—a language
in which all predicates stand for perfectly natural properties and relations
(1986b, p. 61). This approach could be generalized in the case of structure:
one notion is “definitionally more structural” than another, let us say, iff it
has a shorter definition in a fundamental language—a language in which all
expressions are structural.
Lewis’s approach has been thought to face serious challenges (Hawthorne,
2007; Sider, 1995; Williams, 2007), chief among which is that it counts every
two properties that require an infinite definition (of the same cardinality) as
being equally natural.31 But this is a limitation only when properties require
infinite definitions, and it is far from clear that properties of interest do. It
might appear otherwise because such properties have infinitely many
realizations. A complete specification of a certain kangaroo, in perfectly
natural terms, down to the last microphysical detail, is just one realization of
the property of being a kangaroo; and there are infinitely many such
realizations, since (for example) kangaroos can vary continuously in length.
So, let us grant, the property of being a kangaroo has an infinite definition in
a perfectly natural language: the disjunction of its realization-predicates. But
this needn’t be the shortest definition. The property of being a kangaroo
might also have a finite functional definition:
x is a kangaroo =df x has some realization (property) or other that plays
role R
if the perfectly natural language allows quantification over properties, and if
role R is finitely definable in that language.
The class of finitely definable properties is particularly rich if
quantification over arbitrarily high-order properties and relations is available.
Begin with an initial class of finitely definable properties. Then give finite
definitions of further, higher-order, properties, that make reference to the
initial properties. (For example, one might define a space S of the initial
properties, as well as a geometry on S, and then define properties of the form:
having some member of region R of space S, where R is finitely definable.
Also one might make use of causal notions—viz., having some member of
region R which plays such-and-such causal role—provided causal notions
can themselves be finitely defined.) Next give finite definitions of still
higher-order properties, which can make reference to the previously defined
properties. And so on. Given higher-order resources, there is reason to be
optimistic that properties of interest are finitely definable in perfectly natural
and fundamental languages.
(The procedure requires quantification over properties, but such
quantification needn’t be fundamental. Suppose, for example, that the
language in which the definitions are cast includes no quantification over
properties, but does include quantification over sets. Then we could simulate
quantification over properties by identifying fundamental properties and
relations with their extensions, and other properties with quasi-linguistic set-
theoretic constructions out of fundamental properties. The conjunction of p
and q, for example, could be identified with the triple 〈∧, p, q〉, where ∧ is a
symbol standing for conjunction.32)
A further worry about the Lewisian approach is that mere length of
definitions is an inadequate measure. Shouldn’t the degree to which a
definition is “disjunctive” render the defined notion less natural? But there
are strategies for refinement available here. One might, for instance, require
all definitions to be in some standard form (prenex disjunctive normal form,
say, if the language is a first-order predicate calculus); and one might
evaluate definitions, as given in this standard form, by a more complex
measure that takes more into account than length (the number of disjuncts
might, for instance, be taken to count against a definition more than the
average number of conjuncts per disjunct).
The first element, then, is the notion of being definitionally-more-
structural-than. A second element is “lawlikeness”: the degree to which a
notion figures into simple and strong generalizations. Green is more lawlike
than grue because green figures in simpler and stronger generalizations than
does grue. (As with Lewis’s account of laws, the relevant notions of
simplicity and strength must be spelled out.)
Lawlikeness has little utility on its own, since one can cook up simple and
powerful generalizations with even highly non-joint-carving notions—recall
Lewis’s (1983b, p. 367) “law” ∀xFx, which implies everything true because
F is a predicate true only of things in the actual world. But suppose we
restrict our attention to somewhat deffinitionally structural notions—green,
grue, kangaroo, being-a-snail-or-a-kangaroo, and the like. With respect to
such notions, lawlikeness is an interesting measure—green, for example, is
more lawlike than grue. It is only when applied to extremely definitionally
unstructural notions, such as F in Lewis’s “law”, that lawlikeness is
uninteresting. (It doesn’t matter how exactly we sharpen the idea of a
“somewhat” definitionally structural notion; on any reasonable sharpening,
lawlikeness will generate an interesting ordering.)
The third element is really a class of elements. Philosophers of science
have done much subtle and detailed work on explanation in the special
sciences. Lawlikeness is one concept they use to characterize how the special
sciences are explanatory, but there are many others: probability, unification,
cause, and so forth.33 These concepts can be understood without distinctive
fundamental metaphysics; indeed, much of this work is designed to show
how the special sciences can be explanatory in an ultimately physical world.
So any of these concepts could play a role in a definition of a comparative
notion of structure. The degree to which a notion plays a role in causal
statements, for example, could be appealed to in such a definition, provided
the notion of causation is itself given a metaphysical semantics.
So far the elements have been more or less “objective” (modulo some
arbitrariness around the edges). A fourth element is more subjective. Perhaps
economics deserves its place in the hierarchy of “levels” partly because of its
value to us as a tool for prediction and control of things we care about.
Disinterested Martians who for this reason do not develop economic theory
would not be missing out in the way that they would if they did not develop
chemistry or physics. In general, the comparative structuralness of a notion
might be in part a function of how important it is to us. section 2.5 argued
against subjectivism about structure. But the opponent there was a general
subjectivist about structure, including absolute structure. It would be really
bad if there were no objective privilege at all to the physical conceptual
scheme—recall “knee-jerk realism”. But all I am allowing here is a partial
subjectivism about one facet of comparative structure. This is comparatively
benign, and does not conflict with knee-jerk realism, at least of the sort I find
in myself. Absolute structure provides a bedrock of objectivity, on top of
which a modicum of subjectivity in comparative structure may be overlaid.
I can supply, then, these elements. A nice project would be to actually
produce a definition of comparative structure from them. I’d like to do this,
but I don’t know how. To anyone trying this at home: if you get discouraged,
remember three things. First, remember that what you are attempting is a
definition of a complex high-level notion in perfectly fundamental terms. No
one can give a fully adequate definition of this sort for any complex high-
level notion. A reasonable goal in such endeavors is persuading ourselves
that some metaphysical semantics exists, by producing toy accounts that work
well in a decent number of cases and have no in-principle defects. (Producing
the elements of the definition is a start.) Second, don’t be alarmed if a certain
amount of vagueness or arbitrariness creeps into your attempts. These can be
tolerated since comparative structure is not intended to itself be structural.
Third, remember that there is no need to settle on a single definition once and
for all; perhaps different applications call for different definitions of
comparative structure. Granted, my case for primitivism about structure has
been that positing a single notion of structure illuminates multiple domains
(chapter 2, and see also section 7.13). But that single posited notion is
absolute structure, which can be used to define multiple sorts of comparative
structure.
Comparative structure, as defined from these elements, will not itself
carve perfectly at the joints. But it can nevertheless carve reasonably well at
the joints, provided the definition from the elements is a reasonable one. Thus
its explanatory value needn’t be compromised. While good explanations must
be cast in reasonably joint-carving terms, they needn’t be cast in perfectly
joint-carving terms (there can be good explanation outside of physics, after
all). Some vagueness or arbitrariness may enter the definition, which
introduces the potential for nonsubstantive questions. But it does not follow
that all questions about comparative structure are nonsubstantive. Only the
questions whose answers turn on the vagueness or arbitrariness are rendered
nonsubstantive. This is as it should be; some questions about comparative
structure are nonsubstantive. For example, it may not be a substantive
question whether geological notions are more structural than biological ones.
It is compatible with this that the usual hierarchy of “levels” of special-
science notions is by-and-large objective; the nonsubstantivity is only
“around the edges”.

7.11.2 Absolutism and infinite descent

It might be objected that absolutism requires a “ground floor”. For the
alternative to a ground floor is some sort of infinite descent of ever more
fundamental facts or notions, and making sense of infinite descent seems to
require a comparative notion of fundamentality
Distinguish three sorts of “ground floor”: ideological, mereological, and
propositional. Only the first is demanded by my absolutism; but the first is
unobjectionable.
The first sort of ground floor is ideological. The alternative is infinite
ideological descent—a chain of ever more fundamental notions, with no
notions more fundamental than every member of the chain. My account does
indeed require an ideological ground floor. For me, the facts about
fundamentality are in the first instance facts of the form “Notion α is
structural” ( (α)); and it’s hard to see how to construct a metaphysical
semantics for talk of comparative notion-fundamentality in terms of such
facts that would allow infinite ideological descent. (Certainly the account of
definitional structure in section 7.11.1 does not allow it.) But infinite
ideological descent is a seriously weird hypothesis, and seems unproblematic
to deny. The hypothesis denies the existence of a book of the world, a
complete perfectly fundamental description of reality, since for any chosen
concepts, more fundamental concepts could be chosen.
Infinite mereological descent—a.k.a. the metaphysician’s beloved
hypothesis of “gunk”—is comparatively mundane. An object is gunky if each
of its parts has further proper parts; thus gunk involves infinite descent in the
part-whole relation. The corresponding sense of ‘ground floor’ is
mereological: atomism. My theory of fundamentality does not require
atomism, because gunk can be described using perfectly structural
ideology.34 Suppose, to take a toy example, that a certain gunky patch varies
continuously in color from one wavelength to another. As Frank Arntzenius
and John Hawthorne (2005, section V.2) show, the color facts about this
patch could be taken to emerge from the totality of facts of the form the
average wavelength of part x of the patch is λ. And there are various bits of
fundamental ideology one could introduce to characterize these facts about
average wavelength (such as ‘x has a higher average wavelength than y’ and
‘x’s average wavelength together with y’s average wavelength equals z’s
average wavelength’—the details will depend on one’s general approach to
quantities.) For a more physically realistic example, Jeffrey Sanford Russell
(2010) shows how to characterize the topology, mereology, and measure-
theoretic facts about gunky space using a finite list of primitive notions; his
primitive notions could be taken to carve at the joints.35
The final sort of ground floor is propositional, or factual: a ground floor of
facts. A propositional ground floor is opposed to infinite propositional
descent— an infinite descending chain of the in-virtue-of relation over facts
or propositions. Now, as we’ve seen, “in-virtue-of” is not my preferred way
of talking about fundamentality. Still, we can ask: Can I accommodate—in
terms of structure—the kinds of scenarios that would be described by the
friend of in-virtue-of as involving infinite propositional descent? The answer
depends on which of two kinds of scenario is alleged. In the first sort, each
level in the chain is a new “sort” of proposition—a proposition involving new
ideology that is more fundamental than ideology involved in propositions
higher up in the chain. I cannot accommodate this scenario, since it would
require infinite ideological descent. But I can accommodate a tamer scenario,
in which the propositions throughout the descending chain are all of the same
sort—they all involve the same ideology. For example, it might be claimed
that the proposition that a certain gunky object has a certain mass is true in
virtue of a proposition about the masses of its parts under some finite
decomposition, that the latter proposition is true in virtue of propositions
about the masses of still smaller parts under a more fine-grained but still
finite decomposition, and so on infinitely:

a is 1 g mass in virtue of …
… b being 0.5 g and c being 0.5 g (where a = b + c), which holds in virtue of
…
… d being 0.25 g and e being 0.25 g and f being 0.25 g and g being 0.25 g
(where b = d + e and c = f + g), which holds in virtue of …

(“+” signifies fusion). Even though this scenario would be described by the
friend of in-virtue-of as one of infinite propositional descent, my description
of it does not require infinite ideological descent. I can simply state all the
propositions involved in the chain:

a is 1 g
b is 0.5 g
c is 0.5 g
a = b +c
etc.

leaving out the in-virtue-of claims (since I renounce them), and add the only
relevant claim about fundamentality, which is that the ideology common to
all the propositions—namely, mereological and mass-theoretic—is absolutely
structural. (The exact nature of the mass-theoretic ideology common to all the
propositions depends on one’s views about the nature of quantity. One view
would be that the ideology is comparative: “x is more massive than y”, or “x’s
mass together with y’s mass equals z’s mass”.)
So: the only limitation stemming from the absoluteness of structure is that
I cannot accommodate infinite ideological descent. This is no real limitation,
I say, because there is no reason to suppose that this weird scenario obtains.
“Objection: there is reason, namely empirical inductive reason, to believe
in infinite ideological descent. Physicists once thought that everything
depended on the features of molecules. But molecules gave way to atoms,
which gave way to protons, neutrons and electrons, which have given way to
quarks, leptons, and gauge bosons. Each time a new type of particle was
discovered, physicists posited new features of the newly discovered particles,
whose distribution accounted for, but could not be accounted for in terms of,
the distribution of the distinctive features of the older particles. This historical
progression of theories will probably continue forever, so there are no
ultimate features on which everything depends.”—This is a bad argument, for
a few reasons. First, it is an induction from only four cases. Second, by
moving from “finite” observations to an “infinite” conclusion, the argument
makes a big leap. Compare it to the argument that there must be infinitely
many people, since for each person we’ve observed, there exists a taller
person.36 Third, the argument fails by drawing a conclusion that is drastically
dissimilar from the initially observed pattern. The initially observed pattern is
an historical progression of physical theories:
Theory 1: The fundamental features are those of molecules.
Theory 2: The fundamental features are not those of molecules, but are
rather those of atoms.
Theory 3: The fundamental features are not those of atoms, but are rather
those of protons, neutrons, and electrons.
Theory 4: The fundamental features are not those of protons, neutrons, and
electrons, but are rather those of quarks, leptons, and gauge bosons.
The conclusion drawn is that there are no fundamental features, since for any
physical feature had by any particle, there are further features (had by smaller
particles) that do not depend on the first feature. But this conclusion isn’t
inductively suggested by the initial pattern. The conclusion has the superficial
appearance of a kind of limit point of the initial pattern, if that pattern were
infinitely extended. By moving through Theories 1–4, so the idea goes,
scientists have been moving closer and closer to the conclusion. But this
impression vanishes upon closer inspection. Each Theory in the progression
does not add a new layer of fundamental features, but rather ditches the
previous Theory’s layer (since it regards the previous layer as just depending
on the newly hypothesized layer). Extending the pattern indefinitely results in
a series that simply has no intuitive limit. For comparison, imagine a
countably infinite series of chairs: c1, c2,.… Suppose first that in scenario 1,
c1 is filled; in scenario 2, chairs c1 and c2 are each filled; in scenario 3, chairs
c1, c2, and c3 are each filled; and so on. I suppose there’s some sense in which
the limit of this series is a scenario in which all the chairs are filled. But
consider a second series in which only c1 is filled in scenario 1, only c2 is
filled in scenario 2, only c3 is filled in scenario 3, and so on. This series has
no intuitive infinite limit. The imagined infinite extension of the progression
through Theories 1–4 is like the second series.
This third criticism of the inductive argument depends on the fact that I
construed Theories 1–4 in terms of “fundamental features”, by which I meant
absolutely fundamental features. Suppose they were construed instead in
terms of comparative fundamentality:
Theory 1a: Molecules have certain distinctive features.
Theory 2a: Atoms have certain distinctive features, which are more
fundamental than those of molecules.
Theory 3a: Protons, neutrons, and electrons have certain distinctive features,
which are more fundamental than those of atoms.
Theory 4a: Quarks, leptons, and gauge bosons have certain distinctive
features, which are more fundamental than those of protons, neutrons,
and electrons.
If continued infinitely, this progression does seem to have an infinite limit
(it’s like the first chairs series): that for every feature, there are more
fundamental features—infinite ideological descent. But this is not
dialectically effective against absolute fundamentality, since comparative
fundamentality was assumed from the start, in the characterizations of
Theories 1a–4a.
“Even if there is in fact no infinite ideological descent, infinite ideological
descent is epistemically possible, and hence should be allowed by any good
theory of fundamentality.”—If infinite ideological descent is epistemically
possible then my theory of fundamentality is not epistemically necessary.
That’s ok! My theory is intended to be an educated guess about the nature of
the world, not as some sort of a priori deduction that must hold with
certainty. Neutrality on “first-order” questions like that of infinite ideological
descent is not a reasonable constraint on the metaphysics of fundamentality.
A metaphysics of fundamentality is supposed to give the truth about the
nature of fundamentality, not provide a dialectically neutral framework in
which to conduct first-order debates.
“Even if infinite ideological descent is not actual, it is nevertheless
metaphysically possible, so a good theory of fundamentality should permit
it.”—As we will see in section 12.5, the Humean theory of modality to be
defended undermines such arguments from possibility. In brief, the
impossibility of infinite descent amounts to little more than its nonactuality,
so there is no distinctively modal way to support it.

7.12 Determinate?
In addition to being pure, complete, subpropositional, and absolute, I hold
that the fundamental is also determinate.
“The fundamental is determinate” is not particularly clear, and improving
on the situation is difficult because there are so many different ways to
understand what “determinacy” amounts to, but perhaps we can put it thus.
First, no special-purpose vocabulary that is distinctive of indeterminacy—
such as a determinacy operator or a predicate for supertruth—carves at the
joints. Second, fundamental languages obey classical logic. The combination
of these two claims is perhaps the best way to cash out the elusive dogma that
vagueness and other forms of indeterminacy are not “in the world”.
This is not to deny the value of determinacy-theoretic vocabulary, or
supervaluationism, or nonclassical logic; it is just to deny them a place at the
fundamental level. They might yet play a role in explaining vagueness in
nonfundamental languages (see section 10.6).

7.13 Fundamental?
Last question: is fundamentality fundamental?
There are two questions here. First, is fundamentality a fundamental
notion? And second, are facts about fundamentality fundamental facts?
Given my subpropositional approach to fundamentality, the first question is
primary. In my terms it is the question of whether structure is itself structural
—of whether carving at the joints carves at the joints. In the official
regimentation: is it the case that ( )?37 My answer is yes.
My answer to the second question is: not all of them. In my terms, the
question is whether facts about structure are fundamental facts. Facts about
structure that involve only structural notions are indeed fundamental facts;
but facts about structure that involve nonfundamental notions cannot be
fundamental facts, given purity. For example, the fact that grue does not
carve at the joints involves the nonfundamental notion grue, and so cannot be
a fundamental fact. Thus it must, given completeness, have a metaphysical
truth-condition stated in purely fundamental terms. But that truth-condition
can mention structure, since structure is a fundamental notion. The truth-
condition might have the form “G does not carve at the joints” (officially:
“not (G)”), where G is a “metaphysical definition” of ‘grue’.
Back to the first question. My reasons for saying that structure is structural
emerge from considering an opposing viewpoint.
A vivid test for whether a given expression, E, carves at the joints is this:
did God need to think in E-terms when creating the world? Clearly, she
needed to think in terms of quantification, mass, distance, and so on;
accordingly, those notions carve at the joints. But did she need also to think
in terms of structure? It is natural, I must admit, to say no. All she needed to
do was decide which objects to create, how massive to make them, and where
to put them; she didn’t need also to consider whether quantification, mass,
distance, and the rest were structural.
If structure is not structural, then completeness requires all statements
about structure to have metaphysical truth-conditions. Since those truth-
conditions must be stated in perfectly structural terms, they cannot contain
‘structure’; but they can contain terms that are structural—terms of physics,
perhaps. What might such truth-conditions look like? I will sketch an answer
that I will call “Melianism”, since it contains elements of an intriguing view
due to Joseph Melia.38
Let’s simplify by discussing Lewisian (perfect) naturalness, rather than
the more general notion of structure. We are after a definition of ‘natural’ in
terms of natural properties and relations (by hypothesis these do not include
naturalness itself). Where N1,N2 … are the natural properties and relations,
the Melian definition is:
P is natural =df P = N1 or P = N2 or …
This is a highly disjunctive definition. The Melian embraces this. There is
no need for a nondisjunctive, explanatory notion of naturalness, he says,
because naturalness itself is never invoked in explanations. Whenever Lewis
would cite naturalness in an explanation, the Melian cites particular natural
properties. Why are these two electrons exactly alike? Because they have
exactly the same natural properties, Lewis says. The Melian says instead:
because they both have charge c, mass m, and spin s.
Objection: “We may not know which natural properties the electrons
possess.”—The Melian reply is that this is an epistemic limitation of ours, not
a deficiency in the proffered explanation. The best explanation of the
electrons being duplicates cites the particular natural properties they share,
even if we do not possess that explanation.39
Objection: “The Melian’s definition is circular since ‘natural’ was used to
pick out the list N1 …”—There would be objectionable circularity only if
‘natural’ occurred on the right hand side of the definition. But it does not.
‘Natural’ was used to pick out the list by description; and perhaps our only
access to the list—and thus to the definition—is via this description.
Nevertheless, the definition itself contains only the list, not the description.
Further, the Melian could avoid appealing to ‘natural’ in even this indirect
way if he accepted the epistemology of section 2.3. He could then pick out
the list by the description ‘properties that figure in our best theory’. (This
would not be an identification of naturalness with the property of figuring in
our best theory. Naturalness would remain identified with a disjunctive
property of the form being N1 or N2 or …; the description would represent
our best guess as to the identity of the disjuncts.)
“Objection: Melianism diminishes the significance of duplication,
intrinsicality, lawhood, and other notions defined in terms of naturalness.”
Here we approach a more telling objection. Since Melian naturalness is
highly disjunctive, so will be the defined notions. So although the Melian
agrees with Lewis on the first-order questions—on which objects are
duplicates of which, which properties are intrinsic, what the laws are, and so
on—he must regard Lewis’s focus on these first-order questions—questions
of duplication, intrinsicality, law, and so forth—as being arbitrary, because
based on highly disjunctive notions. This is a strange predicament. The
Melian is trying to achieve Lewis’s aims on the cheap, but his theory implies
that these aims are metaphysically arbitrary and not particularly worth
pursuing.
And the predicament is worse than strange; Melianism undermines all of
the applications of naturalness. The Melian admits that a notion so
disjunctive as Melian-naturalness cannot be explanatory, and tries to get
around this by claiming that explanations can always cite particular natural
properties rather than naturalness itself. We can explain why two electrons
are exactly alike, he said, by pointing out that each has charge c, mass m, and
spin s. But that is no explanation; explanations must cite generalizations.
Explanations of similarity-facts require generalizations about similarity;
explanations of meaning-facts require generalizations about meaning;
explanations of substantivity-facts require generalizations about
substantivity; and so on. Moreover, explanations must cite generalizations of
sufficient scope. It would be no good to cite the generalization that any two
electrons sharing charge c, mass m, and spin s are similar; the generalization
is too specific. The generalizations must cite naturalness—or better, structure.
But then structure cannot have a Melian definition, if the generalizations are
to be explanatory.
The argument of this book is that the explanatory power of our overall
theory is enhanced by positing structure. Must the posited notion be itself
(perfectly) structural, or could it have a definition in structural terms? What
we have seen so far is that it could not have a Melian, disjunctive definition,
for that would undermine its applications.
Could it have a nonMelian, nondisjunctive definition? Then even though
structure would not be structural, it might still be capable of figuring in
explanations, just like notions of the special sciences.
But it is hard to see how a definition of structure could avoid being
disjunctive. Chemical kinds have nondisjunctive (metaphysical) definitions
because instances of chemical kinds are reasonably physically alike, and
physical notions are structural. Biological kinds have nondisjunctive
definitions because instances of biological kinds are functionally, if not
physically, alike; and functional notions have reasonably nondisjunctive
definitions in structural terms. But consider the instances of the notion of
structure. These include notions of mass, charge, spatiotemporal distance, set
membership, conjunction, disjunction, and universal quantification, let us
suppose. This class of notions is neither physically nor functionally unified.
Nor does it seem to be unified in any other way that would allow a
nondisjunctive definition.40 Other than the fact that all its members are
structural, the class is highly heterogeneous.
Relatedly, consider the notions to be defined in terms of structure—
similarity, intrinsicality, laws and explanation, meaning, induction, physical
geometry, and substantivity. The instances of any one of these notions are
heterogeneous in physical and functional and other “first-order” ways. When
x is similar to y and x′ is similar to y′, there needn’t be any physical or
functional or other first-order commonality between the pairs 〈x, y〉 and 〈x′, y
′〉; there needn’t be any physical or functional or other first-order
commonality between any two laws of nature; and so on. What unifies all the
pairs of similar objects, and all the laws, is just the fact that they involve
structural notions in certain ways; and the structural notions themselves are
neither physically nor functionally unified.
The “first-order heterogeneity” of structure, and of structure-involving
notions, is an in-principle obstacle to a nondisjunctive definition of structure.
Thus the choice is stark: either adopt extreme realism about structure—
holding that structure is itself structural—or else give up altogether on
explanations that invoke structure, which is tantamount to giving up on
structure itself. My choice is for the former.
The status of metaphysics itself hangs on this choice. In their loftiest
moments, metaphysicians think of themselves as engaged in a profoundly
important and foundational intellectual enterprise. But if fundamentality is
highly disjunctive, the field of metaphysics itself—which is delineated by its
focus on fundamental questions—would be an arbitrarily demarcated one.
Although it offers explanatory power (and a pleasing self-conception for
metaphysicians), extreme realism about structure raises some difficult
questions. I’ll mention three.
First, the argument for saying that structure is structural was that this is
needed to insure that structure can take part in genuine explanations. But look
at the notions other than structure involved in the putative explanations:
simplicity, correctness of interpretation, candidate meaning, and so on. These
do not seem structural either. Doesn’t this already undermine the genuineness
of the explanations? And if it doesn’t—if the genuineness of the explanations
is compatible with their involving nonstructural notions—then why can’t the
explanations remain genuine if structure isn’t structural?
This can be answered. Genuineness of explanation does not require
perfectly structural notions, as we see from the special sciences. It is enough
that simplicity, correctness of interpretation, and the rest, are somewhat
structural. The reason for thinking that structure cannot be merely somewhat
structural is its first-order heterogeneity—if structure is not perfectly
structural then it is disjunctive and therefore highly nonstructural.
The second question I find more challenging. In section 6.3 we considered
two possibilities for regimenting talk of structure, each involving the operator
. On one, had a very flexible grammar, and on the other, the language
needed to be supplemented with dummy variables. Neither smacks of
fundamentality. Each seems to require our fundamental languages to be much
more complex than they would otherwise need to be. The complexity could
be avoided by taking talk of semantic values as fundamental; could then
be a predicate of semantic values. But semantics does not smack of
fundamentality.
The third question is also challenging. Realism about structure requires a
fundamental posit, and such posits are generally to be avoided. The concern
is particularly pressing given my own preference, expressed many times in
this book, for simplicity. I have been disdaining, and will continue to disdain,
primitive modality, law, cause, tense, logical consequence, higher-order
quantification, and other such luxuries. But when it comes to my own pet
concept, structure—it might be alleged—my scruples go out the window.
It’s not that I have no answer to this charge. My answer is that structure is
no luxury, since it cannot be reduced without loss. Still, it smells fishy,
doesn’t it? This is a serious challenge facing the audacious doctrine of
realism about structure.

1 A refined principle would allow a nonfundamental truth to hold in virtue of multiple fundamental
truths taken collectively.
2 On grounding see Fine (2001; 2010; 2011); Rosen (2010); Schaffer (2009a); Schnieder (2011);
and the papers in Correia and Schnieder (2012).
3 This is just one way in which ground improves on the coarser-grained notion of modality; see
especially Fine (2001).
4 Cf. Bennett (2006).
5 See Lewis (1986b, pp. 150–7) and section 12.9 of this book.
6 Thanks to Bruno Whittle for this point.
 7 Similar remarks apply if we construe them in terms of necessitation or truthmaking, etc.
8 So long as facts are taken to exist in the fundamental sense, anyway; otherwise it might be
objected that names like a somehow bring in nonfundamental notions.
9 There is no need for ‘iff’ to have a sense that is somehow distinctive of metaphysical reduction. It
can have the same sense that it has in any explanatory theory—the material biconditional, say.
10 See Dreier (2004); Fine (2001). My solution to the problem is in the vicinity of Dreier’s and
Fine’s: we all agree that the reason expressivism is nonfactualist (Dreier says “irrealist”) has something
to do with expressivism’s implications for how value relates to the fundamental. I prefer my approach
to Fine’s for the reasons given in section 7.9 below.
11 There are other differences between truth-conditional and expressivist semantics, especially
when they are integrated with a broader theory of the mind. For example, truth-conditions and
expressive appropriateness conditions play different roles in communication and deliberation.
12 See Soames (1992) against the Davidsonian approach.
13 Thanks to Karen Bennett here. For more on well-foundedness, see section 7.11.2.
14 I could instead adopt the second definition of ‘fundamental truth’, rendering completeness
(somewhat) trivial. The corresponding version of purity would then be: truths that do not hold in virtue
of any truth involve only fundamental notions. That is: any truth that involves a nonfundamental notion
must hold in virtue of some other truth. Modulo the assumptions of transitivity and well-foundedness of
in-virtue-of, this is equivalent to the result of the other decision.
15 Another is to study how our concepts relate to one another. Even a simplistic semantics might
illuminate the overall shape of a network of concepts consisting of wrongdoing, blame, guilt, shame,
and the like.
16 See, for example, Armstrong (1978a; b).
17 This corresponds to Fine’s (2003) distinction between proxy and non-proxy reductions.
18 On a more plausible view, U would nonfundamental, and defined in terms of mereology and a
fundamental predicate for the strong nuclear force. The approach in the text is for simplicity.
19 I have in mind hydrogen-1 and helium-4, respectively.
20 Well, one could (at the price of artificiality) specify a tricky translation scheme in which ψ(x)
does indeed apply to subatomic particles, namely, those subatomic particles that are “part of hydrogen
atoms: . One would need to make adjustments elsewhere. For
example, the language of chemistry’s predicate P could not be translated as the fundamental language’s
predicate P; otherwise ∃x(Px ∧ Hx) (“something is both a proton and a hydrogen atom”) would be
translated as a truth.
21 Here is a general translation scheme of the desired sort. Let F be the set of sentences in the
fundamental language that express the assumptions we are making about our fundamental predicates. F
thus includes the claim that B is transitive, symmetric, and reflexive-over-protons-and-neutrons. So in
any model, M, of F, M(P) ∪ M(N) is, if nonempty, partitioned into equivalence classes under the
relation M(B) (I use “ M(π)” for the extension of predicate π in M). Think of these equivalence classes
as “nuclei” (they obviously aren’t really nuclei; M is only a model). For each such equivalence class, c,
call the ordered pair (c, h) an “atom”, where e ∈ h iff: e ∈ M(E) and 〈e, o〉 ∈ M(R) for some o ∈ c.
Each “atom” is an ordered pair of a “nucleus”—a set of “protons” and “neutrons”—and the set of
“electrons” that “orbit” the members of the “nucleus”. Next we construct an augmented model, M′, by
adding the “atoms” to M’s domain. (The added “atoms” are to be new; so if any are already present in
M’s domain, first pair them with some arbitrarily chosen object not in the transitive closure of M’s
domain.) Let the extensions of E, P, N, B, and R in M′ be as they were in M. And let M′ also interpret
the extra predicates of the language of chemistry: let the extension in M′ of < include all pairs 〈d, 〈c, h〉〉
where 〈c, h〉 is an “atom”, and either d ∈ c or d ∈ h, plus further pairs so that < satisfies the axioms of
mereology; and assign extensions to H and L in the obvious way. (Place an “atom” 〈c, h〉 in M′(H), for
example, iff h has exactly one member, and c has exactly one member, which is a member of M′(P).)
Finally, say that ϕ in the fundamental language translates X in the language of chemistry iff for every
model M of F in which ϕ is true, X is true in the corresponding augmented model M′. Notice that a
given sentence might now have more than one translation. (Notice also the use of set theory to specify
the translations, even though the fundamental theory in question was nominalistic. There’s no
immediate conflict since only metaphysical truth-conditions themselves, not the description of how to
arrive at them, must be stated in purely fundamental terms.)
22 See also section 9.3.
23 As pursued, for example, by Fodor (1974) and Kim (1992).
14 In chapter 8 I consider views that replace the notion of structure with ground and/or related
notions.
25 To say that this fact is grounded in our attitudes, for example, would turn moral expressivism
into a form of descriptivism.
26 Fine (2001) ultimately prefers to regard ‘ground’ as a sentence operator rather than as a predicate
of facts (or propositions); but similar remarks apply since the sentences to which this sentence operator
may be applied are heterogeneous; they may be either factual or nonfactual.
27 Compare Sider (2008b). deRosset (2010) raises related issues about grounding.
28 Compare Dasgupta (2010) on plural ground and Fine (1994b) on reciprocal essence.
29 Depending on details, this might call for a refinement of the completeness principle.
30 I discuss this variant in Sider (1993a, chapter 3).
31 There is a parallel worry about a Lewisian account of special-science laws: Lewisian laws must
be simple; unnatural properties detract from simplicity; special-science properties are equivalent to
infinite disjunctions of physical properties and hence highly unnatural.
32 There is obviously some arbitrariness in this construction. For a defense of such arbitrariness in
another context, see Sider (2006, section 2).
33 See Strevens (2006) for an overview.
34 I myself reject gunk (for independent reasons); see Sider (2011).
35 See also Arntzenius (2008).
36 Thanks to Cian Dorr here.
37 On some regimentations of talk of structure, one might raise a worry about the intelligibility of
the question: “Being structural is a property of semantic values; the question would be about the
proposition that the property of being structural instantiates itself; there is no such proposition.” I’m not
sure if I accept any part of this objection, but at any rate, my preferred regimentation for talking about
structure avoids the worry because is an operator.
38 Melia put this forward in conversation, so please don’t blame him for what follows!
 39 Compare Melia (1995; 2000).
40 Or better: that would allow a nondisjunctive and objective definition. The class does seem
unified by the fact that its members are all indispensible in our best theories.
8 Rivals
A metaphysics of fundamentality consists of some distinctive concepts for
characterizing fundamentality and a theory of how those concepts behave. I
have been defending one such metaphysics, whose distinctive concept is
structure. In this chapter we will discuss some rival approaches, based on
rival concepts. These rivals are close cousins of my own approach, in that
much of the work I do with structure could be done using my rivals’
concepts. Thus the criticisms I will make should be viewed as sparring
amongst friends.

8.1 Fine’s concepts

Let us begin with the two concepts introduced by Kit Fine in his article “The
Question of Realism”. The first is that of one proposition’s grounding
another.1 When p grounds q then q holds in virtue of p’s holding; q’s holding
is nothing beyond p’s holding; the truth of p explains the truth of q in a
particularly tight sense (explanation of q by p in this sense requires that p
necessitate q). Fine also considers the possibility of taking the fundamental
locution to be a two-place sentence operator, “ϕ because ψ”, rather than a
two-place predicate of propositions, “p grounds q”. I’ll alternate between
these formulations.
The second concept is that of a proposition’s being real. A proposition is
real if it is fundamentally the case, if it describes reality’s intrinsic structure.2
As with ground there is a sentence-operator formulation: “In reality, ϕ.”
The next few sections discuss two views about the metaphysics of
fundamentality based on Fine’s concepts: one based on ground and reality,
the other based solely on reality. Now, although there are elements of these
views that Fine himself would endorse, it should not be assumed that he
would endorse the whole of either of them (hence I call them merely
“Finean”). For one thing, I am particularly interested in the versions of these
views according to which facts about fundamentality are themselves
fundamental; but Fine does not commit himself to such a view.3 For another,
Fine’s purpose for introducing his concepts was not primarily to provide a
metaphysics of fundamentality, in the sense at issue here. It was rather to
clarify what is at stake in debates over “realism”. Fine views such debates as
concerning what is factual, where a factual proposition is one that is either
real, or grounded in propositions that are real. The difference between anti-
realists and realists cannot be articulated in less “metaphysical” terms,
according to Fine, since anti-realists about a subject matter S may well accept
truth, bivalence, and so on, for statements about S. But what they do not
accept is that statements about S are factual in his metaphysical sense.
Today’s ethical anti-realists are not so crude as to deny that murder is wrong,
or that ‘murder is wrong’ is true; but on Fine’s view they must deny that the
proposition that murder is wrong is factual: it is neither real nor is it grounded
in what is real.

8.2 First Finean view: grounding and reality

One rival conception of fundamentality makes use of both of Fine’s notions.
According to this conception, the facts about fundamentality consist of the
facts of the forms: “In reality, ϕ”, and “ϕ because ψ”, or the corresponding
facts in the propositional formulation.
(Following Fine, we can define a basic proposition as one that is grounded
in no proposition.4 This raises the possibility of a simplification: might we
define a real proposition as a basic one? Fine rejects this identification, for
two reasons. First, a nonfactual proposition—for example, an ethical
proposition according to the moral expressivist—might lack a ground and
thus count as basic, but being nonfactual, it would not be real.5 Second, Fine
allows that some real propositions might ground others, for example if matter
were infinitely divisible (2001, p. 27). Thus, for Fine, basicness is neither
necessary nor sufficient for reality. My discussion will be independent of this
issue.)

8.2.1 Ground and purity

Consider grounding facts involving nonfundamental notions, such as:
(1) The proposition that there is a C0 grounds the proposition that there is a
city.

Are such facts fundamental?

Some defenders of the present approach will want to say that they are.
After all, it’s very hard to say in any reductive way how facts about cities
relate to the fundamental. This is the main form of this first Finean view that I
wish to consider.
And my main argument against it is that it violates purity (section 7.2).
Purity prohibits fundamental facts that involve nonfundamental notion, and
(1) involves the nonfundamental notion of being a city.6
This argument must be put carefully in order to avoid begging the
question. My original statement of purity was this: “Fundamental facts
involve only fundamental notions.” But ‘fundamental notion’ is my term—it
means joint-carving— and the dispute with the Finean is over whether to use
his terms or mine. A dialectically appropriate challenge must be formulated
in Finean terms, in terms of ground and reality rather than in terms of carving
at the joints. Rephrased, the argument is as follows. The Finean view under
consideration says that true propositions about ground and reality are
themselves real. So if (1) is true, then the proposition it expresses is real. But,
I say, no proposition about cities—that is, no proposition involving the notion
of being a city—is real. (This premise plays the role of purity.) Thus the
Finean view under consideration is false. It is false because it implies that the
fundamental story of the world includes facts about cities, namely, facts
about how propositions about cities are grounded.
The argument assumes that no propositions about cities are real.
Intuitively, the property of being a city is not the kind of subject matter that
would show up in a fundamental description of the world. The Finean might
reject this assumption by saying that, while propositions about cities that do
not involve ground are indeed never real, propositions of ground about cities
are sometimes real.7 In particular, he might say, the proposition expressed by
(1) is real. This move is intuitively very unsatisfying. What would motivate
the special exception?
The problem with this first Finean view, then, is that it does not heed the
lesson of section 7.3: facts about the connection between the fundamental and
the nonfundamental should not be taken themselves to be fundamental;
otherwise the domain of fundamental facts becomes infected with facts about
cities, smiles, and candy. The problem arises because the view embraces two
things: i) a comparative (recall section 7.11) distinctive concept—ground;
and ii) fundamental facts in which that comparative concept connects
fundamental to nonfundamental matters. There are other rival views that also
embrace i) and ii). Consider Cian Dorr’s (2004; 2005) theory of metaphysical
analysis, for example. Dorr’s key locution for talking about fundamentality,
“to R(x1…) is to ϕ(x1…)” (as in “to be a city is to be a C”) is comparative. So
any follower of Dorr who also held that facts like “to be a city is to be a C”
are fundamental would face the problem as well. Similarly, a view based on
the notion of truthmaking, which held additionally that facts about
truthmaking are fundamental, would also face the problem (see section 8.5).
It is natural to think of grounding as a kind of metaphysical causation. As
a Humean I’m suspicious of metaphysical pushings and pullings; but it’s
awfully hard to stop thinking in terms of them. It’s overwhelmingly tempting
to say that it is either raining or snowing because it is raining (Fine, 2001, p.
22), or that my being a philosopher makes it the case that someone is a
philosopher. But we reductionists can take the same attitude towards
metaphysical causation as towards everyday causation: it reduces in some
way to facts that don’t involve metaphysical causation. As a first pass at the
cases of disjunction and existential quantification, there is a sort of
“metaphysical covering law” underwriting each case. In the first, the law is
that whenever p is true, so is p∨q; in the second, it is that whenever a is F,
something is F. Most crudely: metaphysical causation occurs when an
instance is subsumed under such a metaphysical law. No doubt the account
will need to be refined ad nauseum in familiar ways; but the possibility of
such a project largely dispels the temptation to admit fundamental
metaphysical causation. I’m also inclined to reject fundamental laws of
metaphysics, and so to regard the covering metaphysical laws as being mere
regularities (the regularities are metaphysical laws because they involve only
the abstract and general notions of interest in metaphysics). Opposition to
metaphysical laws is of a piece with opposition to primitive logical
consequence and primitive modality, on which see chapters 10 and 12.
I have been opposing the Finean who says that propositions like (1) hold
in reality. What if the Finean denies this?
If propositions like (1) are not real, they must be grounded in real
propositions (for surely they are factual). It will be difficult to say what the
grounding propositions are. But I am in the same boat: it is similarly difficult
to provide a metaphysical semantics for sentences about metaphysical
semantics.
If propositions like (1) are not real, then what about propositions about
reality? Do they too fail to be real? Answering yes would put the Finean in
the same boat as the Melian of section 7.13. It would then be hard to avoid
saying that facts about ground and reality are grounded disjunctively, in
which case no good explanations could be given using those notions.
But the Finean might say that unlike propositions about ground,
propositions of the form “In reality, ϕ” are real. This view does not face the
purity problem, since in true propositions of this sort, ϕ will presumably not
involve notions like city, smile, or candy. Thus, a powerful position remains
to be considered: the position that the fundamental facts about fundamentality
are those involving Fine’s concept of reality. I discuss this position in section
8.3.

8.2.2 Ground and infinite descent

Friends of ground face the question of whether there could be an infinite
descent of ground, a chain of propositions p0, p−1, p−2 …, where p−1 grounds
p0, p−2 grounds p−1 …, and where no proposition grounds the whole
sequence. Many think that there is something deeply incoherent about the
idea. As Schaffer (2010c) puts it, “Being would be infinitely deferred, never
achieved.”8
Suppose for the sake of argument that infinite descent of ground ought
indeed to be banned. Compliance with the ban might seem easy. It is only
exotic metaphysical hypotheses (such as gunk), one might think, that would
result in infinite descent of ground; and such hypotheses may safely be
rejected. But as we will see, some perfectly ordinary phenomena involve an
infinite descent of ground. Or rather, the ground-theoretic description of the
phenomena involves infinite descent; the structure-theoretic description does
not. Thus the notion of ground gives rise to a sort of paradox that is
immediately dissolved by thinking instead in terms of structure.
Consider first the facts of distance. It’s arguable that distances,
fundamentally, are not direct connections between points, but are rather path-
dependent.9 (Distance is treated this way in general relativity, for example.)
Suppose distance is indeed path-dependent, and consider the fact that a
certain path through space is one meter long. It’s surely overwhelmingly
natural for the friend of ground to say that the path is one meter long because
it is made up of two (nonoverlapping) half-meter long parts. (This is
consistent with saying that it’s also the case that the path is one meter long
because it’s made up of a quarter-meter part and a three-quarter meter part.)
But now, the half-meter long parts of the initial path have the lengths that
they do because they are made of quarter-meter parts, which themselves have
their lengths because of the lengths of still smaller parts, and so on. There is
no end to this descent of grounds, even if space is atomic, provided space is
not discrete. For even if space is atomic (nongunky), there are no shortest
paths.10
An analogous argument can be given for the topological notion of
continuity. A friend of ground should describe a continuous curve as being
continuous because its left and right halves are continuous (and continuously
connected). Those halves are themselves continuous because their parts are
continuous, and so on. As before, even if space is atomic, provided it is not
discrete there is no end to this descent of ground, because there are no “atoms
of continuity”—there are no smallest continuous paths.
These examples do not call for infinite ideological descent. Metrical and
topological facts can be stated using perfectly fundamental notions (‘open
set’, in the latter case, and terminology from the mathematics of differential
manifolds in the former). The demand for infinite descent has a distinctively
fact-theoretic source. Intuitively, the source lies in the way the facts in
question are quantificationally defined in terms of the distribution of the
fundamental ideology over infinite collections of points. Actually this can be
illustrated by a simple mathematical example. Just like topological and metric
facts, set-theoretic facts can be stated with perfectly fundamental notions
(perhaps just a single predicate for set-membership). Set theory thus does not
call for infinite ideological descent. But using the primitive notions of set
theory, we can define the notion of an infinite set. And on the face of it, the
facts about infinite sets call for infinite descent of ground. For example, it
would seem that a certain countably infinite set, A = {a1, a2,…} is infinite
because it is the union of {a1} and the infinite set {a2,a3…}; and the latter is
infinite because it is the union of {a2} and the infinite set {a3, a4…}; and so
on.
8.3 Second Finean view: reality
A second rival conception of fundamentality dispenses with ground (thus
avoiding the objection from purity11), and invokes only Fine’s notion of
reality. On this second Finean view, the facts about fundamentality consist of
the facts of the form “In reality, ϕ.” (I particularly have in mind the version
that adds that all such facts themselves hold in reality.)
Call a sentence structural iff each of its words carve at the joints. Call a
proposition structural iff it is expressed by some structural sentence. And call
a proposition (or sentence, depending on context) a structural truth iff it is
true and structural. Structural truths behave very much like Fine’s real
propositions. Each can be thought of as a way of cashing out the intuitive
notion of a fundamental truth. Nevertheless, there are important differences.
The notion of a fundamental truth behaves differently under the two
conceptions.

8.3.1 Explanation of fundamental truths

The differences come from the fact that structure is subpropositional, whereas
reality is propositional (recall section 7.10). The locus of fundamentality for
the Finean is the whole proposition, whereas for me it is the proposition-part.
Fundamentality is holistic for the Finean, atomistic for me. Suppose, for
example, that it is a fundamental truth that there are electrons. For me, this
boils down to the fact that i) there are indeed electrons; and ii) electronhood
and existential quantification carve at the joints. For the Finean, it amounts
simply to the fact that the entire proposition that there are electrons is real.
The subpropositional locus does seem more intuitive. One wants to say
that it’s a fundamental truth that there are electrons because of something
about electronhood and existential quantification. Now, I cannot lean too
heavily on the thought in this form since the ‘because’ surely has the sense of
ground. But the thought has an epistemic/explanatory reading that has real
bite.
There are patterns in the fundamental truths. Perhaps the following are all
fundamental truths:

e1 is an electron;
e2 is an electron;
There is an electron;
There is a quark;

and perhaps none of the following truths are fundamental:

New York is a city;

Tokyo is a city;
There schmexists an electron;
There schmexists a quark.

(There schmexists an F, recall, iff the property of being an F is expressed by

some predicate in some sentence of this book.) On my account, we can
explain why this is so: the predicates ‘electron’ and ‘quark’, the quantifier
‘there is’, and the names ‘e1’ and ‘e2’ carve at the joints, whereas the
predicate ‘city’ and the variable-binding expression ‘schmexists’ do not. My
theory of fundamental truth is unified by these claims. On the Finean view,
on the other hand, the patterns remain unexplained; they are brute patterns in
the propositions that are real.

8.3.2 Combinatorialism about fundamentality

The propositional/subpropositional difference between the second Finean
view and my own thus leads to a difference in explanatory power. As we will
see in the next three sections, it also leads to differences in flexibility.
My subpropositional, atomistic account implies a sort of combinatorial
principle of fundamental truth:

If S is a fundamental truth, and S′ is any true sentence containing no

expressions other than those occurring in S, then S′ is a fundamental truth as
well.

A fundamental truth, for me, is a structural truth; and a structural truth is just
a true sentence composed only of joint-carving expressions; so any true
sentence composed only of expressions drawn from some structural truth
must itself be a structural truth. The combinatorialism is due to the atomism.
But Fine’s holistic account does not imply the combinatorial principle; Fine
need not accept that if In reality, S is true and S′ is a true sentence
containing no expressions other than those occurring in S, then In reality, S′
 must be true as well.
Are there reasons to deny the combinatorial principle? One might, for
example, want to hold that some sentence ~ϕ is fundamentally true but that
~~~ϕ is nonfundamental (though true).12 Now, this particular example
doesn’t worry me much, since its force would seem to derive from
considerations of ground: the intuitive reason for saying that ~~~ϕ is
nonfundamental is presumably that it is grounded, in a fundamental sense, in
~ϕ. So if I am right that we should reject a fundamental notion of ground, we
have no reason to accept the example. The claim that ~ϕ grounds ~~~ϕ in
some nonfundamental sense of ‘ground’ would not justify the claim that
~~~ϕ is nonfundamental; the “covering law” account of metaphysical
causation/ground mentioned above, for example, would make it
unremarkable for some fundamental truths to ground others.
But there are quite different, and systematic, reasons one might have for
denying the combinatorial principle. Consider, for example, the following
version of “ontological semi-realism”:13
Some ontological facts are fundamental and others are not. Facts about the existence of subatomic
particles (such as electrons and quarks) are fundamental. But there is no fundamental fact, one way or
the other, about whether there exist further things, things like composite material objects, holes,
numbers, and so on.
This semi-realist will presumably grant that the following is a fundamental
truth:
(I) ~∃x(electron(x) ∃ quark(x)) Nothing is both an electron and a quark.
But then, given the combinatorial principle for fundamentality, each of the
following is fundamentally true if true at all:
(2a) ∃x(~electron(x) ∧ ~quark(x)) There is something that is neither an
electron nor a quark.
(2b) ∃∧x(~electron(x) ∧~quark(x)) Nothing is neither an electron nor a
quark.
For both (2a) and (2b) contain only expressions occurring in (1): variables,
∃∧’, ‘~’, ‘∃’, ‘electron’, and ‘quark’. Thus whether there exists anything
other than electrons and quarks is a fundamental matter. If we indulge in the
harmless assumption that electrons and quarks are the only sorts of subatomic
particles, this is precisely what our ontological semi-realist wanted to deny.
The conflict with semi-realism might be regarded as unwanted flexibility
in my subpropositional approach. I view it instead as a powerful argument
against semi-realism!
There are ways to square semi-realism with the combinatorial principle.
The semi-realist could distinguish two quantifiers: i) a quantifier over
subatomic particles, ∃part; and ii) an ordinary quantifier. When ‘∃’ is
replaced with ‘∃part’, (I) is a fundamental truth. Accordingly, any true
statement containing just variables, ‘3 ’, ‘~’, ‘A’, ‘electron’, and ‘quark’ is
also fundamentally true. So when ‘∃’ is replaced with ‘∃part ’, whichever of
(2a) and (2b) is true, is fundamentally true. In particular, the semi-realist
could claim, it is (2b) that is true—only electrons and quarks exist in the
sense of ∃part But it is the ordinary quantifier, not ∃part, that the semi-realist
has in mind when she says that there is no fundamental fact, one way or the
other, as to whether there are any things other than subatomic particles. When
‘∃’ is replaced with the ordinary existential quantifier, the semi-realist will
say, (I) and (2a) are true but not fundamentally true, whereas (2b) is simply
false.
But this ontological semi-realist faces a challenge: to remain distinct from
an ontological realist (see chapter 9) who accepts mereological nihilism—the
denial of mereologically complex entities. Cian Dorr (2002; 2005), for
example, accepts a fundamental sort of quantification, which he uses to state
his nihilism; but he grants that ‘There are composite entities’ is true in
ordinary English. For Dorr as for the purported semi-realist, (1) and (2b) are
true in the fundamental sense of ‘∃’, whereas under the ordinary English
sense, (1) and (2a) are true but not fundamentally true and (2b) is false.
There are ways to meet the challenge. The semi-realist might say that
there are several joint-carving quantifiers, ∃1… ∃n.14 Under each quantifier,
there are electrons and quarks (that is, for each i, ∃ix electron(x) ∃∃ix
quark(x) is true). Under ∃1, only electrons and quarks exist; under the
others, various other objects exist in addition (so to speak). ∃part is ∃1; the
ordinary existential quantifier is semantically indeterminate over two or more
of the ∃is. And it is the ordinary “there is” under which “There is no
fundamental fact of the matter, one way or the other, whether there are
anything other than electrons and quarks” is true. This is a consistent
position, and is distinct from Dorr’s position; but the semi-realist may feel at
this point that he has drifted far from where he began.
The general moral is that the combinatorial principle rules out certain
views about fundamentality: views on which the fundamental level is
distinguished, not by a distinctive vocabulary, but rather by a distinctive set
of claims in a more inclusive vocabulary. One is always free to make such a
view combinatorialism-friendly by “splitting” the inclusive vocabulary.
Whenever one previously used a term T to state both a fundamental truth and
a nonfundamental truth, one now distinguishes a term Tf which occurs
exclusively within fundamental truths from a term Tn which occurs
exclusively within nonfundamental truths. But the result of doing so may not
preserve the spirit of the original view.

8.3.3 Combinatorialism about determinacy

My view implies a further combinatorial principle:

If S is a fundamental truth, and S′ is any sentence containing no expressions

other than those occurring in S, then S′ is determinate—either determi-nately
true or determinately false.

For given my definition of ‘fundamental truth’, such an S′ contains only joint-

carving expressions, and thus must be determinate given section 7.12’s claim
that there is no indeterminacy at the fundamental level.
The continuum hypothesis is sometimes said to be indeterminate. But
suppose that mundane set-theoretic truths, such as the axiom of extensionality
are fundamental. Then by the combinatorial principle, the continuum
hypothesis must be determinate, since it can be stated using only expressions
that occur in mundane set-theoretic truths (namely, logical expressions and
the predicate Є). Thus we have a surprising result: the fundamentality of
mundane truths of set-theory requires the non-mundane continuum
hypothesis to be determinate. The surprise is due to my account’s
subpropositionality. To say that a set-theoretic claim is fundamental, given
my account, is to attribute the status of joint-carving to its constituents, and
so to set-membership in particular. For the Finean, by contrast, to say that a
set-theoretic claim is fundamental is not to attribute any particular status to
set-membership. Thus the Finean has no particular reason to say that other
set-theoretic claims, such as the continuum hypothesis, are determinate.

8.3.4 No fundamental truths

We have been exploring ways in which the Finean account is more flexible
than mine. But there is also a small way in which the Finean account is less
flexible: its positive statements of fundamentality require whole propositions
to be fundamental.
Consider “quantifier variance” (chapter 9). This view is based on the
thought that objecthood, and facts about objects, are not fundamental. A
Finean might want to put this by saying that no propositions involving object-
theoretic concepts are real.15 But there may not be any non-object-involving
propositions at all (or no real ones, anyway). For, one might think, to build up
a sentence for a (real) proposition, one needs to begin with predicates,
whether applied to names or quantified variables; but predication, naming,
and quantification are all object-involving concepts. Thus, the Finean would
be forced to admit, there are no real propositions at all, in which case nothing
positive whatsoever can be said about fundamentality.
With the concept of structure, by contrast, a quantifier variantist could still
say something positive about fundamentality. She could claim that
conjunction, disjunction, negation, and necessity (say) carve at the joints.
Since structure is subpropositional, she wouldn’t need to build up to an entire
fundamental truth in order to attribute structure.
Thus the second Finean view does not combine easily with this sort of
quantifier variance. Not that this is especially damning.

8.3.5 Nihilism and deflationism

A final point about the second Finean account is that it conflates certain
nihilistic and deflationary positions.
For example, consider on the one hand a causal deflationist, who thinks
that causality is not fundamental, and on the other hand a causal nihilist, who
is a causal anti-deflationist with the additional view that, fundamentally
speaking, nothing causes anything. Each, let us suppose, thinks that everyday
statements about causation are true. How to distinguish these positions?
It is clear how my approach distinguishes them. Each thinks that ‘Some
things cause other things’ and other ordinary English sentences about
causation are true. Each thinks that the ordinary English word ‘cause’ fails to
carve at the joints. The causal deflationist thinks additionally that no causal
locution carves at the joints. The causal nihilist, on the other hand, thinks that
there is a joint-carving causal locution, ‘causes*’, in terms of which it is true
to say: “Nothing causes* anything” (recall section 5.3).
But how could the Finean distinguish the causal deflationist from the
causal nihilist? On the face of it, each agrees that:16
Some things cause other things, but it is not the case that in reality,
some things cause other things.
There are various moves the Finean might make here. I can’t conclusively
refute any of them, but none seems wholly satisfactory. The first is to say that
the nihilist is distinguished by her additional belief that in reality, nothing
causes anything. But this would involve denying that ‘in reality’ is factive. I
cannot find an explicit claim of factivity in Fine (2001), but surely that was
his intention.
The second is to make a distinction like my own between fundamental and
nonfundamental languages. (Perhaps the Finean could define a fundamental
language as a language in which In reality, ϕ is true whenever ϕ is true.) In
terms of this distinction, the Finean could say that the disagreement is over
whether ‘nothing causes anything’ is true in a fundamental language: the
causal nihilist thinks it is, and the causal deflationist thinks it is not. But this
seems to go against the spirit of Fine’s approach. That spirit is to do
fundamental metaphysics within an enhanced version of English (English
with the addition of ‘in reality’— and perhaps also ‘grounds’) rather than to
give up on English and do fundamental metaphysics in a different,
fundamental language.
The third was suggested by John Hawthorne: say that only the causal
nihilist admits that certain general laws governing causation are real, such as
the proposition that causation is transitive. The suggestion is intriguing, but it
requires that the nihilist accept some laws of causation. It would not work, for
example, if the causal nihilist denied the transitivity of causation, and further,
that any other general principle about causation is lawlike enough to merit
being real. More generally, it would not work for nihilists about other matters
who are unwilling to accept such laws.
Intuitively speaking, the problem is that deflationisms—of the sort I have
been considering, anyway—are attacks on language, whereas the Finean
ideology is resolutely “object-language” rather than metalinguistic. Faced
with a question Q?, the deflationist wants, in the first instance, neither to give
a positive nor negative answer, but rather to accuse the question of being
faulty in some way. Likewise, the anti-deflationist’s primary goal is not to
take a stand on Q’s answer, but rather to defend Q itself. Now, in Fine’s
ideology, if we don’t want to just answer Q, our remaining option is to ask
whether the proposition that Q is real. But that question doesn’t correspond to
the question of the status of Q. Since reality is factive, to answer it in the
affirmative is to answer Q in the affirmative.

8.4 Truthmaking
According to the doctrine of truthmaking, each truth has a truthmaker—an
entity whose very existence “makes-true” that truth.17
What are these truthmakers? Sometimes they can be perfectly mundane
entities. I, for example, serve as a truthmaker for the truth that there exists at
least one thing. My very existence makes it true that there exists at least one
thing. (In fact, anything at all is a truthmaker for this truth; the truthmaking
relation is not one-to-one.) But other kinds of truths require distinctive
entities as their truthmakers. I cannot serve as a truthmaker for the truth that I
am five feet nine inches tall, since I could have existed but had some other
height. (Truthmaker theorists disagree over the exact nature of the
truthmaking relation, but they all agree that the existence of the truthmaker
must necessitate the truth.) Nor is that truth made true by the property of
being five-nine, since that property could have existed without my
instantiating it. Nor is it made true by the ordered pair of me and the property,
since that ordered pair could have existed without my instantiating the
property.18 Nor does there seem to be any other particular, property, or set-
theoretic construction therefrom whose existence is sufficient for my being
five-nine. So what is the truthmaker here? When mundane entities don’t
suffice, truthmaker theorists tend to posit new entities that are fit to serve as
truthmakers—entities that simply could not have existed if the truths in
question had not been true. Different sorts of entities could be posited; I’ll
focus on “states of affairs”. (Other entities that might be posited include
tropes and “thick particulars”.) The state of affairs of my being five-nine is
the truthmaker for the truth that I am five-nine; it is to be thought of as a
contingent entity, an entity that, necessarily, exists if and only if I am five-
nine. (States of affairs thus construed are like facts on a “concrete”
conception. But I have been using the term ‘fact’ more neutrally, so I will
avoid using that term for truthmakers. Note that the truthmaker theorist’s
“concrete” states of affairs must be distinguished from necessarily existing
“abstract” states of affairs, as in Plantinga (1974).)
 In the next section I will discuss an approach to fundamentality based on
truthmaking; but first I want to distinguish, primarily in order to set them
aside, two vague ideas that are often cashed out in terms of truthmaking.
First, there is the quite plausible idea that there can be a gulf between
fundamental metaphysics and ordinary truth. According to this idea, a
metaphysician who takes the world ultimately to consist of nothing more than
atoms in the void, or of a wavefunction in statespace, or of The One, does not
conflict with common sense or science. For common sense and science do
not concern themselves with questions of fundamental reality, and the truth
of ordinary and scientific beliefs is compatible with a range of positions on
the underlying metaphysics of those beliefs. This idea is often put in terms of
truthmaking: truths of common sense and science are made true by
fundamental states of affairs about atoms or wavefunctions or Ones. The
truthmaking relation needn’t be one-to-one and isn’t tied to ordinary
meaning; thus a state of affairs specifying an appropriate configuration of
atoms can make true both the truth that those atoms are thus configured and
the truth that I am five-nine, and it can make the latter true even though the
ordinary meaning of ‘Ted is five-nine’ does not concern atoms. But
truthmaking isn’t the only way to accommodate the gulf. A Finean could say
that propositions of common sense and science are grounded in propositions
about atoms, wavefunctions, or Ones; and I could say that the metaphysical
truth-conditions of sentences of common sense and science make reference
only to atoms, wavefunctions, or Ones.19
I have a bone or two to pick with the second idea: that truthmakers are
needed in order to catch “ontological cheaters”—those ne’er-do-well
metaphysicians (such as presentists, phenomenalists, or solipsists) who refuse
to countenance a sufficiently robust conception of the fundamental (as
containing past entities, physical objects, other persons) to underwrite the
truths they accept (that there once existed dinosaurs, that there exist external
physical objects arranged thus and so, that there exist such and such people
other than themselves), and who instead state truths using suspect ideology
(tense operators, statements about possible experiences, Arthur Prior’s
(1968b) “person-tenses”). What is wrong with the presentist’s tensed claim
that there once existed dinosaurs, according to this second idea, is that the
claim has no truthmaker. The presentist rejects the existence of past
dinosaurs, and so accepts no entities whose existence would suffice for the
truth of the tensed claim—the claim “lacks a basis in reality”. But there is a
different, more direct way to object to the cheaters: simply reject their
fundamental ideologies. These include tense operators, counterfactual
constructions, and the like; and one can simply argue, on grounds of
parsimony, that ideologically leaner (though more ontologically committed)
rival views are more attractive.20 (It’s important to this assessment that
ideological commitments are coequal with ontological commitments.) This is
not intended as a “methodological” critique of the cheaters; it is a squarely
metaphysical objection. Nor is it intended to be a mechanical recipe; we must
enter the trenches and argue, on a case-by-case basis, that it’s better to posit
merely past things, external physical objects, and the like, than to introduce
tensed, counterfactual, and other ideology.
Similar remarks apply to Truthmaker 2.0, “truth supervenes on being”
(TSB), according to which any two possible worlds that are alike in what
objects exist and what properties and relations they instantiate are alike in
every way whatsoever.21 Its point is to catch the cheaters while avoiding
certain perceived problems with the truthmaker principle. As we saw, the
truthmaker principle requires the existence of distinctive entities (for
example, states of affairs) in the case of predicational truths (like the truth
that I am five-nine). And even more distinctive entities (to put it mildly) are
required for negative statements: one needs negative states of affairs (entities
whose existence somehow suffices for its not being the case that such-and-
such) or “totality” states of affairs (entities whose existence suffices for
certain entities being all the entities). TSB, on the other hand, requires no
such distinctive entities, because the truth values of negative and
predicational sentences about properties and relations do not vary between
worlds that are alike in what objects there are and what properties and
relations they instantiate. But a more direct response to the cheaters is just to
object to their ideology.
Indeed, there is a satisfyingly simple thesis about fundamental ideology
that has essentially the same upshots regarding cheaters as TSB: the correct
fundamental ideology is that of predicate logic. Like TSB, this thesis rules
out primitive tense operators and counterfactual conditionals, since the only
sentence operators in predicate logic are ‘and’, ‘or’, and so on, and the
quantifiers; and it allows predications and negative statements without
requiring the existence of states of affairs, since ‘not’ and predications are
again part of predicate logic ideology. TSB’s similarity to the simple thesis is
hidden by the way the formulation of TSB uses the ideology of predicate
logic. To say that worlds are “alike in what objects exist and what properties
and relations they instantiate” is to say that the worlds contain the same
objects (quantification and identity), and that an object has a property
(predication) in one world if and only if (propositional-logic sentential
connectives) it has that property in the other. As a consequence, true
sentences containing this ideology automatically supervene on being, and
thus pass the TSB test. But if we used other ideology to state the constraint,
this other ideology would automatically pass the test as well. For example,
sentences containing primitive tense operators would pass the test of a tensed
TSB, viz., “Worlds that are alike in which objects instantiate, or used to
instantiate, or will instantiate, which properties and relations, are alike in
every way.” TSB has, therefore, an implicit commitment to the ideology of
predicate logic; better to make this commitment explicit with the simple
thesis.22
Further, it is important to appreciate that the truthmaker principle and TSB
are not mere generic assertions that the facts must be grounded, somehow, in
fundamental reality. Still less is either claim “certain a priori”, as Lewis
claims of TSB (1994, 473). Each is a substantive and distinctive claim of
metaphysics. Consider, for example, what the two principles have to say
about primitivism about laws of nature. The truthmaker principle requires this
primitivism to take the form of the postulation of a distinctive sort of entity.23
TSB requires it to take the form either of the postulation of a distinctive sort
of entity, or of the postulation of a distinctive sort of property or relation.24
But neither allows a third form, a view that takes as primitive a sentence
operator ‘it is a law that’. On this third formulation, there are no distinctive
entities called laws, nor are there distinctive properties and relations
corresponding to laws; two worlds might contain the same distribution of
properties and relations over objects, but might nevertheless differ in that in
one, it’s a law that all Fs are Gs, whereas in the other, all Fs are indeed Gs
but it’s not a law that all Fs are Gs. This third form, in fact, seems to be the
most natural form for primitivism about laws to take; and it’s inconsistent
with both TSB and the truthmaker principle.25 Now, I don’t myself defend
primitivism about laws, but my preferred form of primitivism about structure
—based on the primitive operator described in section 6.3)—conflicts
with both TSB and the truthmaker principle for essentially the same reason
that the third form of nomic primitivism did. So it’s important for me to point
out that TSB—or better, the thesis that the correct ideology is that of
predicate logic—is a substantive thesis of metaphysics. It’s admittedly
attractive, but it can be given up if total theory demands it.

8.5 Truthmaking as a theory of fundamentality

The notion of truthmaking can be used to formulate a third rival view about
fundamentality, according to which the realm of the fundamental is
exceedingly sparse.
Fineans and I allow fundamental facts of arbitrary logical forms:
predicational, negative, quantificational, and so on. (We can defend more
restrictive views— a Finean, for example, might deny that disjunctive truths
are ever real (Fine, 2012). But our theories of fundamentality per se do not
restrict what sorts of truths can be fundamental.) The truthmaking theory of
fundamentality I have in mind, on the other hand, countenances far fewer
fundamental facts. The only fundamental facts, on this view, are certain
singular existential facts, facts of the form “x exists”, where x is a truthmaker.
(It’s an open question whether such facts should be construed
quantificationally, or as consisting of the attribution of a primitive notion of
existence. Alternatively, one could sidestep the question of the nature of these
facts and construe the theory as rejecting the idea of a fundamental fact
altogether, holding simply that the fundamental consists solely of entities.)
Predicational truths (“a is F”), negative truths (“It is not the case that ϕ”),
quantificational truths (“Everything/something is ϕ”—with the possible
exception of assertions of the existence of truthmakers) are all of them
nonfundamental. How do these nonfundamental truths fit into fundamental
reality? They are made true by truthmakers. Thus the making-true relation is
the distinctive concept for this theory. It both distinguishes the realm of the
fundamental—that realm consists of truthmakers, i.e. things that stand in the
truthmaking relation to something—and also connects the fundamental to the
nonfundamental.
It is very difficult to abide by this restrictive conception of
fundamentality. And in fact, truthmaker theorists in practice almost never
abide by it (not that they all intend to). What in fact happens is that by
making ineliminable use of certain bits of ideology, they smuggle in
fundamental facts beyond those allowed by their theory.
One smuggling route is through the canonical names given to
truthmakers, such as ‘the state of affairs of grass’s being green’. These names
are formed by means of distinctive ideology, ‘the state of affairs of ϕ’, which
is a functor that turns a sentence into a singular term naming a state of affairs.
This functor acts as a cache for smuggling unacknowledged fundamental
facts. To illustrate, David Armstrong objects to fundamental predicational
facts, claiming that it is unacceptable simply to include a predicational
sentence ‘a is F’ in one’s fundamental description of reality. To do so
without reifying universals is to be an “Ostrich nominalist” (Armstrong,
1978a, pp. 16–17); but even realists about universals are forbidden simply to
attribute universals to particulars using a predicate of instantiation; the
existence of states of affairs must instead be asserted (Armstrong, 1997,
chapter 8). Armstrong himself, however, apparently smuggles predication
into his own fundamental theory of the world, when he employs canonical
names like ‘the state of affairs of a’s instantiating U’. Armstrong provides no
metaphysical explanation of facts like:

(A) There exists a state of affairs of a’s instantiating U

and it’s hard to see what explanation could be given. So he apparently

accepts that facts like (A) are fundamental. But such facts involve
predication, since ‘instantiates’ is a two-place predicate. So Armstrong
accepts fundamental facts involving predication after all. (Nominalist
truthmaker theorists who reject fundamental predicational facts smuggle in a
similar way when they canonically name their states of affairs thus: ‘the state
of affairs of a’s being F’.) Armstrong might grant a special exemption to (A)
because the predication in (A) occurs within the scope of ‘the state of affairs
of ϕ’; but it’s hard to see why this exemption would be justified.
Alternatively, he might say that his use of canonical names is not part of his
fundamental theory; but in fact I suspect that the canonical names are
indispensable (see below). At any rate, if a truthmaker theorist allows
fundamental predication-involving facts like (A) in this way, then his
conception of the fundamental draws closer to that of the Finean and my
own, since we too allow fundamental predication-involving facts.
A similar smuggling route is employed by those who reject the existence
of fundamental facts involving negation or universal quantification, and in
their place posit negative or “totality” states of affairs. These smugglers tend
to make ineliminable use of canonical names like “the state of affairs of not-ψ
” or “the state of affairs of a1 … an’s being all the objects”, which suggests
that facts like “There exists a state of affairs of not-ψ” or “There exists a state
of affairs of a1 … an’s being all the objects” are fundamental facts, thus
violating their own view.
Truthmaker theorists may respond that these facts are not fundamental
because the canonical names are eliminable: the canonical name possessed by
x is determined by the array of truths that x makes true. x counts as the state
of affairs of a ’s being F, rather than the state of affairs of not-ψ or the state
of affairs a1… an’s being all the objects, because x makes it true that a is F,
and does not make it true that not-ψ or that a1… an are all the objects.
But this just moves the smuggled cargo. The sorts of truthmaking claims
just cited involve predication, negation, and quantification:
(T1) x makes-true thata is F.
(T2) y makes-true that not-ψ.
(T3) z makes-true that a1… an are all the objects.
So if they express fundamental facts then they too violate the truthmaker
theorist’s standards. And there is now particularly strong pressure to say that
truths like (T1)–(T3) are indeed fundamental. The only strategy for
reductively defining ‘makes-true’ would have been to somehow make use of
canonical names. A first attempt might have been:
x makes true that ψ =df for some ψ that entails ϕ, x = the state of affairs that
ψ is true.
This strategy may be doomed, if ‘entails’ is too restrictive and there is
nothing to put in its place. But it seems like the only possible strategy. And it
is unavailable here since it uses the canonical names to define ‘makes-true’;
the goal was to use ‘makes-true’ to define the canonical names.
One way out remains for the truthmaker theorist. She might deny that
(T1)–(T3) are fundamental facts, even in the absence of any way to explain
their truth in terms of other facts, simply by insisting that they have
truthmakers, and that having a truthmaker is sufficient for not being a
fundamental fact. Now, the truthmakers of (T1)–(T3)—call them t1, t2, and t3,
respectively—raise questions like those above. They have canonical names
(t1, for example, might be the state of affairs of x’s making-true that a is F);
they enter into relations of truthmaking (t1 makes (T1) true); and there is
pressure to say that at least one of these two facts is fundamental. But this
charge the truthmaker theorist will again stubbornly deny: each of the two
facts has some further truthmaker and hence is not fundamental. (One could
then raise similar questions about these further truthmakers, and receive a
similar response.)
Thus the truthmaker theorist can simply dig in, and cite further
truthmakers whenever she appears to require fundamental facts involving
ideology she is trying to avoid (predication, negation, quantification). While
this entrenchment may be a stable position, it is very explanatorily
unsatisfying. To bring this out, consider how it would look as offered by a
monistic truthmaker theorist, who accepts only a singular truthmaker: the
Cosmos.26 According to this monist, fundamental reality consists merely of
the existence of a single thing, the Cosmos, X, so that the only fundamental
fact is a fact asserting the existence of X. Now, suppose we ask her to give us
ultimate explanations of various matters. We want her to explain to us, in
terms that require no further explanation, why there are cities, why no one
can eat candy without smiling, and why electrons repel one another. To such
questions she will always give the same answer: because of X. But this is
manifestly unexplanatory. You cannot give satisfying explanations without
citing detailed general laws or patterns or mechanisms. But no detailed
general laws or patterns or mechanisms can be given by someone whose
fundamental conception of reality is so unstructured, consisting solely of the
existence of a single entity.27
The monistic entrencher can give more satisfying explanations by staying
away from the absolutely fundamental level—by appealing to facts other than
X’s existence. She can say that no one eats candy without smiling because
people smile when they’re happy and candy makes people happy; she can say
why candy makes people happy by citing their taste buds and candy’s
chemistry; and she can explain taste buds and chemistry in terms of physics.
But these explanations are not ultimate, since their explananda can be further
explained by citing X. And the monist’s ultimate explanations always consist
simply of citing X. There’s something wrong with a story about fundamental
reality that precludes satisfying ultimate explanation.
It’s tempting to respond that the monist’s ultimate explanations are
satisfying after all because their explanans, X, is a very complex entity, a state
of affairs specifying the entire configuration of the world. But this talk of
what X “specifies” is another instance of smuggling via canonical names. X
“specifies” that the world is configured in a certain way—that, say, a1 is F
and a2 is G and a3 Rs a4 and …—only in the sense that X is the state of
affairs of a1’s being F and a2’s being G and a3’s R-ing a4 and …; and this
latter fact is not fundamental, but rather has a further explanation: namely, X.
So the complexity of X, as captured in its canonical name, cannot be cited in
ultimate explanations.
The position of the nonmonistic entrencher is unsatisfying for the same
reason, albeit not quite so unsatisfying. Ultimate explanations always
terminate in the citation of entities; but since a mere list of entities is so
unstructured, these “explanations” cannot be systematized with detailed
general laws, patterns, or mechanisms.
Think of the point this way. Suppose God hands you a collection of
entities: Alexander, Buffy, Cordelia, Dawn…, and asks you to work out the
rest. Are there helium molecules? Are there cities? Why did Buffy the
Vampire Slayer end after season seven? You wouldn’t have any idea how to
respond. (The problem is most acute given monism; the collection then has
just one member. What are you supposed to do with that?) But according to
the entrenched truthmaker theorist, the fundamental facts consist just of facts
citing the existence of entities. It’s hard to see how all the complexity we
experience could possibly be explained from that sparse basis.
(A symptom of this sparse basis is a certain lack of discipline in the
practice of citing truthmakers, which we might call the “truthmaker free-for-
all”. A paradigm case is Armstrong’s discussion of modality in section 10.1
of A World of States of Affairs. There Armstrong tries to show that
irreducibly modal states of affairs need not be recognized since modal truths
can be given nonmodal truthmakers. The truthmaker for ‘Necessarily,
anything that is 1 kg mass has a part that is 1 pound’ is, he says, the fusion of
the universals 1 kg mass and 1 pound mass. The truthmaker for “It is possible
that aR b and also possible that bRa” is said to be the fusion a +R + b. Other
modal truths are similarly accorded truthmakers that are simply fusions of the
involved entities. The problem isn’t that Armstrong’s claims are wrong. It’s
rather that they are manifestly unexplanatory. The giving of truthmakers is
the truthmaker theorist’s proposed form of metaphysical explanation, but the
entities Armstrong cites clearly do not help to explain modality.28 This is no
accident; it is a natural result of the fact that truthmaking is an unsatisfactory
tool for metaphysical explanation. It is entities that make-true, and entities are
too unstructured to explain. To mention just one other instance of the free-
for-all: in response to the question “What makes it true that a makes S
true?”—this is the analog, for truthmaker theorists, of the question “What
grounds facts about grounding?”—one often receives the unenlightening
answer “a”!)
Unlike the entrenching truthmaker theorist, Fineans and I can give
satisfying ultimate explanations. For we accept structured and plentiful
fundamental truths, and can tell detailed stories about how they ground (Fine)
or are metaphysical truth-conditions for (me) various nonfundamental truths.
In sum, the truthmaker theorist is caught in a dilemma. Either she accepts
fundamental facts beyond those that merely cite the existence of truthmakers,
thus giving up on her theory, or she entrenches (by simply citing further
truthmakers when faced with certain explanatory demands), thus leading to
an unexplanatory theory.
One final point. Recall the two caches for smuggling fundamental facts:
the ideology for giving canonical names, and the ideology of truthmaking
itself. With each piece of ideology, the truthmaker theorist is going to want to
make claims that involve nonfundamental notions, claims like these:29
x is the state of affairs of there existing a city.
x makes-true that there is a city.
Now, given purity, claims like these cannot express fundamental facts.
(Otherwise, telling the fundamental story of the world would require bringing
in the notion of a city.) But truthmaker theorists regularly make claims like
these without giving any hint of how they might be reductively explained.

8.6 Schaffer: entity-grounding

Another rival account of fundamentality is that of Jonathan Schaffer (2009a).
Schaffer’s central concept is a relation of entity-grounding. Unlike Finean
grounding, Schaffer’s grounding relation relates entities of arbitrary sort.
Thus we may speak of concrete particulars, properties, states of affairs,
tropes, facts, and other entities grounding and being grounded. When x
grounds y, x is “prior to” y; y “ontologically depends on” x; y is a “derivative
entity”. Schaffer’s picture is of reality as an ordered structure, a “hierarchy of
being”:
The entities at the bottom of this hierarchy—those entities that are not
grounded by anything—are the fundamental entities, or substances; all else
derives from them.30 On this conception of fundamentality, the grounding
relation is the distinctive concept for talking about fundamentality, the
fundamental facts consist solely of facts asserting the existence of substances
(though see below), and it is the grounding relation that connects the
nonfundamental to the fundamental.
Schaffer tells us how entities are grounded. But a theory of fundamentality
must also address questions like the following: What, in the realm of the
fundamental, makes it the case that a is F? In virtue of what is it the case that
not-ψ? What grounds the fact that a1… an are all the objects? We do not only,
or even primarily, want to know why x; we want to know why ϕ, where ϕ is a
sentence. (We might put this point by saying that Schaffer addresses only
entity-grounding, not propositional grounding, but this is misleading—see
below.)
Finean grounding is built for such questions; with it we can say “the
proposition that X grounds the proposition that a is F”. Truthmaking is also
built for such questions: “x makes it true that a is F”. (As we saw, it is
doubtful whether a mere entity x is a suitable explanans; but the question here
concerns the explanandum.) I can answer the questions as well; I can supply
a metaphysical truth-condition for ‘a is F’. On the face of it, however,
Schaffer cannot answer these questions, since his grounding relation relates
only entities.
Schaffer’s grounding relation applies to all entities. So in particular it
applies to propositions. So, it might be thought, Finean ground is just a
special case of Schaffer’s. But this isn’t right; Schaffer and Fine’s grounding
relations have different significance when applied to propositions. Schaffer’s
grounding relation essentially involves its relata as entities. He characterizes
grounding as “the metaphysical notion on which one entity depends on
another for its nature and existence” (2010a, p. 345). Thus when proposition
p Schaffer-grounds proposition q, the existence and nature of q are due to p.
Finean grounding, on the other hand, is not particularly focused on its
propositional relata as entities, but rather on their subject matters. (Indeed,
Fine is happy with a sentence-operator regimentation, “ϕ because υ”, which
dispenses with relata altogether.) If the proposition that snow is white Fine-
grounds the proposition that either snow is white or snow is blue, this is a fact
about snow and its color, not about the existence and nature of propositions.
(For Fine, the ground of the existence of a proposition, p, would be a
proposition that grounds the proposition that p exists; and the ground of p’s
nature would be a proposition that grounds the proposition that p has such-
and-such features.)
So on the face of it, Schaffer cannot answer the question of what makes a
be F by listing the entity (or entities) that ground the proposition that a is F.
That would explain the existence and nature of that proposition, a certain
abstract entity. It’s hard to say what would ground the existence and nature of
an abstract entity like a proposition, since we know so little about the natures
of propositions (what is important to us about propositions is their theoretical
role, not their innards). Perhaps the proposition’s constituents? Perhaps
nothing? But whatever the answer, it is surely distant from what we were
asking about, which was a, and why it’s F.
I can think of two ways to respond.31 The first is, I take it, not Schaffer’s
own response. It is to become more like Fine, by recognizing a distinctive
sentential/propositional grounding relation. The most natural way to do this
would be to invoke, in addition to entity-grounding, a further relation of
proposition-grounding like Fine’s, which is understood to concern the
content, not the existence and nature, of propositions.32 In that case my
objections to the first Finean theory of section 8.2 apply. Further, once
proposition-grounding is introduced, entity-grounding would seem to be
superfluous: instead of speaking of the entityground of x, we could speak
instead of the proposition-grounds of propositions about x’s existence and
nature.
The second way is, I believe, Schaffer’s preferred way. It is to become
more like the truthmaker theorist, by sticking with entity-grounding but
invoking states of affairs (or other suitable entities) to bridge the gap between
the sentential explanatory questions we want to ask and the explanations of
entities that his grounding relation delivers. Where x is the state of affairs of
a’s being F, the existence of x is tightly bound to a’s being F in at least this
sense: it’s necessarily true that x exists if and only if a is F. So, Schaffer
could say, explaining why x exists (and has the nature that it in fact has)
explains why a is F.
If Schaffer does take the second way, he faces some of the same questions
as the truthmaker theorist. What do ultimate explanations look like? Do they
always terminate in the mere citation of entities? If so then his position is as
explanatorily unsatisfying as the entrenching truthmaker theorist. Or can they
also cite fundamental facts involving canonical names, such as the fact that x
is the state of affairs of a’s being F? If so then the view no longer holds that
all fundamental facts consist of the existence of substances (and the facts of
fundamentality no longer consist solely of the holding of the grounding
relation over the class of entities). My guess is that Schaffer will follow the
second course. His purpose in introducing entity-grounding is expansive
rather than reductive. He’s not trying to make do with the leanest possible
tools for talking about fundamentality; rather, he’s trying to make sure that
we have enough tools to make sense of all sorts of grounding, including
entity-grounding.
It’s worth appreciating the structural role that states of affairs play, both
for Schaffer and for the truthmaker theorist. Schaffer’s grounding relation
relates entities to entities:
x grounds y.
Truthmaking is like the grounding relation on the left, but is propositional on
the right:
x makes-true that ϕ.
(And Fine’s grounding relation is propositional on both sides: ϕ because ψ.)
Mere entities are too unstructured either to serve as explanans or
explanandum in proper explanations; we need instead to appeal to entire
sentences/propositions. Thus the appeal to states of affairs, which bridge the
gap between mere entities and full propositions: given free use of canonical
names, whenever one wishes to use a full sentence ϕ in an explanation, one
can instead cite a corresponding entity: the state of affairs of ϕ. Schaffer
requires the states of affairs to bridge the gap both on the left- and the right-
hand sides of his grounding relation, whereas the truthmaker theorist needs
them only for the left-hand side of the truthmaking relation. (Fine needs them
on neither side.)
One final point. Like the truthmaker theorist, Schaffer needs to appeal to
facts about states of affairs, phrased using canonical names, that involve
nonfundamental concepts. For example:
x is the state of affairs of there existing a city.
But again, purity33 disallows these facts from being fundamental. So like the
truthmaker theorist, if Schaffer wants to respect purity, he faces an
explanatory burden: to explain such facts. And this burden cannot be
discharged simply by citing grounding entities—that would be a return to
entrenching.

8.7 Entity-fundamentality
Finally, I’ll briefly discuss a rival approach that has not (as far as I know)
been defended in print, but which seems to underlie much casual talk of
fundamentality. On this approach, the distinctive concept is that of a
fundamental entity. Fundamental entities are like Schaffer’s substances: they
are the entities on which all else is (in some sense) based. Electrons, perhaps,
are fundamental entities, whereas tables and chairs are nonfundamental.
The approach is usually extended to cover the fundamentality of
properties and other abstract entities. The property grue, for example, is said
to be a non-fundamental entity, whereas unit negative charge is a
fundamental entity.
It would be easy to confuse this approach with my own. Both approaches,
it might be thought, can allow talk of the fundamentality of entities of
arbitrary sort: properties, quantifier meanings, and so on. But my approach is
very different. My official notion of structure (“ ”) is an operator, not a
predicate. To be sure, I often speak casually of properties, words, and “bits of
ideology” as carving at the joints, but this is loose talk. Moreover, there are
several reasons to reject this approach to fundamentality, and they do not
apply to my account.
First, extending the account to properties requires thinking of abstract
entities as the locus of fundamentality. This is especially problematic if we
want to go beyond the predicate; recall section 6.3.
Second, once the approach is extended to abstract entities, it mixes
together two things that are best separated: the fundamentality of an abstract
entity “as an entity” and the fundamentality of what it “represents”; recall
section 8.6.
Third, even when applied just to concrete entities, the approach mixes
together two further things that ought to be kept separate: the fundamentality
of an entity’s existence and the fundamentality of its nature. To illustrate:
what is the status of tables and chairs and other mereologically complex
entities according to a mereological universalist like David Lewis (1986b,
212–13; 1991, 79–81)? As I see it, it is the following. The existence of tables
and chairs is just as fundamental as the existence of electrons (in contrast,
perhaps, with smirks and shadows, which do not exist fundamentally).
However, tables and chairs have nonfundamental natures.34 These claims are
easily distinguished on my own approach: in the fundamental sense of the
quantifier ‘there are’, there are tables and chairs as well as electrons (whereas
‘There are smirks and shadows’ is true only under a nonfundamental sense of
‘there are’); but unlike electrons, tables and chairs satisfy no monadic joint-
carving predicates. It is unclear how to make these claims if we speak only of
entity-fundamentality.
Fourth, the entity-fundamentalist seems to conceive of the fundamental as
consisting solely of a collection of entities; but a theme of the preceding
sections has been that mere entities cannot generate satisfying explanations.
If the entity-fundamentalist responds in certain ways (such as invoking states
of affairs), my objections to the truthmaker theorist and to Schaffer will
apply.

1 Fine (2001, p. 15–16). I’ll ignore certain complications Fine discusses, such as multiple
propositions on the “left-hand side” (“p1, p2,… grounds q”) and the nonfactive sense of ground.
2 Fine (2001, p. 26).
3 He does say that “there is a primitive metaphysical concept of reality, one that cannot be
understood in fundamentally different terms” (p. 1). But “primitive” here could have a conceptual or
methodological sense (note the words ‘concept’ and ‘understood’).
4 This definition uses the predicate-of-propositions formulation. In the operator formulation, we
might define “basically ϕ” as “for no sentence ψ is ϕ because ψ true”, or, alternatively, as “∼∃P(ϕ
because P)”. Neither is ideal. The first renders ‘Basically ϕ’ too weak—it could be true if ϕ had a
ground that our language cannot express; the second requires quantification into sentence position.
5 Fine’s proposition b) (Fine, 2001, p. 17) commits him to this conclusion.
6 For another interesting objection to the grounding approach to fundamentality, see deRosset
(2010).
7 This move is formally available because the Finean needn’t accept the combinatorial principle of
section 8.3.
8 See also Cameron (2008).
9 See Bricker (1993); Maudlin (1993, p. 196).
 10 A priority monist like Schaffer (2010c) can escape by denying that a path has its length because
of the lengths of its parts. A friend of entity-grounding might try to escape by grounding facts about
path length in the plurality of points in the path, rather than in any fact or facts about distance. This
strikes me as an illustration of the evils of entity-grounding; see sections 8.5 and 8.6.
11 But the objection from infinite descent still threatens, insofar as real propositions are to be
intuitively thought of as ungrounded.
12 Thanks to Kit Fine for this example.
13 Chalmers (2009) is sympathetic to this sort of view.
14 See McDaniel (2009); Turner (2010b) on “ontological pluralism”.
15 The view might be put differently since the Finean view does not imply the combinatorial
principle; my point is just that the Finean might put it this way.
16 Let us assume that the causal deflationist and the causal nihilist are both ontological realists, so
the quantificational language here is not a distraction.
17 See Armstrong (1997; 2004); Cameron (2010; 2011).
18 Whatever one thinks of truthmakers, it should not be thought that they are required to
“distinguish facts from mere lists”—to distinguish i) the scenario in which I instantiate the property of
being five-nine from ii) the scenario in which I and the property both exist but I don’t instantiate the
property. Without truthmakers, we cannot say that the scenarios contain different entities. But we can
say simply that I instantiate the property in i) but not ii). Instead of appealing to an ontological
difference, we state the difference with a piece of primitive ideology (‘instantiates’).
19 See also Horgan and Potrč (2000) on indirect correspondence.
20 Sider (2001b, chapter 2, section 3) argued for a general prohibition of ‘hypothetical’ fundamental
ideology. The present approach seems superior.
21 The doctrine is from Bigelow (1988, 130–3) and Lewis (1992, 215–19). For other criticisms see
Merricks (2007, chapter 4); Sider (2001b, chapter 2, section 3).
22 Another problem with TSB is that it is modal. As is often the case, unwanted consequences result
from trying to say indirectly, via modality, what would be better said directly, in terms of
fundamentality In this case, the problem is that statements about necessary subject matters trivially
supervene on being. Anyone willing to say that mathematical claims are noncontingent can, for
example, deny that the fundamental level contains anything mathematical, but can nevertheless say that
mathematical statements are true, without violating TSB.
23 Tim Maudlin seems to accept this view when he calls laws “fundamental entities in our
ontology” (2007b, p. 18), but I suspect he cares more about antireductionism than reifying laws.
24 Armstrong (1983) and Michael Tooley (1987), for example, hold views of the latter type: laws
arise from the holding of a relation between universals.
25 Note also that all three formulations seem on a par, insofar as cheater-catching is concerned. The
thought that it’s illegitimate for presentists to take tense seriously and illegitimate for phenomenalists to
speak of counterfactuals about experiences doesn’t lead to finding the third formulation any more
repugnant than the first two.
26 Compare Schaffer (2010b), except that Schaffer is not (I take it) entrenching.
27 This is not to object to monism per se; a monist might give satisfying ultimate explanations that
terminate in facts about the Cosmos, so long as there is a sufficiently rich set of facts about the Cosmos
(for example, higher-order facts about the properties instantiated by the Cosmos, as discussed in Sider
(2008a)).
28 To be fair, Armstrong says a lot more about modality, in the same book and also in his earlier
book A Combinatorial Theory of Possibility. But what he says is more explanatory when he he attempts
to analyze modality, and not just cite truthmakers.
29 Here I treat ‘makes-true’ as taking a sentence on its right-hand side. If it were taken instead as a
two-place predicate ‘x makes proposition p’ true, the purity problem would be relocated to facts
involving proposition-theoretic ideology such as ‘the proposition that ϕ’.
30 Schaffer thinks that there is but a single fundamental entity (2009b; 2010c), but this monistic
view is detachable from his theory of grounding as such.
31 A third way would be to say that when its arguments are propositions, grounding concerns their
subject matters rather than their existence and nature. But this would implausibly fragment the relation.
32 If the notion takes the form ‘entity x grounds proposition p’ rather than ‘proposition q grounds
proposition p’, then this would move him closer to truthmaker theory rather than to Fine.
33 Side point: purity is the claim that all notions involved in fundamental facts are fundamental
notions. Schaffer’s ideology allows a related claim of “substance-purity” to be formulated: all entities
involved in fundamental facts are substances. Is there pressure on him to accept it?
34 Notice that substance-purity (note 33) is hostile to this sort of mereological universalism. For
surely this universalist will want to say that singular mereological facts are fundamental; but when x is
part of y, surely x and y are not both substances. My own framework, by contrast, is friendly to the
position. The ideology in mereological claims can be held to carve at the joints; and since my
framework allows no notion of substance, substance-purity cannot be formulated.
9 Ontology
“Is there evidence for the existence of black holes? Indeed there is.”—Roger
Penrose
“I turned to speak to God, About the world’s despair; But to make bad
matters worse, I found God wasn’t there.”—Robert Frost
“No definite evidence is yet available to confirm or disprove the actual
existence of unidentified flying objects as new and unknown types of
aircraft.”—United States Air Force
“The Cosmos is all that is or ever was or ever will be.”—Carl Sagan1

Ordinary ontology is no more remarkable than wondering about the

weather. We ask whether there is ice-cream in the freezer, whether there is a
twenty for a cab, whether there is a game on television. In more expansive
moments we ponder the existence of black holes, gods, UFOs, or anything at
all beyond the world of space and time. We understand these questions, and
we know, more or less, how to go about answering them.
Extraordinary, or philosophical, ontology, on the other hand, is
perplexing. Consider the question of whether holes exist. Not black holes.
Holes. Like in Swiss cheese. “Of course holes exist”, one is inclined to
respond; “I have one in my sock right now.” “Well,” the philosopher says; “I
see the sock. And I see that it ‘has a hole in it’, so to speak. But why think
that, in addition to the sock, there really exists another entity, the hole? What
sort of strange entity would that be? Why not think instead that only the
perforated sock exists?”
The question of holes is perplexing partly because it’s unclear how to
answer it. It’s hard to see how to tell by experiment or the senses whether the
hole exists, in addition to the perforated sock. Then again, the same can be
said for all philosophical questions. Part of what makes them philosophical is
that they strain our usual methods of inquiry. The most promising thought
about methodology in ontology, to my mind, is Quine’s (1948) exhortation to
believe locally by thinking globally—to evaluate local ontological claims by
assessing global merits of theories that contain them. These global merits
include such factors as “simplicity”, as well as fit with and prediction of our
evidence. We should believe in the existence of mathematical entities, says
Quine, because our best theories include mathematical physics, which says
(assumes) that there exist such entities as real numbers and functions over
real numbers. Others reject this claim while accepting the Quinean dialectic.
Hartry Field, for example, agrees that if our best theories said that there are
numbers, that would be a good reason to accept numbers; but he formulates a
nominalist version of physics which he claims to be superior to the usual
platonistic version (1980). One finds this methodology, in one form or
another, throughout contemporary writing on ontology.
But our concern here is a different respect in which the question is
perplexing. It can seem unclear just what is at stake. When philosophers ask
whether there exist holes in addition to perforated objects, just what are they
asking? In ordinary circumstances no one hesitates to say “There is a hole in
that sock” if the sock is perforated. By ordinary standards, the sentence
would seem to be true. Yet some philosophers think that it is false. And many
others think that it isn’t obviously true; they think that it’s an open question
whether a perforated sock contains a hole. These philosophers, apparently,
understand the question of whether there exist holes in some extraordinary
sense. But what does it mean to say that there exists a hole, in this
extraordinary sense? What does it take for a sock to contain a hole, over and
above being perforated? The existence of the hole makes no clear difference
to reality. The question itself, rather than our methods for answering it, is
unclear. Or so say many.

9.1 Ontological deflationism

Failing to see what is at stake in philosophical ontology, some are led to
ontological deflationism.2 According to this view, the philosopher’s question
of whether holes exist is confused, because the extraordinary, philosophical
sense of the existence of a hole has not been, and cannot be, clearly specified.
The only coherent question is the more mundane one of what exists
according to ordinary standards. And this mundane question can be settled by
a conceptual analysis of ordinary standards; there is no need to resort to the
Quinean methodology. Likewise for the other questions of philosophical
ontology—questions about the existence of numbers, propositions, events,
past and future objects, tables and chairs, and so on. As the leading
ontological deflationist, Eli Hirsch, puts it:3
… many familiar questions about the ontology of physical objects are merely verbal. Nothing is
substantively at stake in these questions beyond the correct use of language. A derivative claim is that,
since they are verbal, the proper way to resolve these questions is by appealing to common sense or
ordinary language.
The most viable form of ontological deflationism, in my view, holds that
ontological questions are not substantive, in the sense of chapter 4. Their
answers turn on which of the equally nonfundamental candidate meanings we
adopt for their crucial terms. (We’ll discuss other forms in section 9.7.)
According to one subspecies of this sort of ontological deflationism, the
crucial terms in ontological questions are semantically defective.4 Since the
ontologists do not judge sentences like ‘There are holes’ by ordinary
standards, they fail to use the crucial terms in their ordinary senses. They
have supplied no replacement senses; and there are no perfectly fundamental
candidate meanings supplied by the world; so no replacement meanings have
been supplied at all. The crucial terms are therefore semantically empty, or
else semantically indeterminate over a wide range of candidate meanings.
Thus ontological questions are semantically defective, and have no
determinate answers.
According to a second subspecies, the crucial terms have their ordinary
senses. The defender of this second subspecies agrees with the first that
ontologists do not judge ontological sentences by ordinary standards, and that
the world supplies no perfectly fundamental candidate meanings for crucial
ontological terms; but he holds that ontologists nevertheless use the crucial
terms in their ordinary senses—despite themselves, so to speak. After all,
surely the ontologists mean to be speaking English (or another natural
language)!5 The ordinary senses of the crucial terms are whichever candidate
meanings best fit ordinary usage. (The terms may be vague over a small
range of the candidates, but the indeterminacy is less drastic than envisioned
by the first subspecies.) Thus, ontological questions have answers, but the
answers may be ascertained simply by reflecting on the ordinary use of
ontological language.6 The Quinean methodology is as out of place for such
questions as it would be in a dispute over whether an innocent factual mistake
is a lie. It would be hopeless to argue that a globally simpler theory could be
had, if we only regarded Newton as having lied when he said that space and
time are absolute. Any competent speaker of English knows that the word
‘lie’ just doesn’t work like that, simplicity nonwithstanding.7
Both subspecies agree on the bottom line: ontological questions have
multiple equally nonfundamental candidates. Intuitively, there are lots of
things we could mean by ontological questions, none of them better than the
rest, so the only real issue is linguistic/conceptual: which of these is meant in
our language?

9.2 Ontological realism

Opposing ontological deflationism is ontological realism, according to which
ontological questions are “deep”, “about the world rather than language”. In
my view, the most viable form of ontological realism holds that ontological
questions are substantive in the sense of chapter 4.1 further think that the best
way to secure this substantivity is to hold that ontological questions can be
posed in perfectly joint-carving terms. This is the position I will defend (and
it is usually what I mean by “ontological realism”). It is the doctrine of true
believers in ontology.
Ontological realism is a claim about “metaontology”—a claim about the
nature of ontological claims and disputes. As such it is consistent with all
positions on first-order ontology. It is consistent both with the existence and
with the nonexistence of holes, with the existence and with the nonexistence
of numbers, and so on. This is not to say that it is wholly neutral on first-
order ontology. To take just one example, existence monism (according to
which only a single entity exists, the Cosmos) is a nonstarter if ontological
questions are settled by ordinary usage of ontological language. The monistic
denial of the existence of nearly every entity of common sense would then be
like the claim that Newton’s mistakes were lies. But if ontological realism is
true, monism cannot be so quickly dismissed.
Ontological realism meshes with the Quinean methodology for ontology;
ontological deflationism does not.8 Quine says to believe in the ontology of
one’s “best” theory, where one of the determinants of bestness is
“simplicity”. But as we have seen, simplicity is cheap if one is free to choose
any vocabulary one likes. That simplicity is cheap given free choice of
predicate meanings is familiar: recall again Lewis’s (1983b, 367) point that
the theory ∀xFx, where F is a predicate true only of things in possible
worlds exactly like ours, is a massively powerful and syntactically simple
theory. But a similar point holds about free choice of meanings for logical
constants. Let ∀* be an expression with the same grammar as ∀, but
interpreted so that ∀*xϕ is true iff everything satisfying ϕ is such that the
world is exactly as it actually is; in that case, “∀*x electron(x)” is again a
syntactically simple and massively powerful theory. The moral is that
Quinean simplicity is a sensible goal only for theories with appropriate
logical vocabulary. It seems reasonable to draw a further moral: simplicity is
a sensible goal only for theories with joint-carving logical vocabulary.9 Thus,
simplicity is a good guide to ontology only if ontological language carves at
the joints.10
Recent work on ontology nearly always relies on the Quinean
methodology. Nowhere is this more explicit than in Lewis’s On the Plurality
of Worlds, which argues that the best systematic theory of a range of
philosophical and linguistic phenomena requires an ontology of possible
worlds. Another pillar is Peter van Inwagen’s Material Beings, which argues
that there exist no nonliving material objects with proper parts. Although
there exist subatomic particles “arranged tablewise” and “arranged
chairwise”, as van Inwagen puts it, there do not exist any tables or chairs.
Where one billion subatomic particles are arranged tablewise, there is no
billion-and-first entity, the table; there are only the billion particles. Van
Inwagen relies more heavily than Lewis on intuitive verdicts about thought-
experiments—for example, the intuitive conviction that the following is not
the case: a composite object made up of two people is formed if and only if
the people shake hands (1990, chapter 3). But like Lewis, van Inwagen seeks
the simplest theory that accommodates the “data”, which in his case are his
intuitive judgments plus a range of theoretical presuppositions (1990,
preface). Since the Quinean methodology is appropriate only given
ontological realism, ontological realism seems to be an unacknowledged
presupposition of recent ontology.
Recent ontology presupposes ontological realism in a further way. It’s
commonly assumed that a metaphysical theory is given by its ontology and
its ideology. But this is an apt schema only given a certain substantive view:
that ontological notions are fundamental ideology. Otherwise one might
better say that a metaphysical theory is given by its set of possibilities and its
ideology, or its substances and its ideology, or some other schema, if one
thinks of modal or substance-theoretic or some other notions as being more
fundamental than ontological notions. Given ontological realism, the
ontology/ideology division is indeed natural, but this should not be assumed
from the start. (A more general schema: a metaphysical theory is given by its
ideology and its doctrine—its claims phrased in terms of its chosen
ideology.11)
Ontological realism (in my preferred form anyway) says that the crucial
terms in ontological questions carve perfectly at the joints. But what exactly
are these “crucial terms”; what is the form of fundamental ontological
assertions? Alternate views will be discussed in section 9.14, but for now I
will take fundamental ontological assertions to be quantificational:12

∃xFx
There are Fs.

The crucial expression here is the first-order existential quantifier, which,

according to this form of ontological realism, carves at the joints. If F carves
at the joints as well, the question of whether ∃xFx is guaranteed to be
substantive. Inquiry into what there is, is inquiry into the fundamental facts.
To head off a potential misunderstanding, on my conception, the
ontological question is not “Are Fs fundamental?”13 Thus conceived,
‘fundamental’ would be a predicate of entities, and ontology would be the
task of sorting entities (understood under a non-joint-carving sense of the
quantifier?) into two groups, the fundamental entities and the
nonfundamental ones. On my conception, the ontological question is rather
“Are there Fs?”, where ‘there are’ is understood as having a fundamental
sense. (I do not speak of entity-fundamentality at all; recall section 8.7.) Thus
one should not say to me: “Your fundamentality-based approach to ontology
would make it absurd to accept tables and chairs in one’s ontology, since
tables and chairs are obviously not fundamental entities.” This commits the
misunderstanding just described. On my conception, to accept an ontology of
tables and chairs is not to say that tables and chairs are “fundamental
entities”, but rather to say that there are, in the fundamental sense of ‘there
are’, tables and chairs. And saying this is not absurd. For one can say this
while conceding that the property of being a table, the property of being a
chair, and indeed, all other properties of tables and chairs, are not particularly
fundamental. Intuitively speaking, although the natures of tables and chairs
are nonfundamental, their being is perfectly fundamental.
9.3 Ontologese
There is a wrinkle. Suppose that, fundamentally, there are very few things.
Suppose with Cian Dorr (2005), for example, that there exist, in the
fundamental sense, nothing but subatomic particles.14 Given such a sparse
ontology, the most plausible view about natural language quantifiers might be
that they do not carve at the joints. The best metaphysical semantics of an
ordinary sentence like ‘There is a table’ might not be a strict semantics that
interprets it as making the false claim that there exists, in the fundamental
sense, a table, but rather a tolerant semantics, which interprets it as making
the true claim that there exist subatomic particles appropriately arranged. The
English ‘there is’, according to such a semantics, would not express
fundamental quantification. (Section 7.7 gave an example of such a
metaphysical semantics.)
Or suppose that in the fundamental sense of the quantifiers, there are no
holes. The best metaphysical semantics of ‘There is a hole in that sock’ might
not interpret ‘there is’ as expressing fundamental quantification, but might
instead count the sentence as being true if the sock in question is perforated.
So even if there is a joint-carving sort of quantification, the quantifiers of
ordinary language might not carve at the joints. We must therefore refine our
construal of ontological realism. Ontological realism should not claim that
ordinary quantifiers carve at the joints, or that disputes using ordinary
quantifiers are substantive. All that’s important is that one can introduce a
fundamental quantifier, which can then be used to pose substantive
ontological questions.15 Similarly, ontological deflationists should not hold
merely that ordinary-language ontological disputes are nonsubstantive. They
must hold, in addition, that no sense of the quantifier can be introduced on
which ontological questions would become substantive.
The question of whether ordinary quantifiers express joint-carving
quantification is a difficult one. At what point in the following series should a
metaphysical semantics start counting the sentences as false?: ‘There are
hydrogen atoms’, ‘There are dogs’, ‘There are tables’, ‘There are economies’,
‘There are events’, ‘There are smirks’, ‘There are holes’, ‘There are fictional
characters’, ‘There are gods’? To avoid getting embroiled in this question—
which is after all metase-mantic, not metaphysical—ontological realists
might conduct their ontological debates in the metaphysics room rather than
the marketplace (section 5.3). They might introduce a new language
—“Ontologese”—whose quantifiers are stipulated to carve at the joints.
Ontological questions in Ontologese are substantive, even if those in ordinary
language are not. Moreover, Ontologese is a better language, since its
structure better matches reality’s structure.
How is Ontologese to be introduced? The crucial thing is to stipulatively
remove any normal metasemantic pressure towards tolerant interpretations
that assign non-joint-carving meanings to quantifiers. Ontologese quantifiers
are to have meanings that carve at the joints, but are otherwise as similar as
possible (in inferential role, for instance, as well as in extension), and similar
enough, to the meanings of the ordinary quantifiers.
Of course, ontological deflationists will think that the attempted
introduction of Ontologese misfires, since the world lacks the necessary
structure. They might compare the attempted introduction of Ontologese to a
failed attempt to introduce ‘dirt’ as a natural kind term: “by ‘dirt’ I shall mean
the element of the periodic table in which trees grow, which forms mud when
combined with water, and which is flattened to make country roads”.
Whether the introduction of Ontologese succeeds depends on the facts, on
whether there is a joint-carving sort of quantification, just as the imagined
introduction of ‘dirt’ depends on whether the periodic table contains an
appropriate element.
The shift to Ontologese is an instance of the following schematic reply to
ontological deflationism: “While it may be absurd to deny that there’s any
sense in which—for example—there is a hole in a perforated sock, it’s not
absurd to deny that in ‘a metaphysical sense’ there exist holes.” Anyone
pursuing this schematic reply needs to characterize the alleged metaphysical
sense. I do this by appealing to a metaphysical notion, the notion of carving
at the joints. Others attempt to do this using more linguistic notions. For
example, the ontological question of whether there are holes is sometimes put
as the question of whether there strictly speaking exist holes, or whether there
literally exist holes. But I doubt these notions of linguistic theory are up to
the task. “Literal” is opposed to things like metaphor and hyperbole; “strict”
casts off things like quantifier domain restriction and loose talk (as when
people who live in Cherry Hill, New Jersey say they’re “from
Philadelphia”).16 But ‘There are holes’ is neither metaphorical nor hyperbolic
nor restricted nor loose.
In what follows I will generally continue to speak as if the
metaontological question concerns ontological disputes in English; but take it
as read that we may need to construe it instead as concerning disputes
conducted in Ontologese.

9.4 Predicates not the issue

Ontological deflationists say that questions of philosophical ontology are
nonsubstantive. These questions, let us continue to assume, have the form:
∃xFx? Now, a (near enough) sufficient condition for a question to be
substantive is for all of its terms to carve at the joints. So in the cases in
question, the ontological deflationist must argue either that the predicate F
fails to carve at the joints, or that the quantifier fails to carve at the joints. I
will argue in this section that they must make the second choice. They cannot
rely solely on claims about predicates.17
To take a concrete example, consider the dispute over the ontology of
composite material objects. Van Inwagen (1990) claims that there are no
tables; his opponents claim that there are. And the ontological deflationist
says that this is not a substantive dispute. There are multiple equally joint-
carving candidates for ontological language; under some, ‘There are tables’ is
true; under others it is false. (The sentence is false under strict candidates that
require the parts of would-be composite objects to satisfy some strong
condition—being “caught up in a life”, say—and true under lax candidates
that make no such requirement.) Now, could the ontological deflationist grant
that the quantifiers carve at the joints, but claim that these candidates are for
the predicate ‘is a table’?
It is intuitively clear that he could not. The deflationist’s thought is not
that the nonsubstantivity is merely about how to classify things—that there
substantively is a certain thing, x, which counts as a “table” under some
conceptions of tablehood but not under others. The deflationist’s thought is
rather that it is not a substantive matter whether x exists at all, never mind
how it is classified.
Distractingly, the predicate ‘table’ does indeed fail to carve at the joints,
and has multiple candidate meanings. But this noise can be eliminated. We
can simply drop the predicate ‘table’, and instead construe the dispute as
concerning whether ∃xτ(x) is true, for some appropriate τ stated in purely
fundamental terms. τ might specify the exact mass, or exact boundaries, or
exact arrangement of parts, of some particular putative table. Van Inwagen
will reject the existence of τs; his opponents will accept the existence of τs;
and the ontological deflationist will want to claim that the dispute is
nonsubstantive. Since τ carves at the joints, this requires the deflationist to
say that ∃ does not carve at the joints.
Many ontological deflationists seem to be tempted to resist this
conclusion, to blame the predicates instead, and to say things like this:
The dispute over whether “there are tables” (or “there are τs”) is nonsubstantive because its resolution
depends on whether one adopts a strict or lax definition of ‘table’. The lax definition takes ‘table’ to
mean “collection of appropriately arranged particles”, while the strict definition adds some further
requirement (being alive, say). It’s obvious that ‘There are tables’ is true under the lax definition and
false under the strict definition, so what is the dispute about?
Meta-ontology 101: never say such things. The problem is with the phrase
‘collection of appropriately arranged particles’. What does it mean? This is a
predicate, so we can rephrase: what must something, x, be like in order to
count as a “collection of appropriately arranged particles”? A natural answer
is: x must be an object containing parts that are appropriately arranged. Given
this understanding, it is not at all “obvious” that there exist tables in the lax
sense. The debate between van Inwagen and his opponents is precisely over
whether there exist things whose parts are appropriately arranged. Van
Inwagen’s claim is that there simply are no such things.
It would not help to take ‘collection of appropriately arranged particles’ as
instead meaning ‘set of appropriately arranged particles’, for this would not
support deflationism about a debate between ontologists who have stipulated
that they intend ‘table’ in a mereological, rather than set-theoretic sense. Van
Inwagen and his opponents would be happy to stipulate τ(x) to mean that x is
composed of (i.e., made up from parts that are) appropriately arranged
particles, and to recast their debate as being over whether ∃xτ(x). Indeed,
that is how they understood the debate all along. (Also, the disputants might
not accept the existence of sets.)
A further reason not to blame ‘table’ is that van Inwagen and his
opponents also disagree over how many entities there are, and such
disagreements do not concern ‘table’. Consider a world in which there exist
exactly two subatomic particles. Ofthat world, van Inwagen would reject,
while some of his opponents would accept:

—that is, “There exist at least three things.”18 (Van Inwagen may think that
subatomic particles are invariably accompanied by abstract objects—sets for
example—in which case he too would accept the displayed sentence. But we
could simply restrict the sentence’s quantifiers to concreta.) This sentence
contains only quantifiers, truth-functional connectives, and the identity sign.
Surely the deflationist will want to say that a debate over its truth is
nonsubstantive, and thus cannot accept that all of its expressions carve at the
joints. But it’s hard to see how they could blame the nonsubstantivity on the
truth-functional connectives or the identity sign (or the predicates needed to
restrict the quantifiers to concreta).

9.5 Quantifier variance

Thus ontological deflationists cannot accept that quantifiers carve at the
joints. In my view they should go further and uphold quantifier variance, to
use Hirsch’s term: the claim that there are multiple candidates to be meant by
quantifiers, none of which carve perfectly at the joints, but none of which are
exceeded in joint-carving by any other quantifier candidate. It’s a bit unfair to
commandeer this term, since Hirsch himself does not speak of joint-
carving.19 Still, the view I call “quantifier variance” is in the spirit of Hirsch’s
own view, a spirit he attributes to Hilary Putnam and describes as follows:
… the quantificational apparatus in our language and thought—such expressions as “thing”, “object”,
“something”, “(there) exists”—has a certain variability or plasticity. There is no necessity to use these
expressions in one way rather than various other ways, for the world can be correctly described using a
variety of concepts of “the existence of something”. One of [Putnam’s] favorite examples concerns a
disagreement between mereologists and anti-mereologists as to how many objects there are in some
domain. Suppose we are evaluating the truth of the sentence, “There exists something that is composed
of Clinton’s nose and the Eiffel Tower”. Mereologists will accept this sentence, whereas anti-
mereologists will reject it. Putnam’s doctrine of quantifier variance implies that the expression “there
exists something” can be interpreted in a way that makes the sentence true or in a way that makes the
sentence false. Since both interpretations are available to us, we have a choice between operating with a
concept of “the existence of something” that satisfies the mereologist or operating with a different
concept that satisfies the anti-mereologist.
Quantifier variance, on my formulation, says that “there are” many
candidates for being meant by quantifiers; but the quantifier variantist needn’t
take this quantification seriously. Granted, he could reify quantifier-meanings
(under his current quantifier-meaning!), but he needn’t. He could instead
rephrase his claims (as Hirsch does) so as not to reify the candidates; he
could say, for example, that we could choose to use the sentence “There
exists something that is composed of Clinton’s nose and the Eiffel Tower” so
that it comes out true, or we could choose to use it so that it comes out false;
and under neither choice would our words carve at the joints better than
under the other. My discussion will quantify over candidates since otherwise
it’s hard to state the view in any generality, but I won’t take such
quantification as part of the view’s metaphysical core.
How should we think about these quantifier candidates, these alternate
meanings for ‘all’ and ‘some’? At the very least, quantified sentences are
supposed to have truth-values relative to them. Hirsch’s sentence
(H) There exists something that is composed of Clinton’s nose and the
Eiffel Tower
is supposed to be true relative to one of them and false relative to another.
But can anything more be said?
One approach associates quantifier meanings with alternate languages:
languages whose sentences look just like English sentences—same grammar,
same-looking lexicon—but have different truth-conditions. The different
quantifier-meanings are the meanings that quantifiers have in these
languages. For example, following a suggestion of Cian Dorr (2005) on
behalf of the quantifier variantist, we could introduce a language LU in which
a sentence ϕ is true if and only if the following sentence of ours is true:20
If composition had been unrestricted (that is, if it had been the case
that for any x and y there exists a mereological sum of x and y), then it
would have been the case that ϕ.
If composition had been unrestricted then there would have existed
something composed of Clinton’s nose and the Eiffel Tower; so Hirsch’s
sentence (H) is true in LU. But (H) is false in a second language, LN, in which
ϕ is true if and only if the sentence that results from ϕ by restricting each
quantifier with the predicate ‘is a mereological atom’ is true in English. (The
idea isn’t that speakers of these languages think the content of ϕ via its
“translation” into English. They think natively)21
It might be objected that these languages do not really correspond to
different meanings for quantifiers. The specification of truth-conditions for
LU, for example, did not give any distinctive rule for interpreting quantifiers.
It’s hard to see how to divide the difference between English and LU into a
difference exclusively concerning quantifiers and other differences. But this
may not bother quantifier variantists. For what they really care about is
vindicated simply by the claims that: i) there are multiple possible languages
of this sort; ii) quantificational claims can have different truth-values in the
different languages; and iii) none of the languages are more joint-carving
than the rest.
Quantifier variance might at first sound trivially correct. Language is
conventional, after all, so who would deny that quantifiers could have meant
different things? Everyone agrees that we could have used the bare symbol
‘there exists’ as a sign for negation, a predicate for faculty of Harvard
University, a name for Rudolf Carnap, or even as a punctuation mark.
But remember that quantified sentences (same grammar as ours, and
same-looking lexicon) are supposed to have truth-values relative to these
meanings. So the quantifier meanings can’t correspond to taking ‘there
exists’ to be a name, predicate, or punctuation mark. Moreover, the meanings
are supposed to be candidates, in the sense of section 4.2. “Semantically
alien” meanings (relative to the actual meanings of quantifiers) are thus
disqualified. Section 4.2’s characterization of ‘candidate’ was admittedly
thin, but in the present case we can say a little more: the candidates here must
be inferentially adequate in that the core inference rules of quantification
theory must come out truth-preserving (for example, any candidate must
count ‘Something is a philosopher’ as true if it counts ‘John is a philosopher’
as true).22 And the fact that the candidates are said to be unsurpassed in joint-
carving by any other quantifier-candidate renders the claim still less trivial.
It’s not hard to concoct arbitrary and bizarre assignments of truth-values to
quantified sentences while preserving the inference rules of quantification
theory (there exist assignments of this sort for each model for the language in
question, for example). But such concocted meanings will, in general, fail
horribly to carve at the joints.

9.5.1 Quantifier variance and domain restriction

There is a different way in which quantifier variance threatens to be trivial,
however; which raises delicate issues and is not so easily answered. There is
a sort of “quantifier variance” that is wholly mundane and accepted by
everyone: quantifier domain restriction.23 For example, the sentence
“Everyone was at the party” can be used to say that everyone who works at
the office was at the party, that all the Hollywood celebrities were at the
party, or even (in admittedly extraordinary circumstances) that absolutely all
human beings were at the party. The meanings of ordinary restricted
quantifiers are reasonably joint-carving candidates, yet they are clearly not
what quantifier variantists have in mind, since otherwise they would have no
objection to substantive ontology conducted with unrestricted quantifiers.24
Quantifier variantists need to answer this threat of triviality. This section
explores two ways of doing so, corresponding to two types of quantifier
variance.
The quantifier variantist claims that some pairs of possible languages are
otherwise similar except for what their quantifiers mean. In some such pairs,
he will want to say, one quantifier is an expansion of the other. Let’s use “q1
expands q2” and “q2 contracts q1” synonymously. The intuitive idea of
expansion is that q2’s domain is a proper subset of q1’s; but there are
obstacles to regarding this as a definition. For one thing, the definition would
function as intended only when, intuitively, the domains of both q1 and q2 are
properly contained in the domain of the quantifiers that are implicit in the
definition—the quantifiers that would reveal themselves when ‘the domain of
q2 is a proper subset of the domain of q1’ is unpacked.25 But set this aside for
now; let’s just give the quantifier variantist ‘expands’ and ‘contracts’.
Question: does the quantifier variantist think of contraction as domain
restriction? When q2 contracts q1, could a speaker using q1 characterize the
meanings of existentially q2-quantified statements thus: “They mean what I
would mean by ‘there is a member of S that is …’, for some restricting set (or
class) S”? The two types of quantifier variance to be discussed differ over
how they answer this question.
The first type says: yes. Contraction is restriction. But this faces an
obstacle. The quantifier variantist might well recognize a maximal quantifier
—a quantifier of which all other quantifiers are contractions. Cannot
substantive ontology then be resuscitated—as concerning the maximal
quantifier? For this maximal quantifier seems to enjoy a privilege: it alone
ascribes unqualified being. For any other quantifier, to say that there is a ϕ in
its sense is to say that there is, in the maximal sense, some ϕ that has some
further, restricting, feature. Users of the nonmaximal languages are just
ignoring some of what there unrestrictedly is. The nonmaximal quantifier
meanings are nothing more than second-rate restrictions r1(q), r2(q)… of a
single privileged unrestricted quantifier meaning, q:
 The defender of the first type of quantifier variance replies that quantifier
meanings are not “objectively” restricted or unrestricted. Restrictedness is
relative to which quantifier one is using “as one’s unrestricted quantifier”.
The picture above is from the perspective of someone who adopts q as her
unrestricted quantifier; and from this perspective the meanings ri (q) are
indeed restricted. But one can just as well adopt one of the other meanings as
one’s unrestricted quantifier; and from this perspective the adopted meaning
is unrestricted. The notation “q” and “ri(q)”, and the boldfacing and
enlargement of ‘q’ in the picture, misleadingly depict an asymmetry between
q and the other meanings ri (q), whereas in fact the quantifier meanings are
all on a par, and are better depicted thus:

If one adopts a nonmaximal quantifier meaning, then some of the other

meanings cannot be defined as restrictions of one’s own. Does this generate
an unwanted asymmetry amongst the meanings, since some cannot be used to
define others? The defender of this type of quantifier variance will say no.
Let qe expand qc. One can adopt qe and define qc by restriction; but
alternatively, one can adopt qc and define qe in some other way. For example,
one might, following Dorr again, define qe in counterfactual terms: there
existsqe an F iff there would have existedqc an F, had composition been
unrestrictedqc.26 The fact that contracted meanings cannot define expanded
meanings by restriction does not privilege the expanded ones because
counterfactual expansion is just the converse of restriction. Just as quantifier
meanings are not objectively restricted or unrestricted, they are also not
objectively counterfactual or noncounterfactual. Indeed, it would be natural
to pair this type of quantifier variance with a very coarse-grained conception
of content.27
This first type, then, claims that contracted quantifier meanings are indeed
restrictions of expanded ones, but denies that this privileges the maximal
meaning in a way that would resuscitate substantive ontology since
restrictedness and unrestrictedness are relative to one’s adopted quantifier
meaning.
Recall the crucial question for quantifier variantists: is contraction just
domain restriction? The second type of quantifier variance answers: no. None
of the quantifier meanings distinctive of quantifier variance are restrictions of
any of the others. This type instead embraces an objective distinction
between restricted and unrestricted quantifier meanings, and claims that there
are multiple objectively unrestricted quantifier meanings q1,q2.…Each qi
can, of course, be restricted in various ways, but the results are objectively
restricted quantifier meanings r1(qi), r2(qi).…So the quantifier meanings may
be pictured as follows:

This type requires hyperintensionally individuated meanings. For some ri

(qj) might be “necessarily coextensive” with some qk; they would
nevertheless be distinct, since the former is objectively restricted and the
latter is objectively unrestricted.
The defender of this type has an easy response to the worry motivating
this section: his position differs from the recognition of mundane domain
restriction because mundane domain restriction results in objectively
restricted quantifiers whereas his qi s are objectively unrestricted.
But what does it mean to say that the qi s are objectively unrestricted? It
can’t mean that “each qi includes absolutely everything in its domain”. For
“There is no more than one domain that includes absolute everything” would
seem to be something like a logical truth (if domains are distinct then
something is included in one but not the other, in which case they can’t both
include absolutely everything); combining this with the putative definition
would imply that there is only one of the qi s. The defender of this type will,
in fact, deny that his qi s should be understood as corresponding to domains.
What, then, are the qi s? The defender might refuse to answer in reductive
terms. They are entities with respect to which sentences have truth-values,
but one cannot define any qi using any distinct qj.
Even granting this much—that the qi s do not correspond to domains and
indeed cannot be reductively defined—we still need an account of objective
unrestrictedness: what is it about each qi that makes it objectively unrestricted
and its restrictions rj(qi) objectively restricted? The quantifier variantist might
appeal to joint-carving: although no qi carves perfectly at the joints, each
carves better at the joints than do its restrictions.28 Alternatively, he might
appeal to facts about meaning identity: a quantifier meaning m is objectively
restricted if in some possible language some explicitly restricted quantifier
“Some member of S is …” (for some name S of some restricting set or class)
means m; m is objectively unrestricted if no such explicitly restricted
quantifier means m (even if some such explicitly restricted quantifier is
“necessarily coextensive” with m). Yet another alternative would be to
simply take the notion of objective restriction as primitive.
It is unclear whether either of the two types of quantifier variance can be
adequately spelled out; and there are further questions about how to
formulate quantifier variance.29 But let us rest content with what is on the
table so far, and move on to discuss objections to the doctrine.

9.6 Objections to quantifier variance

Ontological deflationism is probably the most popular form of metaphysical
deflationism. This is ironic since it is particularly hard to sustain. A dispute
involving a crucial term T is nonsubstantive only if there is “T-variance”—an
array of multiple, equally joint-carving candidates to be meant by T; and as
we will see, quantifier variance is a particularly problematic form of variance.
(There are no general problems with claims of predicate-variance, for
example).

9.6.1 The semantic argument

I begin, however, with an objection to quantifier variance that can, I believe,
be answered. The objection is due to Matti Eklund (2007; 2009) and John
Hawthorne (2006c). Consider two characters, Big and Small. Big speaks an
“expansive” language, Biglish, in which speakers freely quantify over tables.
Big introduces a name, ‘a’, for a table, and thus accepts ‘Table(a)’. Small
speaks a “smaller” language, Smallish, in which speakers refuse to quantify
over tables. But Small is a quantifier variantist, and thinks that he does not
genuinely disagree with Big. So Small says to himself, speaking in Smallish:
“Even though there are no tables, the sentence ‘Table(a)’ is true in Biglish.”
But this commits Small—and all quantifier variantists, who must accept the
scenario as described—to rejecting familiar Tarskian ideas about semantics.30
According to Tarskian semantics, for any language, L, a subject-predicate
sentence is true-in-L iff its subject term has a denotation-in-L that is a
member of the extension-in-L of its predicate. If this quantified biconditional
is true in Smallish, then in order for ‘ “Table(a)” is true in Biglish’ to be true
in Smallish, ‘There is something that “a” denotes-in-Biglish’ must be true in
Smallish. But what would this object—in the Smallish sense of “object”—be,
if not a table? So runs the Eklund-Hawthorne argument.
The quantifier variantist should, I think, reply as follows. When we vary
what the quantifiers mean, we thereby also vary the meanings of all other
expressions that are tied up with the “idea of a thing”: names, predicates,
function symbols.31 Indeed, the meanings of these categories, construed as
semantic categories, must vary. For the notions of name, predicate, function
symbol, and quantifier are interconnected. A name (in predicate logic,
anyway) is an expression that refers to a single thing; a predicate is an
expression that applies to one or more things; a function symbol is an
expression that stands for a function mapping one or more things to a further
thing. These connections to quantification are definitive of ‘name’,
‘predicate’, and ‘function symbol’, and are what underlie Tarskian ideas
about semantics. It is natural to expect, therefore, that different conceptions
of quantification would result in different conceptions of what it is to be a
name, predicate, or function symbol.32 In particular, Small should deny that
Big’s expression ‘ a ’ is a name (i.e., deny that it is a nameSmall), since ‘ a ’ in
Biglish doesn’t have the semantic function of referring to a thing.33 By
denying that ‘a’ is a name, Small can deny that ‘Table(a)’ is a subject-
predicate sentence, and so can deny that the Tarskian biconditional applies to
‘Table(a)’.
The reply so far is, I think, correct, but it doesn’t fully answer Eklund and
Hawthorne. For even if Small is right to deny that Biglish contains names or
subject-predicate sentences, it would be hard for Small to deny that Big’s use
of language is in some sense compositional. And so, shouldn’t Small say
something systematic about how Big’s sentences get their truth-conditions?
Yes; but Small need not do so inTarski’s way.Anyone can agree that some
extreme cases call for novel semantic ideas in order to make sense of alien
but compositional linguistic behavior. From Small’s point of view, the case
of Big calls for a departure from the Tarskian paradigm. One strategy Small
might pursue would be to say that Big’s sentence ⌜Fα⌝ is true iff there are
some referents (plural) of the “subject” term α that are in the “extension” (in
a plural sense) of the “predicate” F. The resulting theory might be complex
and ugly. But if a full semantics is difficult (or even impossible) to give using
Small’s language, that wouldn’t undermine quantifier variance. Granted, it
would be an asymmetry between Small and Big, for there is no corresponding
disadvantage to speaking Biglish. But quantifier variantists can admit that
bigger is better for certain purposes; all they are committed to saying is that
neither language adheres better to nature’s joints.

9.6.2 No foundation
A more promising objection presses the following questions on the quantifier
variantist: what is fundamental, if quantification is not?34 What is the world
fundamentally like? How will you write the book of the world? As we’ll see,
these questions are difficult for the quantifier variantist to answer.
Whereas the previous section’s argument was specific to quantifier
variance, this argument—as well as the arguments of the next two sections—
applies to all ontological deflationists, since all ontological deflationists must
deny that quantifiers carve at the joints.
As we saw in section 7.1, it’s natural to assume that the fundamental is
“complete”, which I spelled out as meaning that it must be possible to give a
metaphysical semantics for each nonfundamental language, using a
fundamental metalanguage. So in particular, it must be possible to give a
metaphysical semantics for the language of physics, using a fundamental
language.
Now, the ontological realist can do this in familiar ways. It’s easiest if the
ontological realist has a reasonably permissive ontology, for then he can take
the physicist’s language—including the mathematical bit—to be a
fundamental language.35 It will be harder if the ontological realist accepts a
more restrictive ontology—say, one that contains no mathematical entities.
Still, there are well-known programs for at least attempting to formulate
physical theories in nominalistic terms.
But no serious work on the foundations of physics and mathematics has
been done in a quantifier-free setting. So the quantifier variantist must begin
from scratch. He must choose some alien quantifier-free language as his
fundamental language, and then he must somehow give a metaphysical
semantics for the quantificational language of physics in its terms.
Some attempts to do this evaporate under closer inspection. Suppose, for
example, that the quantifier variantist says the following:
I have no need for objects in my fundamental description of the world.
The world fundamentally consists of the distribution of properties over
spacetime. One can then introduce the ordinary notion of an object in various
ways atop this foundation.
This is just confusion. Far from renouncing quantifiers in his fundamental
language, this variantist helps himself to quantification over points and
regions of spacetime.36 Quantification is deeply embedded in all physical
theories as normally understood, as well as in the mathematical theories they
employ. Imagine trying to state a physical theory without quantifying at all.
You couldn’t quantify over points or regions of space or time or spacetime.
You couldn’t quantify over points or regions of configuration space or phase
space or any other higher-order space. You couldn’t quantify over real
numbers, or functions of real numbers, or vectors, or tensors, or any other
mathematical entities. Your attempt wouldn’t even get off the ground.
It is sometimes said that we need no things at the fundamental level, only
stuff. To help in the present context, the suggestion must be that we can
replace quantification in a fundamental language with stuff-theoretic lingo,
and then use this to give a metaphysical semantics for physical theories.
Now, the stuff gambit tends to be paired with the idea that stuff lingo is
basically like natural language talk using mass nouns. Thus, the stuff
language would include sentences of the form “Some α is ϕ”, “All a is ϕ”, “δ
is the same α as γ”; “δ is part of γ”, and so on, for mass nouns α and ß and
“mass singular terms” δ and γ. After all, in English one can say “Some water
is polluted”, “All water is wet”, “The water on the floor is the same water as
the water that was in the tub”; and “The water in Lake Erie is part of the
water in the Great Lakes.” Thus, the stuff language contains “stuff
quantifiers”, “stuff-predication”, “stuff-identity” and “stuff-parthood”. One
then wonders whether this stuff language differs in any nonterminological
way from the thing language it is supposed to supplant.37
Resolving whether two proposed fundamental languages differ genuinely
or merely notationally can be very difficult. Indeed, it’s often hard to know
what’s at stake. But here we can bypass the issue. For even if the stuff
language differs genuinely from the thing language, it won’t scratch the
quantifier variantist’s real itch. The variantist’s reason for introducing the
stuff language was that he regarded questions about the existence of things as
being nonsubstantive, but conceded that questions in a fundamental language
are substantive—the stuff language was to be a fundamental language in
which questions about the existence of things cannot be raised. But analogous
questions about the existence of stuff can be raised in the stuff language, such
as the question of whether for any stuff and any distinct stuff, there exists
some further stuff containing them as stuff-parts. The quantifier variantist
should at this point retrace his steps, for he will be no happier admitting that
this new question is substantive than he was admitting that the question of
whether any two things are parts of some further thing is substantive.
Similar remarks apply to some other attempts to construct quantifier-free
fundamental languages. The variantist might think to use the predicate-
functor formulation of predicate logic popularized by Quine (1960b).38 In
this language, one replaces quantifiers and variables with predicate functors.
Grammatically, predicate functors turn predicates into predicates (with
perhaps different ’adicies). We can get the hang of the system by considering
the predicate functor that takes the place of the existential quantifier: c. c
turns a (perhaps complex) n + 1-place predicate Fn+1 into an n-place
predicate cFn+1, which can be thought of, intuitively, as being true of u1… un
iff for some u, Fn+1 is true of u,u1. ..un. Thus in predicate functorese one says
cF instead of ∃xFx. (We just used the existential quantifier to explain the
meaning of c in an intuitive way, but the proposal on the table is that c and
the other predicate functors are metaphysically primitive.) Other predicate
functors are needed to gain the full expressive power of predicate logic. ~ is
for negation: ~Fn is a predicate that is (can be thought of as being) true of
u1… un iff Fn is not true of u1… un. ∧ is for conjunction: Fn∧Gm is true of
u1… umax(n m) iff Fn is true of u1… un and Gm is true of u1… um. σ “rotates
argument places”: σFn is true of u1 …un iff Fn is true of un, un,u1,…un−1.l
permutes the first two argument places: lFn is true of u1 …un iff Fn is true of
u2, u1, u3… un. p “pads” a predicate by adding a vacuous argument place: pFn
is true of u1… un+1 iff Fn is true of u2…un+1. It can be shown that predicate
functorese has the expressive power of predicate logic (without individual
constants); so the variantist might regard it as a suitable replacement for
predicate logic in his fundamental language. But this would again be
misguided, since it would saddle the variantist with substantive questions that
are precisely analogous to the substantive questions he was trying to avoid.
For example, the variantist was trying to avoid substantive questions like this:
“Does there exist something containing an F and a G as parts?” But
corresponding to this question’s predicate logic symbolization:

there is the following sentence of predicate functorese:

(This is a “sentence” in the sense that it’s a zero-place predicate; it

“corresponds” in the sense that if one gives the obvious definition of truth in
a model for the language of predicate functorese, the functorese sentence is
true in exactly the same models as the predicate logic sentence. cF
corresponds in this sense to ∃xFx.) The variantist we are discussing thinks
that the second displayed sentence is couched in joint-carving terms, and so
must admit that its truth-value is a substantive matter. But this will surely be
no less objectionable to him than regarding the original, quantificational,
question as being substantive.
Here’s another thought the variantist might try. The fundamental language
does contain quantifiers, but these are “particle quantifiers”—quantifiers over
subatomic particles. The philosophical ontologist’s quantifiers, on the other
hand, are not fundamental; rather, they can be defined in terms of the particle
quantifiers in multiple, equally good ways.
The putative particle quantifiers, ∃p and ∀p, are conceded to range over
particles. But we can now raise a philosophical question: do they range over
more entities? For example, where < is a predicate for parthood, we can now
ask whether the following is true: Since ∃p and ∀p are
conceded to be fundamental, the quantifier variantist cannot invoke quantifier
variance to dismiss this question as nonsubstantive. ∃p and ∀p were called
“particle quantifiers”; but what does that mean? The claim that they range
only over particles is simply a bit of doctrine, and can be coherently
challenged. Given these quantifiers, there is nothing conceptually incoherent
about the claim that they range over more than just particles.
There is one final nonquantificational proposal I’d like to consider. It’s
just a picture, really; but it’s arguably the best intuitive fit with quantifier
variance. The picture is that of a “fact-level” metaphysics. According to this
picture, the fundamental facts are indivisible wholes, with no privileged
decomposition into subfactual components. Joint-carving notions are
somehow all at the level of the complete sentence, the entire fact. All
descriptions of the world that refer to subfactual matters—objects, properties,
and relations—are nonfundamental, and may be understood in multiple,
equally good ways, in terms of the fundamental unstructured facts. I suspect
that Hirsch has something like this in mind (minus the bits about joint-
carving).39 It may perhaps be thought of as the metaphysical correlate of
Frege’s (1884, Introduction) context principle, which says that sentence-
meaning is in some sense prior to word-meaning. But it’s no good stopping
with just this intuitive picture. What exactly is the proposed array of fact-
level fundamental notions, how can they be shown to underlie the
quantificational languages of science, and do they really not re-introduce
analogs of the very questions they were designed to avoid? As we have seen
in this section, these questions need to be faced.
It’s a pipe dream, this idea of a nonquantificational language that is
adequate to science and ordinary life while not re-introducing substantive
questions just like those of traditional ontology. The quantifier variantist
could keep chasing the dream. But perhaps a better course would be to reject
the assumption that the fundamental is complete, thus eliminating the need
for the chase. He could then say: even though quantificational languages fail
to carve at the joints, we can do no better. There are many quantificational
languages (corresponding to the many candidate quantifier meanings); we
can use some of them to give semantic accounts of others; none of them is a
fundamental language; and there is no fundamental language in which one
can give a semantics for all of them. What is reality fundamentally like, one
wants to ask? This quantifier variantist denies the need for an answer. There
is no way that reality is fundamentally like; there simply is no book of the
world.
 Is this a defensible position? Well, it is utterly at odds with the intuitive
idea of fundamentality—of there being facts that underlie everything else.
Further, it faces a dilemma. Do some predicates (or other object-theoretic
expressions) carve at the joints?
Horn 1: no predicates carve at the joints. Here only two unattractive
options seem open. One is Goodmania: all talk of objective joints in reality is
simply mistaken. The other is reverting to the pipe dream—hoping that some
nonquantificational fundamental language might one day be discovered
which would allow one to recover some sort of inegalitarian distinction over
predicates, even though the predicates don’t themselves carve at the joints.
Horn 2 is a more likely resting place: some predicates do carve at the
joints, even though the quantifiers don’t. But this is hard to square with purity
(section 7.2). If some predicate F carves at the joints, then surely some
sentence S containing F is a fundamental truth. But S must surely also contain
either a name or a quantifier. Given quantifier variance, quantifiers don’t
carve at the joints, and surely names don’t either. So S, a fundamental truth,
contains an expression that doesn’t carve at the joints—a violation of purity.
We have seen substantive questions of ontology re-emerge, even from the
barren soil of fundamental ideology constructed expressly to inhibit their
growth. The pattern here is instructive, and not specific to ontology. The
philosophical instinct is an almost perverse tendency to ask questions that
push us out of our epistemic comfort zone. Just as a child asks her parents
who made God, so the philosopher asks who or what makes torturing wrong,
how we know we’re not all dreaming, how we can be blamed when our
actions are causally determined, and so on. Our normal epistemic practices
easily answer many questions about right and wrong, knowledge and blame,
but are perplexed by these philosophical questions. Just so for ontological
questions: the philosopher takes the notion of there being something, a notion
integral to ordinary and scientific practice, and asks an extraordinary question
using it. There’s nothing wrong with the notion itself; and so there’s nothing
wrong with extraordinary questions asked in terms of it. They’re just hard to
answer.

9.6.3 No epistemic high ground

A third argument attacks a certain putative reason for accepting quantifier
variance (and other forms of ontological deflationism) rather than the view
itself. Many are attracted to quantifier variance because they think it will give
them the epistemic high ground. They see ontologists perennially searching
for answers to the same old questions, sometimes with new methods,
sometimes with old, but never with much success. Given ontological realism,
ontological questions are “epistemically metaphysical”: they resist direct
empirical methods but are nevertheless not answerable by conceptual
analysis. Epistemically metaphysical questions can seem unanswerable. But
given quantifier variance, ontological questions are no longer epistemically
problematic in this way: now they can be answered by conceptual analysis.
I argued in abstract terms in section 5.6 that such attempts to gain the
epistemic high ground are misguided, and this particular case is no exception.
For quantifier variance doesn’t dispel the unanswerable questions—not even
those in the vicinity. First, as we saw in section 9.6.2, once the quantifier
variantist has banned quantifiers from her fundamental language, there is
pressure to specify their replacements. And on many of the replacements
discussed there, new epistemically metaphysical questions—analogous to the
rejected questions of ontology—re-emerge. Moreover, quantifier variance
itself—the assertion that quantifiers do not carve at the joints—seems to be
epistemically metaphysical. So it’s not as if the quantifier variantist can avoid
such questions altogether. In fact, there is a sense in which even ontological
questions remain epistemically metaphysical for the quantifier variantist. For
since direct empirical methods and conceptual analysis don’t tell us that
quantifier variance itself is true, such methods don’t tell us that ontological
realism is false; nor do they, conditional on the hypothesis of ontological
realism, rule out the truth of ‘There are no tables’ under the joint-carving
meaning of the quantifiers. So for all direct empirical methods and conceptual
analysis tell us, tables might fail to exist (at least in the fundamental sense).
Quantifier variance is just another metaphysical story about ontology,
alongside views like mereological universalism and mereological nihilism
that presuppose ontological realism. The appearance that it is an epistemic
panacea is sustainable only if one forgets both i) the need for a replacement
for quantifiers in one’s fundamental ideology; and ii) the epistemic status of
quantifier variance itself. Of course, quantifier variance might still represent a
modicum of epistemic advance. Still, I expect that for some, once a decisive
march to high ground is seen as unattainable, the position will lose its appeal.

9.6.4 Indispensability
The best argument against quantifier variance, and indeed against all forms of
ontological deflationism, is really quite simple. As I argued in section 2.3, the
way to tell which notions carve at the joints is broadly Quinean: believe in
the fundamental ideology that is indispensable in our best theories. This
method yields a clear verdict in the case of quantification. Every serious
theory of anything that anyone has ever considered uses quantifiers, from
physics to mathematics to the social sciences to folk theories. And as we saw
in section 9.6.2, there is no feasible way to avoid their usage. Quantification
is as indispensable as it gets. This is defeasible reason to think that we’re onto
something with our use of quantifiers, that quantificational structure is part of
the objective structure of the world, just as the success of spacetime physics
gives us reason to believe in objective spacetime structure.
Further, just as the success of particle physics suggests that ‘charge’,
‘mass’, and the like correspond to unitary structure, rather than fragmented
structure as with ‘jade’, so the indispensability of quantification suggests that
quantificational structure is unitary. It therefore argues against ontological
pluralism, the view that there are multiple fundamental sorts of
quantification.40
Questions framed in indispensable vocabulary are substantive; quantifiers
are indispensable; ontology is framed using quantifiers; so ontology is
substantive— that’s the best argument for ontological realism.

9.7 Easy ontology

There are forms of ontological deflationism other than quantifier variance.
My last three arguments against quantifier variance were arguments that
quantifiers carve at the joints, and hence are effective against any form of
ontological deflationism. Still, let us consider some of these other forms in
detail.
The main other form is what we might call “easy ontology”. Its defenders
say that questions of philosophical ontology are settled by analytic truths
together with certain obvious facts. For example, the question of holes is
settled by the analytic truth that if a sock is perforated then there is a hole in
that sock, together with the obvious fact that some socks are perforated.
Ontology is “easy”, according to this conception, because it is easy to
know what is analytic—we simply need to reflect on how we use language.
Given easy-to-know analytic truths, and given further information that’s easy
to know (such as the fact that some socks are perforated), it’s easy to answer
(in the affirmative) many ontological questions. This is not to say that the
resulting entities owe their existence to our language use—that their existence
is caused by, or counterfactually dependent on, our language use. It’s a
linguistic convention that ‘There are electrons’ means what it does, but this
does not make electrons themselves dependent on us; similarly, though it’s a
linguistic convention to count ‘There is a hole in that sock’ true whenever the
sock is perforated, this doesn’t make the hole dependent on us. Indeed, the
linguistic conventions governing ‘hole’ likewise secure the analyticity of
sentences like these:

Before there were people, when something was perforated (e.g., a certain
mountain) there was a hole in that entity (e.g., a cave). If there had been no
people and something had been perforated, there would have been a hole in
that entity.

Easy ontology has many defenders, in different branches of philosophy. In

the philosophy of language, Stephen Schiffer (2003) has argued that the
existence of properties, propositions, events, and various other entities is
easy, being secured by what he calls “something-from-nothing
transformations”—analytic conditionals like the following:
If Lassie is a dog then Lassie has the property of being a dog.
Schiffer imagines a linguistic community that does not speak of properties,
asks what it would take for them to come to know that properties exist, and
answers:
What it would take, and all that it would take, would be for them to
engage in a certain manner of speaking, a certain language game—namely,
our property-hypostatizing practices, in particular our property-yielding
something-from-nothing transformations… How can merely engaging in a
linguistic, or conceptual, practice give one knowledge of things that exist
independently of that practice? Because to engage in the practice is to have
the concept of a property, and to have the concept of a property is to know a
priori the conceptual truths that devolve from that concept, such as the
conceptual truth that every dog has the property of being a dog. (Schiffer,
2003, p. 62)
In the philosophy of mathematics, an instance of easy ontology is the neo-
Fregeanism discussed by Crispin Wright and Bob Hale. According to
neoFregean-ism, Hume’s Principle is analytic:

If the Fs and the Gs are equinumerous, then the number of Fs equals the
number of Gs.

Thus we have an easy route to the existence of numbers. Suppose there are
two apples and two oranges in the bowl. Then the apples in the bowl are
equinumerous with the oranges in the bowl, whence by Hume’s Principle the
number of apples in the bowl equals the number of oranges in the bowl. But
then, there is such a thing as the number of apples in the bowl. So, there are
such things as numbers.41 In the philosophy of art, Amie Thomasson has
defended the view that conceptual truths assure us of the existence of works
of art (2005; 2006) and fictional characters (2003). Unlike holes and
numbers, works of art and fictional characters are quite generally brought into
existence by human beings. But as above, this does not mean that the
adoption of linguistic conventions brings these entities into existence. It
rather means that the conceptually sufficient conditions for their existence
make reference to certain human activities, and specify that the entities come
into existence only when the activities are performed. Thomasson writes:
Our literary practices … definitively establish the existence conditions for
fictional characters … According to those criteria, what does it take for an
author to create a fictional character? This much is clearly sufficient: That she
write a work of fiction involving names not referring back to extant people or
characters of other stories, and apparently describing the exploits of
individuals named … (2003, 148)
Thomasson also defends a similar view about the ontology of material
objects.42 On her view, statements giving the existence and persistence
conditions of ordinary objects such as tables, sticks, and baseballs are
analytic. For example, it is analytic that if some particles are arranged
baseball-wise, then there exists a baseball.
Easy ontology rests on the notion of analyticity But what does it mean to
say that a sentence is analytic?

9.8 Analyticity
While I reject the claims of Quine (1936; 1951b; 1960a) and his minions that
analyticity is an incoherent notion, much of Quine’s critique is, I believe,
correct. There are several notions in the vicinity, some legitimate, some not;
and often an illusion of progress comes from illicitly shifting between
different notions.43
Two of these notions are truth by convention and definitional constraint.
Truth by convention is a putatively metaphysical notion: a sentence is said to
be true by convention when the conventions governing that sentence
somehow suffice on their own to produce the sentence’s truth. Definitional
constraint is a semantic notion: a sentence S is a definitional constraint on
expression E when S plays a certain distinctive role in helping to determine
the meaning of E. The role, intuitively, is that there is “metasemantic
pressure” towards interpretations of E under which S is true. A natural
thought about the source of this metasemantic pressure is that it results from
our intending (in a meaning-constituting way) E to mean something under
which S is true. Examples: ‘All bachelors are married’ is a definitional
constraint on ‘bachelor’; a sentence expressing the transitivity of ‘=’ is a
definitional constraint on ‘=’; ‘Nothing is red and green all over’ is a
definitional constraint on ‘red’ and ‘green’ jointly. We may also speak of
rules of inference as definitional constraints: the idea here is that there is
metasemantic pressure towards interpretations under which the rule comes
out truth-preserving. The rules of ‘and’-introduction and ‘and’-elimination,
for example, are definitional constraints on (the logician’s) ‘and’.
As we saw in section 6.5, truth by convention was discredited long ago, in
large part by Quine. This part of the Quinean critique is beyond reproach. But
with many I reject the part that is directed at defnitionality. Quine argued in
“Two Dogmas of Empiricism” that confirmation holism implies meaning
holism, which he then alleged to be inconsistent with the coherence of
analyticity—here clearly meaning definitionality; but this argument had a
dubious verification theory of meaning as a premise.44 A second argument
from “Two Dogmas” was that it is difficult to reductively define ‘analytic’
(again, in the sense of definitionality) without appealing to related notions
such as meaning or rule of language. But as Grice and Strawson (1956)
quickly pointed out, inability to define a term should not make us doubt that
term’s sense. Definitionality depends, somehow, on dispositional,
psychological, historical, and environmental facts about language users.
Turning this bland observation into a philosophical definition is a
monumental task. But so are all tasks of philosophical definition. Failure to
define ‘definitional’ is no more cause for alarm than failure to define ‘city’,
‘smile’, or ‘candy’.
Though I cannot define ‘definitional’, I can say a bit about it. Meaning is
determined, somehow, by a complex array of facts about us—call these the
facts about how we “use” language—as well as a complex array of facts
about our relation to the world. Definitionality is primarily a function of the
first array of facts, the facts of use.45 Think of definitional constraints as
messages we send to the semantic gods: “Insofar as you can, interpret our
words so that these sentences come out true, and these inferences come out
truth-preserving.” Definitionality presumably comes in degrees: sentences or
inferences can be more or less defi-nitional (perhaps even: more or less
definitional with respect to different terms). I think of definitional constraints
as being in-principle available to competent speakers, or groups of competent
speakers, though not infallibly or immediately so.
Definitional sentences need not be true; and definitional rules need not be
truth-preserving.46 For definitionality can be trumped. A definitional sentence
or rule of inference governing certain terms might not be the only
metasemantic pressure on those terms’ interpretation. Countervailing
metasemantic pressure from other sources might lead the semantic gods to
assign those terms no meanings at all, or to assign them meanings that render
definitional sentences or rules false or non-truth-preserving. The
countervailing pressure might come from other facets of our usage of the
terms (including, but not limited to, other definitional sentences or rules), or
it might have a more metaphysical source.
Some examples. 1. Arguably, ‘There is absolutely no space between two
objects in contact’ is false but definitional of ‘contact’. The countervailing
pressure comes primarily from use: other things we say using ‘contact’ lead
the semantic gods to sacrifice this one definitional constraint. 2.
Compatibilists reject the sentence ‘No free action is determined by the past +
laws’; but this sentence is arguably definitional of ‘free’. Part of the
countervailing pressure comes from the usage of ‘free’ in other contexts, but
another part has a metaphysical source: the nonexistence (and even
incoherence, on some views) of incompatibilist freedom. 3. The introduction
and elimination rules for ‘tonk’ are definitional, but are not (both) truth-
preserving. Here the countervailing source has nothing to do with use, but
rather is purely metaphysical (or perhaps we should say logical): the semantic
gods cannot acquiesce to the definitional constraints on pain of contradiction.
4. Pretend that physical theory includes a primitive theoretical predicate of
times, ‘earlier-than’. No physicist, let us suppose, questions the assumption
that this relation is connected—that for any two distinct times, one of them is
earlier than the other. Indeed, if asked to give “a definition” of ‘earlier-than’,
physicists might cite connectedness as one of its defining characteristics. So a
sentence expressing connectedness is at least somewhat definitional of
‘earlier-than’. Nevertheless, we can imagine the sentence to be false. Suppose
there is just one joint-carving relation in the vicinity and that it is not a
connected relation. Under this joint-carving relation, times are grouped into
two separate time-lines, say. Surely this relation is what ‘earlier-than’ would
denote. The definitional sentence about connectedness would be trumped by
countervailing metasemantic pressure of a metaphysical source: the pressure
to assign the joint-carving relation to ‘earlier-than’. This pressure could be
due to reference magnetism, but it needn’t be. It could instead be based on
the fact that ‘earlier-than’ is a theoretical term: it is intended to stand for
whatever joint-carving notion is in the vicinity. 5. Consider an intuitionist of
a rather metaphysical stripe,47 who thinks that there is just one joint-carving
notion in the vicinity of negation, which satisfies an intuitionistic logic. Such
an intuitionist does not accept the rule of double-negation elimination, but
might nevertheless concede that this rule is definitional, in ordinary English,
of ‘not’. On this view the definitional claim of that rule to be truth-preserving
would again be trumped, as in example 4, by the presence of a rogue joint-
carving notion that does not satisfy the definitional constraint.
Though definitional sentences can turn out false, often they turn out true.
We need a word for true definitional sentences. I propose: ‘analytic’.
Analytic sentences thus understood are meaning-constraints that, as it
happens, succeed. It might be thought inadvisable to re-use that old word
here, since (as we will see) true definitional sentences do not play all of the
traditional role of analyticity. But it plays some of that role; and as for the
rest, nothing plays it. True + definitional is what we ought to mean by
‘analytic’.
Analyticity, thus understood, carries no commitment to truth by
convention. To count as analytic, a sentence must first be true—true on its
own steam, so to speak. To say that a sentence is analytic is to say that it is a
certain kind of truth; it is not to explain why it is true. As a result, analyticity
does not have the epistemic significance it is often taken to have—we will
return to this.
Analyticity is often accompanied by nonsubstantivity in the sense of
chapter 4. For the point of a definitional constraint is often to (help) select
one out of a range of equally joint-carving candidates. We are aware, for
example, that there are many, equally joint-carving, pairs of meanings we
could assign to the words ‘foot’ and ‘inch’. So we partially constrain the
interpretation of this pair by laying down (F) as a definitional constraint:
(F) Something is a foot long iff it is twelve inches long.
Since some alternate constraints that we could have laid down, such as ‘11
inches make a foot’, would not have been semantically alien, (F) is
nonsubstantive (conventional, in fact, in the sense of section 4.3). Further,
since there is no countervailing metasemantic pressure against interpreting
‘foot’ and ‘inch’ so that (F) is true, (F) is indeed true, and hence analytic.
However, analyticity and nonsubstantivity can come apart. One instance is
when there is a fortuitous convergence of definitionality and joint-carving.
Suppose (what I do not believe) that parthood is a fundamental relation, and
that the sentence expressing the transitivity of ‘part’ is definitional of ‘part’.
If this sentence is true, then it is analytic; but it is substantive: any candidate
meaning for ‘part’ on which the sentence is false would presumably fail to
carve at the joints. An analogous point can be made for “analytic inferences”:
truth-preserving inferences that are definitional. The inferences that are
distinctive of conjunction (from a conjunction to either of its conjuncts; and
from the conjuncts together to the conjunction) are both truth-preserving and
definitional, but are substantive, I would argue. Conjunction carves at the
joints; and there are no equally joint-carving candidates for ‘and’ under
which the rules fail to be truth-preserving.
Nothing can fully play the role traditionally associated with analyticity,
for much ofthat traditional role presupposed the doctrine of truth by
convention.48 For instance, a sentence’s analyticity was supposed to explain
how we could know it to be true—we could know it to be true because the
conventions we know ourselves to have adopted sufficed to make it true. That
is why the positivists were so keen to demonstrate the analyticity of logic and
mathematics. The downside of this favorable epistemic status was the
apparent triviality of analytic truths.
Analyticity, on my conception, does not explain knowledge. Obviously,
analytic sentences must be true, given my definition, but knowing that a
given sentence is analytic is the rub. At best, we have unproblematic
knowledge of definitionality (and even that is not really epistemically
unproblematic). But no matter how definitional a sentence (or inference rule)
is, it may still fail to be true (or truth-preserving), in the ways described
earlier. Nor are analytic truths trivial. As we learned from Quine’s critique of
truth by convention, even the logical truths say something about the world.
Analytic sentences do not comprise some weird representational species,
somehow managing to be true without really saying anything. Like any other
sentence, a definitional sentence must measure up; the world must be as it
says in order for it to be true.
The failure of analyticity to generate knowledge is especially significant
in logic, given the aspirations of the traditional conception. The case of ‘tonk’
shows that definitional inference rules in logic needn’t be truth-preserving.
How, then, a skeptic might ask, do we know that the standard rules, such as
‘and’-introduction and elimination, are truth-preserving? Perhaps ‘and’ is
tonk-like, in that its introduction and elimination rules somehow lead to
contradiction. Or perhaps candidate meanings vindicating those rules are
simply missing. Or perhaps joint-carving candidate meanings vindicating
those rules are missing, and ‘and’ is a theoretical term intended to stand for a
joint-carving meaning. Or perhaps ‘and’ is a theoretical term, and there is a
single joint-carving candidate in the vicinity that fails to vindicate ‘and’-
introduction and elimination. We don’t believe any of these odd possibilities
to be actual, but how can we claim to know this, without presupposing the
sort of logical knowledge that is at issue? This hard problem in the
epistemology of logic cannot be solved by reflection on meaning.49
There is a kernel of truth in the traditional view. In contexts in which the
existence of equally joint-carving candidate meanings is not in question, and
in which it is known that there are no conflicting facets of use, definitional
sentences have the traditional features “relative to the context”. Consider a
context in which everyone takes it as wholly obvious that there exist equally
joint-carving properties being an unmarried male, being an unmarried male
eligible for marriage, being an adult unmarried male, being an adult
unmarried male eligible for marriage, and so on (and in which a suitable
logic is being taken for granted). Someone then offers up a stipulative
definition: ‘bachelor’ is to mean the same as ‘unmarried male’. Everyone in
the conversation would then regard ‘Something is a bachelor iff it is an
unmarried male’ as epistemically secure, and they would regard assertions of
this sentence as trivial, as contributing nothing to conversations. But that is
only because the facts required to bridge the gap between defnitionality and
analyticity are not in question in the context.

9.9 Against easy ontology

Let’s return to the easy ontologist’s claim that we can tell, just by reflecting
on language, that sentences like (T) are true:
(T) If some things are arranged tablewise then there exists a table.
As we saw in the previous section, linguistic reflection can deliver at best the
conclusion that (T) is definitional. And being definitional is insufficient for
truth: (T)’s definitional status might be trumped by some other factor.
In particular, suppose first that there is a joint-carving candidate to be
meant by ‘there exists’ that satisfies enough of the other definitional
constraints on ‘there exists’, but which does not satisfy (T). (This is, in fact,
exactly what “mereological nihilists” believe.) And suppose, second, that the
simple charity-based descriptivist metasemantics discussed in section 3.2 is
true. In that case, this joint-carving candidate would be meant by ‘there
exists’ (its superior eligibility having trumped its failure to satisfy (T)), and
(T) would be false.
Or, make the first supposition again; but now, instead of assuming simple
charity-based descriptivism, suppose instead that ‘there exists’ is a theoretical
term—a term intended to stand, if possible, for a joint-carving meaning.
‘There exists’ might again mean the joint-carving candidate, rendering (T)
false. Or— continuing to take ‘there exists’ to be a theoretical term—if the
definitional claim of (T) is strong enough, ‘there exists’ might then be
semantically defective (the conflicting metasemantic pressures being too
strong to resolve), in which case (T) would be untrue because semantically
defective.
In these ways, the presence of a joint-carving sort of quantification could
render (T) untrue, even if (T) is definitional. Easy ontologists cannot,
therefore, claim merely that (T) is definitional. They must also reject joint-
carving quantification.50 And so my arguments for ontological realism in
sections 9.6.2–9.6.4 tell also against easy ontology.
In section 9.3 we saw that even an ontological realist might admit that
English quantifiers do not carve at the joints. Something similar is true here:
even an ontological realist might admit that English ontology is easy.
Suppose, for example, that English quantifiers aren’t theoretical terms;
English speakers aren’t “trying” to carve at the joints with such terms. Then,
assuming the falsity of simple charity-based descriptivism, there would be
comparatively little danger of the joint-carving sort of quantification
trumping English usage. Alternatively, even given simple charity-based
descriptivism, suppose that in the joint-carving sense, all that exists are
subatomic particles. In that case, it might be argued, since so many ordinary
statements about macro-entities would come out false under that austere
candidate, this mismatch with usage outweighs its superior joint-carving.
Either way, the ontological realist might then say that English quantifiers
have a meaning that fits their ordinary usage, that (T) has an untrumped
definitional status, and thus that (T) is analytically true in English. Thus,
when applied to English quantification, the easy ontology picture might well
be correct, even if ontological realism is true. But in that case, the appropriate
language for conducting ontology would be Ontologese, in which the
quantifiers are stipulated to carve at the joints, and in which sentences like
(T) are not definitional. English is second-rate; the value of inquiry in joint-
carving terms is superior (recall section 4.5). Ontology in Ontologese remains
hard—and better.

9.10 Other forms of easy ontology

There are other approaches to ontology that one might classify as “easy”. For
example, Thomas Hofweber (2000; 2005; 2007) distinguishes between
“internal” and “external” uses of quantifiers. External uses are
“metaphysically loaded”; they are used to make statements about objects that
are “‘out there’ as part of reality” (2007, p. 27). Internal uses, on the other
hand, have an entirely different linguistic function—roughly, to form
sentences that inferentially connect to their “instances” in the usual way, even
if the “singular terms” in those instances do not in fact have the semantic
function of referring to entities. For example, the existential quantifiers in (J)
and (F) are internal:51
(J) There is a number which is the number of moons of Jupiter.
(F) There is someone who is admired by Fred and many other detectives.
(J) follows from ‘The number of moons of Jupiter is four’, and (F) follows
from ‘Sherlock Holmes is admired by Fred and many other detectives.’ But
in the latter sentences, ‘the number of moons of Jupiter’ and ‘Sherlock
Holmes’ are not referential expressions. (‘The number of moons of Jupiter is
four’ is, according to Hofweber, a syntactic variant of ‘Jupiter has four
moons’, which contains no singular term purporting to refer to a number.)
Internal ontology is easy. You don’t need to know that there are numbers
“out there” in order to know that there are numbers in the internal sense; it is
enough to know that Jupiter has four moons. Nevertheless, Hofweber’s view
does not conflict with ontological realism (not that he says it does), for the
simple reason that even if he is right about natural language quantifiers,
ontologists could always relocate to the metaphysics room, and conduct their
dispute in Ontologese.52
A similar point can be made about Stephen Yablo’s (1998; 2000) view
that some quantificational language is metaphorical, and indeed about any
other view according to which some natural language quantifiers are
“lightweight”. Even if correct, such views do not challenge ontology as
practiced in Ontologese. (Such views may, of course, help to support sparse
positions in first-order ontology, by undermining arguments such as the
following: “(i) John walked; so (ii) there is something John did; so (iii)
nominalism about events is false.” Nominalism about events says that there
are no events in the heavyweight sense; so the quantifier in (ii) must be
heavyweight; but, one might argue, the move from (i) to (ii) is valid only if
the quantifier in (ii) is lightweight.)
D. M. Armstrong (1997, section 2.12) defends another form of easy
ontology:
[A supervenient entity] is not something ontologically additional to the subvenient, or necessitating,
entity or entities. What supervenes is no addition of being. Thus, internal relations are not ontologically
additional to their terms. Mereological wholes are not ontologically additional to all their parts … One
may call this view … the doctrine of the ontological free lunch.
Although Armstrong does not give a precise meaning to “no addition of
being”, the phrase has intuitive content.53 It connects to the notion of the
world being more or less “full”: if accepting x makes the world no fuller, then
x is no addition to being. It connects to commitment: the admonition to
choose theories with fewer commitments (whether ontological or ideological)
tells us to choose sparser, less full, worlds. The term remains vague; but even
so, I think we can see that Armstrong is wrong that the supervenient is ipso
facto no addition to being. For additions to being should be measured at the
fundamental level. If you add to fundamental reality, you add to being, make
the world a fuller place, commit yourself further—even if what you added is
supervenient. Supervenience is just a kind of modal connection, and such a
connection might hold between equally fundamental entities or facts.
Suppose, for example, that quantifiers carve at the joints, and that there is
something (in the joint-carving sense), x, that is composed of y1…yn. The
existence of x is an addition to being, even if it’s necessarily true that a
composed entity must exist whenever y1…yn exist. For the composed entity’s
existence isn’t “metaphysically reducible” to the existence ofy1…yn;
‘Somethingis composed of y1…yn’ is a true sentence in a fundamental
language, and has no metaphysical truth-condition (except perhaps a
homophonic one). This can be put in Fine’s terms as well: even if the
existence of a composed object is necessitated by the existence of y1…yn (and
even if this necessitation holds in reality), if the proposition that something is
composed of y1 …yn holds in reality—i.e., is a fundamental truth—then
intuitively, the composed object is still an addition to being. But suppose, on
the other hand, that it’s not the case in reality that there is a composed object;
rather, the proposition that something is composed of y1…yn is grounded in
the proposition that y1…yn exist and stand in a certain relation. Or suppose, in
my terms, that ‘Something is composed of y1…yn’ is not true under the
fundamental sense of the quantifiers; rather, that sentence is true under a
nonfundamental sense of the quantifiers, and is governed by the metaphysical
truth-condition that y1… yn exist (in the fundamental sense) and stand in a
certain relation. Either way, we can then say that the composed object is no
addition to being to y1…yn. For it is not a fundamental addition.
The supervenient lunch doesn’t even look free in certain cases. Given a
certain form of platonism, numbers are necessary existents and so are trivially
supervenient (on anything), yet they clearly seem extra. Given a certain
theology, there necessarily exists a god (though his states are presumably
contingent); such a god would obviously be extra. In these cases, there is no
tendency at all to say that supervenient entities are no additions to being.
What went wrong? “Platonic” numbers and the god would exist
fundamentally; and supervenience is a sign of a failure to be extra only when
the supervenient is nonfundamental. The ontologi-cal free lunch is based on
the thought that modality is the measure of all things metaphysical. We have
yet another example of the inadequacy of that thought, of modality as a
foundation for metaphysics.
9.11 Metaontology and conceptions of fundamentality
My discussion of ontological realism and ontological deflationism has mostly
assumed my framework for talking about fundamentality. How might
alternate frameworks affect the discussion?
Consider, first, Fine’s framework, which appeals to the notions of
grounding and reality. As noted in section 8.3.5, this ideology has a hard time
distinguishing deflationary views from nihilistic views. For example, the
causal deflationist holds that statements about causation are nonsubstantive,
whereas the causal nihilist holds that statements about causation are
substantive, and furthermore, that fundamentally, nothing causes anything. In
my terms, the difference is that the causal nihilist thinks that there is a sense
of ‘cause’ that carves at the joints, whereas the causal deflationist denies this;
but in Fine’s terms, the views are hard to distinguish.
This same problem emerges in the present context. How can a Finean
distinguish an ontological deflationist from an ontological nihilist, who thinks
that ontological questions are substantive but that nothing fundamentally
exists? Each will presumably want to deny that existentially quantified
propositions—propositions of the form ∃xϕ—are real. Might the difference
be that the ontological nihilist will say that in reality, nothing exists? No.
Since ‘in reality’ is surely factive, it would follow that nothing exists, which
our ontological nihilist needn’t accept.
Consider, next, Schaffer’s framework, which appeals to a relation of
grounding between objects. Schaffer takes his framework to support one sort
of deflationism and another sort of anti-deflationism. The supported
deflationism is about the Quinean question do there exist Fs?:
… while the Quinean will show great concern with questions such as whether numbers exist, [I] will
answer such questions with a dismissive yes, of course. (2009a, p. 352)

The supported anti-deflationism is about the question assuming that F s exist,

are they grounded in further objects, and if so, in which further objects?
Philosophical ontology, according to Schaffer, takes any given category of
interest (numbers, for example), assumes that there are objects in that
category, and tries to figure out whether and if so how those objects are
grounded.
On the face of it, though, it’s hard to see why dismissive affirmative
answers to the Quinean question are justified. Suppose we believe in entities
that would successfully ground Fs, if Fs exist. Even so, what would be wrong
with denying that Fs exist? If such a denial would not be incoherent, then the
Quinean ontological question of whether Fs exist at all would seem to remain
open.
Quantifier variance, or some other form of easy ontology, would justify
the dismissive affirmative answers. Given quantifier variance, it will be
obvious (in many cases, anyway) that there are quantifier meanings on which
‘There are Fs’ comes out true, meanings that carve at the joints as well as any
on which the sentence comes out false. Given that our usage of ‘There are Fs’
favors its truth, we can be confident that it is in fact true (modulo our
knowledge of quantifier variance). But Schaffer does not base the dismissive
affirmative answers on such thoughts (2009a, p. 360). His idea, rather, is that
belief in numbers, composites, and so on, is just so much common sense. It’s
not that disbelieving in them is incoherent; it’s rather that disbelieving in
them would be maximally unjustified. But now the framework of grounding
is playing no role in the deflation of the Quinean question; the question’s
answer is obvious purely because of the Mooreanism. An ontological realist
of my sort who accepted this Mooreanism would agree, for precisely the
same reason, that it’s obvious that there are numbers, composites, and so on.
I must grant, though, that grounding does play this role: it preserves, for
anyone who accepts Mooreanism, a domain for philosophical inquiry. For
even if it’s obvious that there are numbers, composites, and so on, given
Schaffer’s framework there remain questions of how numbers, composites,
and the rest are grounded. But even an ontological realist can say something
similar. The propositions that are certain, given Mooreanism, are surely those
expressed by ordinary sentences. So even if it’s quite certain that numbers,
composites, and so on, exist in the ordinary sense, there would remain the
question of what exists in the fundamental sense—of what is true in
Ontologese.54
Consider next what Matti Eklund (2006) calls “maximalism”, according to
which reality is in a sense “full”. Within certain limits,55 all objects that can
coherently be supposed to exist, do exist. Thus there exist numbers, sets,
properties, propositions, events, mereological composites, and so on.
Maximalism, moreover, does not rest on quantifier variance. The
maximalist’s picture, rather, is that these objects exist under the one and only
meaning for ‘exist’.
Eklund thinks of maximalism as a form of ontological deflationism; but
on the face of it, maximalism seems instead to be just another stand on
substantive first-order ontology, and indeed, an extreme one. But perhaps a
second look is warranted. Maximalism is a quite general hypothesis about
reality, such that given the hypothesis, there can be no further doubt about the
existence of Fs (for appropriate choices of F). If you accept that existence is
quite generally maximal, you should have no further doubt about whether
composites, or numbers, or other “philosophical” entities exist. Of course,
you can doubt maximalism itself. But compare the situation with quantifier
variance. There too, it’s only under the assumption that quantifier variance is
true that there is any guarantee that there are tables. Even a quantifier
variantist could entertain doubts about tables, since he could entertain doubts
about quantifier variance. The most that any ontological deflationist can hope
for, I think, is to show that doubts about ontological questions in particular
domains are misguided, granted some quite general hypothesis about the
nature of ontology.56 So there is a sense, after all, in which maximalism is a
kind of deflationism.57
But notice this. As a substantive metaphysical hypothesis, maximalism
should be assessed by comparing its ontology and ideology to its rivals. Even
setting aside the objection that its ontology is profligate, there is a more
serious worry: its ideology is extremely profligate.58 For the maximalist
presumably recognizes distinctive fundamental ideology for each of the
groups of entities he embraces: integers, sets, composites, events,
propositions, and so on.59 (Reducing successor to set-membership, for
example, rather than taking it to be primitive, would seem to turn
maximalism into a more standard sort of reductionism.)

9.12 Ontological commitment

The remainder of this chapter will be a bit of a digression: a discussion of a
scattered set of ontological issues, in the light of ontological realism,
beginning with ontological commitment.
Contemporary use of the term ‘ontological commitment’ is perplexing in a
few ways. To take a representative example, it is debated whether saying that
there is a hole in one’s sock “ontologically commits” one to holes. This is
supposed to be an open philosophical question, alongside the other greats
such as the question of whether we have free will and the question of whether
we know anything. But ‘ontological commitment’ is a technical term, and has
no semantic life outside of what we stipulate for it. Moreover, the most
straightforward stipulation is that ‘ontologically committed to Fs’ just means
‘committed to Fs’, which, in addition to being a waste of screen real-estate,
precludes what people seem to take to be an open position: believing in holes
in one’s sock without ontologically committing to holes.
Ontological realism clears all this up. Define ‘ontologically committed to
Fs’ as meaning ‘believing that there are Fs, in the fundamental sense of
“there are”’. This allows someone to say in English that there are Fs without
ontologically committing to Fs, if English quantifiers are nonfundamental.
And it makes sense of the fact that ontological commitment is supposed to be
a big deal.
Other talk of ontological commitment can be cleared up in similar fashion.
We can speak of ontological commitment for quantifiers (or quantifier-
interpretations), for example: a quantifier (interpretation) is ontologically
committing iff it carves at the joints. Thus an ontological realist could claim
that whereas some quantifiers are ontologically committing (those in
Ontologese, and perhaps also some natural language quantifiers), other
quantifiers (perhaps those in “There are many ways to win this chess match”
and “Some detectives are fictional”) are not. If each singular term is
associated with a unique quantifier—a natural assumption—then we can also
speak of ontological commitment for singular terms: a singular term is
ontologically committing iff its associated quantifier is ontologically
committing. For example, ‘Sherlock Holmes’ would fail to be ontologically
committing if it is associated with the quantifier in “Some detectives are
fictional” and that quantifier fails to be ontologically committing.
Another perplexing thing is the common focus on unwitting ontological
commitments. There is supposed to be a question of whether a person who
says that an apple is red has ontologically committed herself to a universal (or
property) of redness, even if she claims not to believe in universals.
Nominalists live in fear that they will be shown, by philosophical argument,
to have unwittingly committed themselves all along to the position of their
opponents!
We would be better off not talking this way, I think. If you think that
apples can’t be red without there being universals, then just say that. It
muddies the waters to put this by claiming that anyone who says that the
apple is red commits herself to universals.
It might be argued that one unwittingly commits to abstracta via the true
metaphysical semantics.60 If the true metaphysical semantics regards ‘The
apple is red’ as having the metaphysical truth-condition that the apple
instantiates the universal of redness, then, we might say, one is
“metaphysically committed” to universals simply by saying that an apple is
red—the fundamental fact that one is getting at when one says ‘The apple is
red’, after all, involves universals. Fine; but the only way to establish this
metaphysical commitment—to establish that the metaphysical truth-
conditions for natural-language predications involve universals—would be to
argue that there are, fundamentally, universals. Thus it’s just a distraction to
bring up metaphysical commitment to universals in a debate over universals.
Further, ‘metaphysical commitment’ is a misleading name. ‘Commitment’
suggests some sort of obligation to believe, whereas metaphysically
committing to Fs means, in essence, making a statement whose underlying
metaphysics involves Fs; and ignorance about underlying metaphysics is
typically blameless. One might just as well say that anyone who says “There
are tables” is “physically committed” to subatomic particles.

9.13 Quantifiers versus terms

I formulated ontological realism as concerning quantifiers. Thus quantified
sentences express fundamental facts. But against this, a “Tractarian” form of
ontological realism might be advanced: the fundamental facts are
individualistic, not general; they are facts of the form Fa, Rbc, and so forth,
not facts of the form ∃ x F x or ∃ x ∃ y Rxy.
How we make sense of Tractarianism depends on how we think about
funda-mentality. Under my approach, we must locate some expression or
expressions crucial to ontology as conceived by the Tractarian, and claim that
those crucial expressions carve at the joints. If quantifiers are not the crucial
expressions, then what are?
One answer would be that each name carves at the joints. Another answer
would be that some names carve at the joints. A final answer, elusive and yet
perhaps more satisfying, would be that the grammatical category of the
singular term carves at the joints. Objecthood, so to speak, carves at the
joints.61
Each of these may seem strained. The strain is caused by my “atomistic”
approach to structure, which requires the ontological realist to locate a crucial
joint-carving expression. The strain would be eased by adopting Fine’s more
holistic approach for talking about fundamentality (chapter 8). The Tractarian
could then simply say that individualistic propositions are real, without
needing to identify any particular expression to which this is due.
Tractarianism, in either form, faces an immediate problem. Consider, first,
its formulation in my terms: quantifiers do not carve at the joints; names (or
the category of the name) do. If quantifiers do not carve at the joints,
quantified statements require a metaphysical semantics in fundamental terms.
But what metaphysical truth-condition for the sentence ∃xF x can be given
in purely individualistic terms? Any disjunction Fa1 V Fa2 V… faces the
modal objection that there could have been some further entity b, distinct
from the ai s. Fa1 VFa2 V… is therefore not modally equivalent to ∃xF x,
and so, it might be alleged, cannot be a metaphysical truth-condition for it.
For similar reasons, ∀xFx seems to have no purely individualistic
metaphysical truth-condition; in particular, it is not modally equivalent to any
conjunction Fa1 ∧ Fa2 ∧.…
In Fine’s terms, the modal argument could be put as follows. If
Tractarianism includes the view that no quantified statements hold in reality,
then all (factual) quantified statements must presumably be grounded in
individualistic statements. But no collection of individualistic statements is
modally sufficient for ∀xF x. Since grounding requires modal sufficiency,
there is no individualistic ground for ∀xF x. (The fact that Finean
fundamentality need not obey the combinatorial principle of section 8.3.2
opens up a weakened position according to which some but not all quantified
claims hold in reality. Perhaps only one quantified claim holds in reality:
this, together with various individualistic claims—
including negative individualistic claims—suffices to ground all quantified
claims.62 I wonder, though, what the attraction of this position is.)
The modal argument is inconclusive given “modal deflationism”—given,
that is, that modal words are not particularly fundamental and have multiple
candidates.63 For then the Tractarian might deny that a sentence’s
metaphysical truth-condition must be necessarily equivalent to it, or that
grounding requires necessitation. In principle these denials are open to
anyone, but they are more defensible given modal deflationism. For
defenders of fundamental modality surely regard modality as a “measure of
the facts”, in the following sense: if p can be true even though q is not, then p
and q are “entirely separate facts”; and so, p cannot ground or be a
metaphysical truth-condition for q. Modal deflationists, on the other hand,
could regard modal distinctions as metaphysical epiphenomena. They might
regard ∀xFx and Fa1 ∧Fa2 ∧… as really being “the same fact”, and so
might embrace the view that the latter is the metaphysical truth-condition, or
ground, of the former—even if our idiosyncratic concept of possibility counts
“possibly Fa1 ∧ Fa2 ∧… but not ∀xFx” as being true.64 (This is not to say
that knowing or asserting the generalization is the same as knowing or
asserting the conjunction. The sense of sameness of fact here is metaphysical,
not epistemic or semantic.)
Why might one be drawn to Tractarianism, or more generally, to
fundamental individualistic facts? I will consider three arguments. First, there
is an admittedly powerful intuition that, for example, the general fact that
∃xFx holds because of the individualistic fact that Fa. I wonder, though,
whether this intuition should be trusted. It seems to concern a distinctively
metaphysical sort of production or causation—the latter makes-true the
former—and I reject such thoughts unless they can be rephrased in
explanatory terms (chapter 8). And in this particular case, the thought does
not survive the rephrasal.65 The rephrased thought would be that a theory
without the quantifiers is more explanatory. But this thought is not at all
compelling, since quantificational explanations are usually better than purely
individualistic explanations.
Second, there is an elusive but somewhat compelling argument in favor of
recognizing something individualistic at the fundamental level in addition to,
rather than instead of, fundamental quantification. Suppose your fundamental
ideology includes existential and universal plural quantifiers in addition to
the usual existential and universal singular quantifiers. Suppose, further, that
you believe in some fundamental predicates. Now, these fundamental notions
all seem interconnected, through a single notion of “objecthood”. By this I
don’t mean a distinction between entities that are objects and those that are
not. I mean something more basic (and hard to express): a notion of
objecthood is “presupposed” by all quantifiers and predicates.66 The singular
quantifiers quantify over objects; the plural quantifiers also quantify over
objects, albeit in a different way; and predicates, too, are distinctively object-
theoretic. Intuitively: singular quantifiers, plural quantifiers, and predicates
are intimately connected notions; and so, according to the argument, you
ought to recognize, in your fundamental ideology, a notion of objecthood for
their overlapping content.
The basis of the argument is the thought that something needs to be
explained: namely, the connection between the various object-theoretic
notions. Now, there would indeed be something in need of explanation if the
various notions were all varying in sync. To illustrate, consider things from
the quantifier variantist’s point of view. He will agree that each object-
theoretic notion can be (equi-structurally) varied. But variation of one
requires correlated variation of the others. You can’t change what you mean
by the ordinary singular quantifiers, say, without also changing what you
mean by the plural quantifiers and predicates. This constrained variation
needs to be explained. And it would be natural to explain it by saying the
following. First, there is an ur-notion of objecthood—of how to “carve the
world up into objects”. We have an arbitrary choice of how to understand this
ur-notion. And this ur-notion acts as a parameter for all the object-theoretic
notions. Once we select a value for the parameter (a way of carving the world
up into objects), we thereby fix meanings for all the interconnected notions:
the quantifiers, both singular and plural, and the predicates. Second, every
predicate and quantifier has a “hole” in it, corresponding to the parameter.
For example, ∃ is really ∃X, where X is filled in with a value for the
parameter.67 The explanation of the constrained variation is then this: an
object-theoretic notion can be varied only by varying the parameter, which
then induces a corresponding variation in all the other object-theoretic
notions.
But for an ontological realist, there is no constrained variation to be
explained. And so, there seems to be no reason to posit the “holes” in object-
theoretic notions. Positing such holes would be positing a parameter-slot for
which there’s only one value. And if the holes aren’t posited, then it isn’t
clear what explanatory gain is to be had by positing the parameter—by
positing a joint-carving notion of objecthood.
Third, there is an argument for recognizing fundamental individualistic
facts— again, in addition to fundamental quantification rather than instead of
it—based on completeness (section 7.1). If fundamental languages have only
quantified sentences, and no names, then how will we give a metaphysical
semantics for individualistic statements of nonfundamental languages, such
as ‘Socrates is snub-nosed’?
But this argument, too, can be resisted. For we can give a purely general
metaphysical semantics for a language with names: each named object could
be associated with descriptive identifying conditions, and the metaphysical
semantics could translate names as appropriate Russellian descriptions. This
is not to lapse into early twentieth-century philosophy of language. For
metaphysical semantics, recall, is distinct from the linguistic semantics
pursued by linguists and philosophers of language. Descriptivist metaphysical
semantics is compatible with antidescriptivist linguistic semantics. “Natural
language proper names are never synonymous with definite descriptions”
might well be true in the metalanguage of Kripkean linguists, even if that
metalanguage has a descriptivist metaphysical semantics.68

9.14 NonQuinean first-order ontology

The ontological question, Quine (1948) says, is that of what there is. Quine
belittles the fictitious Wyman (a proxy for Meinong), who thinks of the
ontological question as being instead: which of the many things that there are
have a further distinction of being actual? Other neoMeinongian terms for
this or related further distinctions include ‘exists’ and ‘subsists’.69 According
to a Quinean approach to ontology, what the opponent of holes denies is that
there are holes; but according to neoMeinongians, the opponent of holes
concedes that there are holes while denying that holes exist (are actual,
subsist, etc.) According to this picture there are holes, numbers, unicorns,
fictional characters, and round squares; it’s just that many (or perhaps all) of
these objects do not exist.
Quine and many following him regard (neo)Meinongianism as being
conceptually confused.70 The distinction between what there is and what
exists simply isn’t there to be made. Now, I think that, as a matter of
metaphysics, this criticism is correct: there are no joints in reality
corresponding to the neoMeinongian ideology. But in what sense is the
neoMeinongian conceptually confused? A neoMeinongian form of
ontological realism could hold that quantifiers and the predicate ‘exists’ both
carve at the joints. Reality includes both quantificational and existential
structure. There are two sorts of fundamental ontological facts: facts about
what objects there are, and facts about which objects exist. The division
between existent and nonexistent objects is distinguished from other
fundamental divisions (other joint-carving properties) by being more abstract.
There are charged particles on either side of the division, spatial relations
hold on either side of the division (though perhaps not across it, depending on
what we want to say to Quine about fat men in the doorway), and so on.
What is conceptually confused in this picture? It makes a clear choice of
fundamental ideology, and says clear things in terms of that ideology. I do
think the picture is wrong; Quine’s ideology is metaphysically superior. But
the thought that the neoMeinongian is confused seems to be a product of the
magical-grasp picture of understanding rejected in chapter 2. One can fault
the neoMeinongian for inadequately explaining the theoretical role that his
ideology is supposed to play, or for giving bad arguments for his theory; but
we shouldn’t fault him simply for being different.
Other nonQuinean conceptions of ontology are possible. Rather than a
quantifier + predicate fundamental ideology, one could instead have multiple
quantifiers. Kris McDaniel (2009; 2011a; b) and Jason Turner (2010b) have
explored the idea that there are different kinds or modes of being, which they
understand as the idea that there are two or more joint-carving quantifiers.71
Shamik Dasgupta (2009a) has explored the idea of an “objectless”
fundamental ontology, according to which fundamental languages do not
have quantifiers at all, but instead have versions of Quine’s (1960b) predicate
functors. Various people have flirted with rejecting the Fregean “bare”
quantifier in favor of a quantifier with a slot for a sortal predicate—one
cannot ask whether there is a ϕ; one must always ask whether some F is a ϕ,
for sortal predicate F.
There is a lot going on behind the scenes, in orthodox Quinean ontology.
Choices have been made, both about the ideology in which to pose the
ontological question, and about the methodology used to answer the question.
These choices are not inevitable, and they embody a tacit, substantive
metaphysical commitment concerning the fundamental structure of reality, to
the view that the joint-carving ontological notions are the quantifiers of first-
order logic. This is the unacknowledged foundation of contemporary
ontology. It is a good foundation I think. But it would be better to
acknowledge the commitment.

9.15 Higher-order quantification

Quantificational extensions of first-order logic include the addition of
second-order quantifiers, plural quantifiers, and generalized quantifiers such
as ‘most’ and ‘there are infinitely many’. These additions add expressive
power: given a fixed stock of predicates, one cannot define the new
quantifiers in terms of the first-order quantifiers ∀ and ∃.
It is common to do model theory for these new quantifiers in a first-order
set-theoretic metalanguage. In the model theory for second-order logic, for
example, the second-order quantifiers range over sets of n-tuples of the
domain, and the second-order predication Xy1 …yn is taken to be true
(relative to a variable assignment) iff the n-tuple of the referents of y1 … yn is
a member of the referent of X. But philosophers sometimes claim that the
new quantifiers can be interpreted “innocently”, as not meaning their first-
order set-theoretic interpretations, but rather as expressing sui generis new
contents.72 Interpreted innocently, the second-order sentence ∃X∀y Xyy is
not supposed to mean that there exists a set containing the ordered pair 〈y, y〉
for each object y. Nor is it supposed to mean that there exists a relation such
that for every y, the relation is instantiated by y and y. Under the innocent
interpretation, it does not make a claim about sets or relations or any other
things at all. What does it mean? Well, it means ∃X∀y Xyy!
Many questions about innocent second-order quantification (and innocent
plural quantification, etc.) have been asked: whether it is legitimate, whether
it really commits one to sets, whether we understand it, whether it is
intelligible, and so on. But in addition to these methodological, normative,
and psychological/semantic questions, there is also a squarely metaphysical
question: do second-order quantifiers carve at the joints? Does reality contain
second-order quantificational structure?
Anyone who thinks that reality does contain second-order quantificational
structure is in a position to affirmatively answer the methodological,
normative, and psychological/semantic questions. Imagine, for example, a
nominalist who believes in second-order quantificational structure, and who
introduces second-order quantifiers and variables by i) specifying the new
symbols’ grammar and envisaged inferential role; and ii) stipulating that they
are not to be understood as quantifiers over sets or relations or anything of
the kind. She can then reasonably claim, first, that nothing has gone wrong
methodologically, since her introduction of the second-order quantifiers is on
a par with the positing of electrons. Such ontological posits are justified by
their theoretical payoff; her ideological posit should be evaluated in the same
way. And she can reasonably claim, second, that in the following sense she is
not committed to sets: if her overall position is correct, then her second-order
quantifiers are meaningful despite the nonexistence of sets. For her overall
position includes the claim that reality contains irreducibly second-order
quantificational structure; and given that claim, her stipulations surely
succeed in giving a non-set-theoretic meaning to her second-order
quantifiers. And she can reasonably claim, third, to grasp her second-order
quantifiers in the way we grasp all theoretical terms: by understanding their
inferential role. The thought that the second-order quantifiers would be
psychologically/semantically suspect seems rooted in the misguided magical-
grasp model of understanding (section 2.1).
The nominalist’s introduction of the second-order quantifiers is
methodologically, normatively, and psychologically/semantically faultless.
Still, something may well have gone wrong. If reality lacks second-order
quantificational structure (and also lacks an ontology of sets—see below),
semantic disaster will strike the nominalist’s attempt to stipulate a meaning
for her second-order quantifiers. Those quantifiers may be semantically
indeterminate, or even semantically defective. Perhaps “disaster” overstates it
since some second-order sentences may have their truth-values settled by the
stipulated inferential role. For example, if the nominalist stipulates that ∃Y
Ya is to be true whenever Fa is true, then if some sentence Fa is indeed true,
perhaps ∃Y Ya becomes true as well. But second-order sentences whose
truth-values are not thus settled will presumably be semantically
indeterminate.73
Suppose reality lacks second-order quantificational structure but includes
an ontology of sets. Then the true metaphysical semantics might interpret the
second-order quantifiers as first-order quantifiers over sets (that might be the
only way for second-order sentences to be determinate). Thus interpreted, the
second-order quantifiers would “metaphysically commit” their users to sets.
But as we saw earlier, metaphysical commitment barely deserves the name.
(We are also metaphysically committed to there being subatomic particles by
saying “There are tables” if its metaphysical truth-condition quantifies over
subatomic particles; but that doesn’t make the denial of subatomic particles
irrational.) What is important is that the nominalist holds a coherent overall
position according to which second-order claims can be true, and
determinate, despite the nonexistence of sets.
Coherent, yes; but reasonable? The addition of second-order quantifiers to
fundamental ideology immensely increases the complexity of one’s
fundamental theory, and should be undertaken only to gain some great
explanatory benefit—to fundamental theory. Here again the term ‘ideology’
is unfortunate, leading us to focus on questions of psychology and semantics.
To posit the second-order quantifiers as fundamental ideology is to go out on
a metaphysical limb, to posit that the world contains the requisite structure.
Why think that the world contains that structure? Why not retain, at the
fundamental level, a simpler quantificational ideology, that of good old first-
order logic?
Many of the arguments in favor of innocent higher-order quantification do
not look powerful when taken metaphysically, as arguments for the
conclusion that second-order quantifiers carve at the joints. (This is not a
criticism; the arguments typically aren’t intended in this way.) For example,
in their defense of innocent second-order quantification, Agustín Rayo and
Stephen Yablo (2001) argue that ordinary English already contains such
quantification, extending Arthur Prior’s (1971, p. 37) example of “I hurt him
somehow”. Suppose they are right; and suppose further that, as they argue,
English second-order quantifiers are innocent in the sense that using those
quantifiers while denying that one is quantifying over sets does not
compromise one’s semantic competence. It’s hard to view this putative datum
as supporting the idea that second-order quantifiers carve at the joints.74
English second-order quantification might be—like much natural-language
talk—metaphysically second-rate. It might have a complex or disjunctive
metaphysical semantics, and it might be semantically indeterminate.
Alternatively—and I suspect this is indeed true—English second-order
quantifiers might have a first-order set-theoretic metaphysical semantics. The
fact that a competent speaker can coherently think “I hurt him somehow, but
there do not exist sets” does not rule out a set-theoretic metaphysical
semantics, any more than the fact that a competent speaker can coherently
think “There is a chair but there do not exist electrons” rules out a
metaphysical semantics based on modern particle physics. The way we, or
our brains, conceptualize ‘There are tables’ and ‘I hurt him somehow’ is one
thing, and the underlying metaphysics of these matters is quite another;
there’s no reason to take the former as a guide to the latter.
A famous argument due to George Boolos (1984) for irreducibly plural
quantification is similarly unconvincing if taken as concerning
metaphysically fundamental ideology. (Again, this is not a criticism; Boolos
didn’t intend his argument in this way.) Ordinary English speakers say many
things that cannot be regimented in first-order logic without using set-
theoretic predicates. There is the Geach-Kaplan sentence, for example: ‘Some
critics admire only one another’. Moreover, ordinary English speakers say
other things that are best represented in non-first-order terms, even though
their truth-conditions can be given in first-order terms; for example: “There
are some monuments in Italy of which no one tourist has seen all.”75 Boolos
concludes that i) the best formal regimentation of such sentences is into the
language of monadic second-order logic; ii) the sentences of monadic
second-order logic may be interpreted by translating them into a (somewhat
augmented) fragment of ordinary English containing plural quantifiers; and
iii) the use of the existential plural quantifier “There are some things such
that…” in this fragment of English does not commit (his word) one to sets.
For example, we may regiment the Geach-Kaplan sentence in second-order
terms as follows:

We may then translate this into plural English as follows:

There are some things such that i) there is at least one of them; ii) each of
them is a critic; and iii) anything that one of them admires is distinct from
that one of them, and is itself one of them.

Finally, according to Boolos, this final sentence, and other sentences in which
we quantify plurally over critics, monuments, and other nonsets, do not
commit one to sets. As he memorably puts it (1984, p. 448), “It is haywire to
think that when you have some Cheerios, you are eating a set—what you’re
doing is: eating THE CHEERIOS.”76
The questions Boolos was addressing were i) how should we
“represent”— “symbolize”, “regiment”—certain sentences in English; and ii)
do English plurally quantified sentences “commit” one to sets? I suspect that
his answers are correct, given a natural way of taking his questions. There is
indeed a certain level of logico-linguistic representation of English
(connected with capturing natural-language logical relations) at which the
Geach-Kaplan and other sentences should be given second-order
representations; and I also agree that there’s a good sense in which English
speakers don’t commit themselves to sets when they utter plural sentences.
But none of this tends to show that plural quantification carves at the joints.
First, none of it conflicts with the idea that the metaphysical semantics for
English plural quantifiers is second-rate in one or more ways: it could be
complex and/or disjunctive and/or semantically indeterminate. Second—and
again, this is where my own money is—the underlying metaphysical
semantics for English plural quantification might be set-theoretic. This
wouldn’t conflict with Boolos’s claim that English plurally quantified
sentences don’t “commit” their users to sets. Compare: even if the right
metaphysical semantics for “There is a table” refers to subatomic particles,
there’s a perfectly straightforward sense in which normal English speakers
aren’t saying that there are subatomic particles when they utter that sentence,
and a perfectly straightforward sense in which they don’t commit themselves
to subatomic particles when they utter it. Similarly, even if the right
metaphysical semantics for the Geach-Kaplan sentence is set-theoretic,
ordinary speakers aren’t saying that there are sets when they are uttering it.
Nor are they committing themselves (in the only sense that matters) to sets:
uttering the sentence while disbelieving in sets is neither irrational nor a sign
of linguistic incompetence.
A different argument for irreducibly plural quantification, given by
Boolos and others, fares a bit better—but only a little bit—if taken as
concerning metaphysically fundamental plural quantification. Ordinary
English speakers who believe that there exists at least one F are generally
disposed to accept: “There are some things such that they are all the Fs.” So,
ordinary English speakers who believe in sets are disposed to accept:
(AS) There are some things such that they are all the sets.
But given standard ZF set theory, there is no set of all the sets; thus we
cannot interpret the plural quantifier in (AS) set-theoretically if we want (AS)
to be true.
There is no significant pressure here to admit metaphysically fundamental
plural quantification. First, plural sentences might have non-set-theoretic but
metaphysically second-rate metaphysical truth-conditions. Second, the
metaphysical semantics of English plural quantification might be
nonuniform: some plurally quantified sentences might have set-theoretic
truth-conditions while the metaphysical truth-condition of (AS) might be
simply that there exists at least one set. (The metaphysical truth-condition of
any sentence of the form “There are some things such that they are all the Fs”
could be that there exists at least one F.) Compare the truth-conditions for
ordinary English first-order sentences. Many of them may have quite
“straight” metaphysical truth-conditions. For example, the ordinary English
“There is an electron” may well have the metaphysical truth-condition that
there exists, in the fundamental sense, an electron. But it’s unlikely that all
first-order English sentences have such truth-conditions; think of A man, a
plan, a canal; Panama!”, ‘There are five ways to win this chess match’, ‘My
sock contains a hole’, ‘Every smirk disappeared from every face at the return
of the exams’, and so on. It seems likely that the metaphysical truth-
conditions will be nonuniform across the range of first-order English
sentences. (Remember, these are metaphysical truth-conditions; the idea isn’t
that competent speakers know about all this.) Third, (AS) might be false,
ordinary belief in its truth be damned.
It may be objected that each of these alternatives is unpalatable, and hence
that we would be better off positing irreducibly plural quantification. But it
takes more than a casual observation about what we ordinarily say and
believe to justify a truly dramatic complication of our fundamental theory of
the world. The argument from (AS) to fundamental plural quantification
doesn’t look much better than the old argument from the apparent truth of
“John and Ted share many vices” to the fundamental existence of universals
as abstract entities. Note also that the objection depends on two attitudes that
are hard to sustain simultaneously: squeamishness about error-theories and
squeamishness about nonuniform metaphysics. Squeamishness about error
theories is most compelling when the terms in question are everyday rather
than theoretical, so that we’re entitled to confidence that paradigmatic uses of
sentences containing them are true. (No one contemplates an error theory
about our belief that Newton didn’t lie when he said that space and time are
absolute; ‘lie’ is an everyday term.) But there’s nothing wrong with a
nonuniform metaphysical semantics for such terms (which is largely why the
confidence is justified). It’s only for theoretical terms (‘charge’, ‘gold’) that
we think a good metaphysical semantics must be uniform; but then we’re
more open to error theories.
A far more convincing strategy for defending fundamental plural
quantification would be to argue that it is required for some important
theoretical purpose. And a further argument of Boolos’s may seem to do this:
the argument that we need plural (or second-order) quantification to give an
adequate axiomatization of ZF set theory. The standard first-order
axiomatization of ZF set theory contains as axioms all of the infinitely many
instances of the first-order replacement schema. A strictly stronger, second-
order, formulation dispenses with the infinitely many first-order axioms in
favor of a single second-order sentence. Boolos shows that the second-order
quantifiers in this second-order sentence cannot be regarded as short for first-
order quantifiers over sets. Our choice is therefore between a first-order
axiomatization of ZF and a second-order axiomatization in which the second-
order quantifiers are either taken as primitive or—Boolos’s preferred
alternative—interpreted plurally. Boolos then goes on to argue that the first-
order axiomatization is insufficient.
… to rest content with a set theory formulated in the first-order predicate calculus with identity … must
be regarded as a compromise, as falling short of saying all that we might hope to say. We accept [the
first-order formulation of ZF] because we accept a stronger theory consisting of a finite number of
principles, among them some for whose complete expression second-order formulas are required. We
ought to be able to formulate a theory that reflects our beliefs. (1984, p. 441)
At first glance the argument looks very weak (again, if taken as
concerning metaphysical fundamentality—which is not what Boolos intended
by this argument). “We ought to be able to formulate a theory that reflects
our beliefs”: in the present context this would have to mean that the beliefs in
question have fundamental plural content, that they be stated in terms of
fundamental plural quantifiers. But why suppose this? The fundamental
contents of our beliefs are not in general transparent to us. On my view,
neither causation nor modality is fundamental; on Hirsch’s view,
quantification is not fundamental; neither I nor Hirsch can be refuted simply
by pointing to the fact that people have causal, modal, and quantificational
beliefs.
However, the argument can be formulated so as not to concern our beliefs
at all; it can be formulated as the claim that second-order ZF is a better
theory. We generally prefer unified theories, theories that explain diverse
phenomena using a small number of posits. First-order ZF posites a
disunified infinite array of facts, the infinitely many instances of the
replacement schema. By replacing this infinite array with a single principle, it
may be thought, second-order ZF is more explanatory. First-order ZF doesn’t
have the right kind of “laws”; second-order ZF does. But is our overall theory
made simpler by introducing the higher-order quantifiers?77 Our laws of set
theory no longer include the replacement schema, but our laws of logic now
must include the plural comprehension schema:

And we must make use of instances of this schema to draw consequences

from our second-order set theory. So the argument seems inconclusive.
The final argument I’ll consider—the best I know—for fundamental
higher-order quantification derives from Timothy Williamson’s (2003)
argument that we need second-order quantification to state semantic theories
for languages with absolutely unrestricted quantifiers. Standard semantic
theories introduce entities known as interpretations, and then use those
interpretations to define various familiar semantic and logical concepts. (For
example, a set of sentences G is said to logically imply a sentence S iff S is
true in each interpretation in which every member of G is true.) But, as
Williamson shows, one can derive an analog of Russell’s paradox for
interpretations. Williamson’s solution is to adopt a different approach to
semantics that isn’t subject to the paradox. On this approach one does not
introduce interpretations as entities at all. Rather, one uses primitive second-
order quantification (with dyadic predicate variables) to formulate a second-
order notion of an interpretation, with which one can formulate second-order
analogs of familiar definitions of logical consequence and other semantic and
logical notions. Thus, Williamson argues, we need second-order
quantification in order to do semantics without falling into paradox. And
finally, one might add that the second-order quantifiers must be fundamental
if the logical and semantic notions we define using them are to be fully
determinate.
The argument here is formidable (though one worries it will demand still
higher-order ideology for defining interpretations of the second-order
language). But note that the demand for extra ideology is coming from
linguistic theory and metalogic. Elsewhere we resist demands from high-level
sciences to add significant complexity to our fundamental account of the
world. Such resistance is exactly what drives familiar projects to naturalize
mental and semantic content. As Fodor (1987, p. 97) puts it:
I suppose that sooner or later the physicists will complete the catalogue they’ve been compiling of the
ultimate and irreducible properties of things. When they do, the likes of spin, charm, and charge will
perhaps appear upon their list. But aboutness surely won’t; intentionality simply doesn’t go that deep.

Williamson might reply that we need the extra ideology to do logic, not
semantics. But notice that the ideology is needed, at most, for metalogic, for
giving fully general theories of such metalogical notions as logical
consequence. It is not needed for the use of logical notions (such as
conjunction, quantification, and so on), within science and mathematics. It is
hard to give up on beautiful metaphysics for the sake of metalogic.
 1 Penrose (2005, p. 711); Frost (1936); USAF Maxwell Blue Book 1, p. 8; Sagan (1980, p. 4).
2 See Carnap (1950); Chalmers (2009); Hirsch (2002a; b; 2005; 2007; 2008; 2009); Hofweber
(2009); Putnam (1987); Sidelle (2002; 2010); Sosa (1999); Thomasson (2007; 2009).
3 Hirsch (2005, p. 67). Hirsch is a deflationist about only some of philosophical ontology.
4 Compare the indeterminacy strategy of section 5.1, and see Chalmers (2009).
5 Compare the common-sense strategy of section 5.1, and see Hirsch (2005).
6 The second deflationist would presumably turn into the first if ontologists were to enter the
philosophy room, and introduce new ontological vocabulary stipulated to carve at the joints (section
5.3).
7 Although see Weatherson (2003).
8 Another connection to Quinean methodology: why follow Quine’s advice to regiment in first-
order terms before assessing ontological commitments? One possible answer is that i) first-order
quantifiers carve at the joints; ii) second-order quantifiers do not (section 9.15); and iii) ontological
commitments are given in terms of joint-carving quantifiers (section 9.12).
9 Compare section 3.3.
10 Further, if ontological language doesn’t carve at the joints, then why think that positing more
entities would be worse? Occam’s (ontological) razor is based on the thought that “emptier”
possibilities are prima facie more probable than “full” possibilities. But the measure of the fullness of a
possibility should depend on its description in fundamental terms; if ontological language doesn’t carve
at the joints, then a possibility with more entities might not be fuller in the relevant sense.
11 I take this use of ‘doctrine’ from Fine (2005).
12 Compare Quine (1948), though he would reject the backdrop of realism about metaphysical
structure. The first formulation is in first-order logic, the second in English; but English quantifiers
have a different syntax from those of first-order logic. I’ll take the official formulation to be the first,
though I’ll often paraphrase in English.
13 As it is according to Jonathan Schaffer (2009a).
14 I defend a similar view, except that I accept sets in addition. See chapter 13 and Sider (2011).
15 Compare Dorr (2007, section 1).
16 See Sperber and Wilson (1986); Wilson and Sperber (2004).
17 To be sure, some ontological disputes are nonsubstantive or equivocal because of their predicates
— for instance, a dispute over whether “God” exists in which some mean by ‘God’ the center of mass
of the universe. But this is not so in the disputes to be discussed.
18 Could the crucial expression be the predicate ‘is a thing’? In that case, the deflationist would
have to admit that a metaphysical dispute could be reinstated simply by recasting the debate as being
over whether there exist tables at all, as opposed to tables that are things. Van Inwagen and his
opponents would be happy to rephrase things in this way, since that’s how they understood the debate
in the first place. See Williamson (2003, p. 420).
19 See Hirsch (2002a; b; 2005; 2007; 2008; 2009), and also Putnam (1987) and Eklund (2008).
20 Dorr (plausibly) assumes that not all “counterpossible” conditionals are true.
 21 I specified the semantics of L and L by giving truth-conditions for their sentences using
U N
English as a metalanguage; but the quantifier variantist might instead specify partial, implicit
definitions governing quantifiers within alternate languages, as in Båve (2010).
22 Compare Hirsch (2002b, p. 53).
23 Whether this is conceived semantically or pragmatically is irrelevant here.
24 Even a defender of “indefinite extensibility” should allow that we can disavow all restrictions
except those concerning indefinitely extensible domains; this would allow for more substantive
ontology than a quantifier variantist will want to admit.
25 Some of the issues in this neighborhood are discussed in Sider (2007a).
26 One might do it plurally instead, taking ‘There is a table’, for example, to mean that there
qe
areqc some things that are arranged tablewise.
27 For this reason, I believe this type to be closer to Hirsch’s own position than the second.
28 This might require a commitment to primitive comparative joint-carving. For otherwise, the
claim would have to be that each qi has a more direct basis in the perfectly joint-carving notions than its
restrictions; but what conception of the perfectly joint-carving notions could the quantifier variantist
adopt that would sustain this? (See also section 9.6.2.)
29 See Sider (2007a; 2009).
30 Eklund and Hawthorne focus on atomic sentences, but Small must also reject the Tarskian clause
for quantified sentences. Small thinks that ‘∃x Table(x)’ is true in Biglish, but is unwilling to assert
“There exists something that is in the extension-in-Biglish of ‘Table’.”
31 Plausibility argument: pretend that giving meaning to a language is just a matter of describing its
intended model. Models are described using quantifiers in the metalanguage. One uses metalanguage
quantifiers to specify a domain, which fixes the meaning of the object-language’s quantifiers; and one
uses metalanguage quantifiers to give the meanings of object-language names, predicates, and function
symbols (a constant means an object in the domain; a predicate means a set of tuples from the domain;
a function symbol means a function on the domain). So if one changes the meanings of the
metalanguage quantifiers, one would change the meanings of the object-language’s quantifiers, names,
predicates, and function symbols.
32 This is so for a semantic conception of grammatical categories, anyway. On a purely syntactic
conception, the Tarskian ideas lose force.
33 Compare Hirsch (2002b, p. 57).
34 Not much has been written on this, though see Hawthorne and Cortens (1995); Turner (2008).
35 At least, after the physical theory has been appropriately formulated.
36 This point is regularly missed, for example by those who claim that modern physics has no need
for “objects”. Perhaps modern physics has no use for particles, but this doesn’t show that it has no use
for quantifiers.
37 Ned Markosian (2004a) distinguishes between stuffand things, accepts the existence of both, and
assigns them different roles in a theory of material objects (for example, stuff obeys unrestricted
composition and mereological essentialism, whereas things do not). My criticisms of the stuff defense
of quantifier variance do not apply to Markosian, since he is not trying to replace thing language with
stuff language, nor is he trying to avoid substantive questions about either things or stuff
 38 Turner (2010a) discusses a related proposal critically Dasgupta (2009b) appeals to the predicate-
functor formulation, but in defense of the idea that a permutation of individuals across qualitative roles
doesn’t change the world, rather than in defense of quantifier variance, so my criticism does not apply
to him. See also Burgess (2005).
39 “… we can retain the notion of an unstructured fact. I think this is indeed our most basic notion
of ‘reality’, ‘the world’, ‘the way it is’, and this notion can remain invariant through any changes in our
concept of ‘the things that exist’.” (2002b, p. 59)
40 The argument is defeasible; ontological pluralists might argue that the balance of evidence favors
their view. (Here it’s important that the common belief in the incoherence of ontological pluralism is
mistaken, as McDaniel (2009) and Turner (2010b) have effectively argued.) Note that ontological
pluralism would not reinstate deflationism; rather, it would split hitherto univocal ontological questions
into multiple questions, each as substantive as the original.
41 And we can go much further: the entirety of second-order arithmetic can be derived in second-
order logic from Hume’s Principle. See Wright (1983); Hale (1987); Hale and Wright (2001).
42 See Thomasson (2007; 2009). There is much else in this broad-ranging and careful work that
deserves comment; I’ll indulge in a few all-too-brief remarks. 1. Thomasson writes as if the “qua-
problem” for the causal theory of reference argues for her view: in order for singular and sortal terms to
have determinate reference, i) they must be governed by analytic descriptive conditions; and ii) these
must have the form of what she calls “application” and “coapplication” conditions. But the opponent of
easy ontology can grant i) while denying ii): the conditions might instead have the form of conditions
that select which entities we are referring to. For example, the name ‘Orky’ could be governed by the
condition ‘Orky’ refers to the animal in such-and-such a location (if there is no such animal then
‘Orky’ refers to nothing); and the sortal term ‘animal’ could be governed by the condition ‘animal’
refers to something iff it has such-and-such features and thus-and-so persistence conditions. This is
consistent with denying that analytic principles secure the existence of things with the specified
features and persistence conditions. 2. Thomasson regards sortal-relative vs. bare quantification as a
battle-line between her and her opponents. But a friend of easy ontology could regard the relevant
analytic sentences as governing the bare quantifier. In some cases, the sentences would involve no
sortal predicates at all; in others, they would. For example, the analogs of Thomasson’s application and
co-application conditions would be analytic sentences of the forms “∃ x (F x ∧ ø (x,t))” and“∃ x (F x
∧ ø (x, t) ∧ ψ (x, t′))”, respectively, where F is a sortal predicate and t and t′ name times. Conversely,
a foe of easy ontology could argue that the bare quantifiers do not carve at the joints; what does carve at
the joints is a quantifier that has a “slot” for a sortal predicate. 3. Thomasson doubts that her opponents’
quantifiers could have determinate meanings; why? Because, I think, there is no consensus amongst her
opponents over cases (over, e.g., whether ‘There exist tables’ is true) or useable rules (over, e.g.,
whether this rule holds: “‘There exist tables’ is true iff there exist things arranged tablewise”). (By a
“usable” rule I mean—vaguely—one that a linguistic community as a whole, perhaps through a
subcommunity of experts, could knowledgeably apply Everyone can agree on the homophonic rule
“‘There exist tables’ is true iff there exist tables”; but this isn’t useable, since Thomasson’s opponents
disagree over whether the right-hand side holds.) However, realists quite generally say that terms can
have determinate meanings despite disagreement over cases and useable rules of use (think of
theoretical terms in areas of physics in which there is controversy). Why couldn’t the same hold for the
quantifiers? (Reference magnetism is one model of how quantifiers might nevertheless be semantically
determinate.)
43 See in this vein Boghossian (1997).
44 See Boghossian (1997, section II).
 45 I do not say that the relevance of use to meaningis exhausted by its determination of the
definitional sentences.
46 See also Eklund (2002); Tappenden (1993).
47 In contrast to the more usual, semantically motivated, sort, such as Dummett (1973).
48 See Harman (1999, chapters 5–7).
49 It is no answer to the skeptic to say, as Paul Boghossian (1997; 2003) does, that since skeptical
doubts are phrased using logical concepts, if our logical constants lacked meanings then we could not
even entertain those doubts—the sentences expressing the doubts would fail to express propositions.
The fact—if it is a fact—that the doubts are unentertainable if true is peculiar but not probative.
(Eliminativism about propositional attitudes or people (Churchland, 1981; Unger, 1979) also cannot be
doubted if true, but so what?) We cannot, for example, combine the fact with the additional premise
that we clearly are capable of entertaining the doubts, and thus conclude in Cartesian fashion that the
skeptic is wrong; for the added premise is dialectically inappropriate. To “entertain” the doubts in the
relevant sense requires bearing the propositional attitude of doubt towards skeptical propositions; and
our ability to doubtingly wield skeptical sentences is no guarantee that the skeptical sentences express
propositions.
50 I’ve been describing easy ontology as an alternative to quantifier variance; but the two can be
combined. On this combination, the function of laying down (T) as a definitional constraint is to select
one of many quantifier-candidates to be meant by ‘there exist’. Sider (2007a) describes in more detail
the combination of neoFregeanism about mathematical ontology with quantifier variance. Note also
that even given quantifier variance, the easiness of ontological questions is only modulo knowledge of
quantifier variance itself—see section 9.6.3.
51 See Hofweber (2005, section 4.3) and (2007, section 6.4).
52 Three further points here. (i) An ontological realist might take Hofweber’s arguments to show
that internal uses of the quantifiers do not carve at the joints. Such arguments based on Hofweber’s
linguistic analysis of the internal use of quantifiers must be distinguished from the metasemantic
argument for non-joint-carving quantifiers considered in section 9.3. (ii) An ontological realist
following Hofweber might say that external quantifiers in English carve at the joints. But she might say
instead that, for metasemantic reasons, even these fail to carve at the joints. For even if external
quantifiers don’t carve at the joints, they might still relate to the world in a different way from how
internal quantifiers do, a way that lets us justifiably say that they are about objects that are “out there”.
(Such an ontological realist would need to do ontology in Ontologese.) (iii) A quantifier variantist
might accept Hofweber’s arguments, and thus distinguish between two types of non-joint-carving
quantifiers.
53 Armstrong does say that if neither of two objects is an addition of being to the other, then the
objects are identical.
54 I argue against this sort of Mooreanism, however, in Sider (2011).
55 There are real questions about how exactly to formulate maximalism; but set them aside.
56 For more on this see Sider (2007a, sections 3 and 4).
57 This way of thinking might, incidentally, be adopted by Schaffer in place of the Mooreanism.
58 Compare this analogous argument: since Lewis’s pluriverse contains myriad “alien” natural
properties and relations, modal realism sins against Ockham’s razor in a big way. Perhaps Lewis could
respond that his pluriverse has a corresponding simplicity since it lacks arbitrariness—all possible
natural properties and relations, so to speak, are present. (Compare Bennett (2004, section 4).) This
raises interesting issues about how to evaluate simplicity
59 Phillip Bricker (1992) gives an interesting defense of maximalism both about entities and about
natural properties and relations, within the realm of mathematics. An alternative is combining a sort of
maximalism about mathematical entities with a much smaller, solely logical, albeit higher-order,
ideology, as in Lewis’s (1991) set-theoretic structuralism.
60 Analogous points could be made about commitment via the true linguistic semantics.
61 See section 11.8 on grammatical categories carving at the joints.
62 If there are infinitely many things this would require an infinitary disjunction to hold in reality.
63 The Humean theory of chapter 12 is a form of modal deflationism.
64 If the latter is the metaphysical truth-condition of the former, then infinitary disjunction must be
fundamental.
65 Contrast the rephrasal of the argument concerning the source of structure in section 8.3.1.
66 Jason Turner’s (2010a) notion of pegboard structure is a good metaphor here.
67 An intriguing possibility would be for a quantifier variantist to say that these meanings with
holes do carve at the joints. Does this allow a reply to the argument of section 9.6.2?
68 Dasgupta (2009a) defends a “generalist” metaphysics in a different way against this sort of
objection, and also gives an interesting positive argument in favor of generalism. See also Dasgupta
(2010).
69 Recent neoMeinongian work includes McGinn (2000, chapter 2); Parsons (1980); Routley
(1980).
70 Compare van Inwagen (1998).
71 This needs to be distinguished from the idea that there is just one fundamental quantifier which
may be restricted by multiple fundamental predicates. I will not take up this issue here (McDaniel and
Turner have much to say about it), except to say that quantifiers grammatically are very different from
predicates since they bind variables; thus, the fundamental facts, intuitively, have a very different shape
if there are two fundamental quantifiers.
72 See, for example, Boolos (1984); Rayo and Yablo (2001).
73 Timothy Williamson (1994) says that there’s a particular hair that makes a bald man bald, but
this hair has no particular metaphysical significance. I wonder whether he would hold the analogous
position in this case, e.g., that a second-order sentence corresponding to the continuum hypothesis has a
truth-value, but this truth-value has no metaphysical significance.
74 Moreover, it would be hard to view the falsity of the putative datum as evidence against the idea;
why think that English is maximally metaphysically expressive?
75 The most natural regimentation is ∃X(∃xX x∧∀x(X →Mx)→~∃y(Ty∧∀x(X x→S y x))),
which is equivalent to the first-order sentence ∃x M x ∧ ~∃y(Ty ∧ ∀x(Mx→Syx)).
76 Boolos’s confidence about what we’re eating is unjustified; perhaps the underlying metaphysics
of eating Cheerios is set-theoretic. What’s haywire is to think that we are saying that we’re eating sets
when we say that there are some Cheerios that we’re eating.
77 Thanks to Cian Dorr here.
10 Logic
If one can query expressions from all grammatical categories for joint-
carving, then one can query logical expressions. This leads to interesting
questions about the metaphysics of logic, which bear on substantivity,
objectivity, determinacy, and so forth in logic, and also on the nature of joint-
carving itself.
Questions about joint-carving can be asked about both logical and
metalogical concepts. Logical concepts are those concepts expressed by
logical constants, concepts such as conjunction, disjunction, negation, and
quantification; metalogical concepts are theoretical concepts such as logical
consequence, logical truth, provability, semantic consequence, and the
concept of a logical constant. Questions about joint-carving for logical and
metalogical concepts should be presumed independent unless shown
otherwise.
It may seem odd to ask metaphysical questions about logic. But recall
Russell’s conception of the subject, as differing from other inquiries only by
being more “abstract and general”. This (vague) picture of the continuity
between logic and other disciplines is a crucial presupposition of this chapter
(and book). If a hangover from logical conventionalism leads you to distrust
the questions: pop a couple of aspirins, re-read your Quine (1936) (and
section 6.5), and report back.

10.1 Fundamental logic

First question: do any logical concepts carve at the joints? There is a
powerful argument that at least some of them do. The best guide to joint-
carving is a Quinean criterion of ideological commitment: it is (defeasibly)
reasonable to regard indispensable ideology as carving at the joints. But we
cannot get by without logical notions in our fundamental theories. In
particular, since Frege (1879) it has become clear that the notions of first-
order predicate logic are indispensable in serious foundational theorizing.
Section 9.6.4 pressed this argument for the first-order quantifiers. But a
parallel argument suggests that the identity sign and the sentential
connectives of propositional logic also carve at the joints: these are as
indispensable as the quantifiers.
The case for joint-carving is much weaker for certain other “logical”
notions. Modal operators and nontruthfunctional conditionals, for example,
seem dispensable in fundamental theorizing, even though they are deeply
embedded in our ordinary conceptual scheme. The material “conditional” of
first-order logic suffices for conditionality in mathematics and the
fundamental sciences, and neither discipline seems to need modality. Our
best theories are extensional, and talk simply about what is. This is not to say
that there are no modal or conditional facts; it is simply to call for
metaphysical reductions of those facts. And reductions seem possible
—chapter 12 defends a reduction of modality, for example.

10.2 Hard choices

Next question: which logical concepts carve at the joints? I said a moment
ago that the sentential connectives of propositional logic carve at the joints.
But which ones? Just ∧ and ~? Just ∨ and ~? Or perhaps the only joint-
carving connective is the Sheffer stroke |? Similarly, which quantifier carves
at the joints, ∀ or ∃?
You don’t have to be a logical positivist to feel that something is wrong
with these questions. Yet my approach to fundamentality seems to force them
upon us. Structure is structural, and applicable to notions of arbitrary sort; so
we can ask of any logical notion whether it carves at the joints and expect a
substantive, objective answer.1
What’s particularly hard to swallow is the need to choose amongst the
proposi-tional-logic connectives. It’s comparatively easy to swallow the idea
that propo-sitional logic as a whole is metaphysically on-track. But one wants
to treat the choice of which connectives to take as primitive and which to take
as defined as a conventional one (assuming classical logic anyway), whereas
my account elevates this choice to one of high metaphysics.
One might try to avoid the choice by saying that, although it’s
determinately the case that some connectives carve at the joints, it’s
indeterminate which ones do. But section 7.12 argued against indeterminacy
at the fundamental level.
Sometimes one can move to a different, more fundamental, theory to
make hard choices like these go away. Which is fundamental: earlier-than or
later-than? The following theory gets us off the hook: neither is fundamental;
the fundamental temporal relation is temporal betweenness; time is
fundamentally symmetric; the past-future asymmetries derive from certain de
facto asymmetries about how matter is distributed in the universe.2 Or again,
which distance function is the fundamental one? Which function from pairs
of points of space to real numbers is fundamental: the distance-in-meters
function, or the distance-in-feet function, or a function corresponding to some
other unit? We get off the hook by saying: none of them is; the fundamental
metrical facts are facts of spatial congruence.3 Unfortunately, escape of this
sort seems unlikely in the case of logic: what more fundamental theory could
we shift to?
I don’t see how to escape the hard questions, provided we’re willing to
speak of joint-carving at all, since I see no principled way to shield logic
from such questions. And anyway, we face similar hard questions outside of
logic: believers in mereology face the question of whether it is parthood or
overlap that carves at the joints; believers in an intrinsic direction of time face
the question of whether it is earlier-than or later-than that carves; and so on.
So I am led by argument, against inclination, to accept the questions. There is
a real question about which of propositional logic’s connectives carve at the
joints, and similarly for ∀ and ∃.
Here are two points to make this go down easier. First, egalitarian answers
can be given to the questions. For example, one might hold that both ∀ and
∃ carve at the joints, or that all the truth-functional connectives do,4 and thus
avoid drawing invidious metaphysical distinctions.
You might think that a “nonredundancy” constraint is constitutive of the
notion of joint-carving, and that this rules out egalitarianism. Lewis (1986b,
60), for example, says of his perfectly natural properties that “there are only
just enough of them to characterise things completely and without
redundancy”. But the nonredundancy constraint should be rejected.
Suppose first that the nonredundancy constraint is taken modally: no joint-
carving notions can supervene on others. This would prohibit all
mathematical joint-carving notions, given the usual dogma about the
necessity of mathematics, since those notions supervene trivially on any
notions whatsoever. It would also prohibit certain kinds of higher-order
structure. For example, it’s an attractive position for a platonist about
properties to say that higher-order relations generate the determinate-
determinable structure of families of properties, and underlie the use of
numbers to represent quantities (Mundy, 1987). One such higher-order
relation would be a relation ofgreater-than, which holds between two
determinate mass properties m1 and m2 iff objects that instantiate m1 are more
massive than objects that instantiate m2.5 It would be natural for such a
platonist to say that i) the holding of these relations is a matter of necessity
(and so they supervene trivially on any basis whatsoever); and ii) these
relations carve at the joints.6
These problems could be avoided by taking the nonredundancy constraint
in a logical rather than modal sense, so as to prohibit joint-carving
expressions E where every sentence containing E (and perhaps other joint-
carving expressions) is logically equivalent to a sentence not containing E.
But this sort of special pleading is suspect. The examples of mathematical
and higher-order structure show that the redundancy constraint is based on a
misguided idea—the idea that there cannot be “necessary connections”
between distinct joint-carving expressions. Even if facts about numbers bear
the modal relation of necessitation to all other facts (and thus trivially
supervene), they are genuine facts, and involve joint-carving notions; missing
out on such facts would mean missing out on some of reality’s structure.
When ϕ modally entails ψ, the fact that ψ may nevertheless be a distinctive
fact, with a distinctive structure, and so we may wish to say that expressions
in ψ, as well as those in ϕ, carve at the joints. And all of this remains the case
even if the “necessary connection” is a purely logical one. That ψ is a logical
consequence of ϕ does not mean that there isn’t a fact that ψ, or that the fact
that ψ has no structure distinct from the structure of the fact that ϕ. It means
merely that a certain connection holds between the facts: one logically entails
the other. So there is no reason to treat logical redundancy differently from
modal redundancy. And so, since we shouldn’t prohibit modally redundant
joint-carving notions, we shouldn’t prohibit logically redundant joint-carving
notions either.
Rejecting the idea that nonredundancy is constitutive of joint-carving is
consistent with accepting some sort of nonredundancy constraint on
reasonable belief about joint-carving. (This should have been the status of
nonredundancy all along.) In general, we shouldn’t multiply ideology beyond
necessity; and redundant ideology is often unnecessary ideology. Redundant
in what sense? Not a modal or logical sense: the principle shouldn’t
summarily rule against the rationality of adopting fundamental mathematical
or logical ideology. Rather, an explanatory sense: we shouldn’t posit
ideology that is not needed for reaching our explanatory goals. For example,
in a language containing (sentential) disjunction, primitive predicates F and
G, and a further primitive predicate for the disjunctive property of being
either F or G, the latter predicate can simply be deleted without explanatory
loss. This ideological Ockham’s razor is consistent with modally or logically
redundant ideology because the razor, like most constraints on reasonable
belief, can be defeated by opposing considerations. One potential opposing
consideration comes from symmetries: other things being equal, one
shouldn’t draw distinctions between symmetrical hypotheses without good
reason; to do so would be arbitrary. For example, one might regard the
symmetry between ‘earlier’ and ‘later’ (or that between ‘part’ and ‘overlap’)
as a reason to ignore the razor and regard both as carving at the joints. A
second opposing consideration comes from explanatory work done by the
candidates for deletion. Deleting a given bit of ideology from one’s theory
might incur theoretical loss, despite modal or logical redundancy. Simply
deleting mathematical or determinate/determinable-theoretic vocabulary
would (arguably) result in a worse theory, because explanations couched in
the deleted vocabulary would be forfeited. While it is unclear whether the
second opposing consideration is significant in the case of the propositional
connectives (are explanations stated in terms of the Sheffer stroke worse than
those stated in terms of negation, conjunction, and disjunction?), the first
does seem significant, and is the intuitive source of egalitarianism.
So: the availability of egalitarianism is the first point to make the hard
questions go down easier. The second point is that we should not dismiss the
questions simply because of the interdefinability of propositional
connectives, for there is a parallel sort of interdefinability in cases where
there are clearly legitimate questions about joint-carving. Given temporal and
observational vocabulary, ‘grue’ and ‘bleen’ are interdefinable with ‘green’
and ‘blue’; but if talk of joint-carving is ever legitimate, it is surely legitimate
here. Indeed, despite the interdefinability, inegalitarianism is surely right
here: ‘green’ and ‘blue’ carve at the joints, whereas ‘grue’ and ‘bleen’ do
not.7 (Inegalitarianism is hard in the case of the truth functions, but perhaps
not impossible. Perhaps parallels with Boolean algebras in mathematics,
together with a preference for commutativity/symmetry at the fundamental
level, privileges conjunction, disjunction, and negation over the other truth
functions.)
It might be argued that the question of which truth functions are joint-
carving arises only as an artifact of my approach to joint-carving, and would
disappear under other approaches. On my subpropositional approach (chapter
8), the locus of joint-carving is (to put it linguistically) the word; it is words
that do or don’t carve at the joints. One can, as a result, raise meaningful
questions about whether ‘and’, or ‘or’, or both, or neither, carves at the joints.
But suppose instead that the locus is the sentence, or rather, the proposition
(to put it, now, nonlinguistically): it is entire propositions that do, or don’t,
carve at the joints. Then one could not directly raise the question of whether
logical words carve at the joints. One might try to indirectly raise the
question by asking whether propositions expressed using ‘and’, or
propositions expressed using ‘or’, or both, or neither, carve at the joints. But
suppose further that propositions are “coarse-grained”, so that “a is F and b is
G” expresses the same proposition as its Sheffer stroke equivalent: “(a is F | b
is G) | (a is F | b is G)”. This would block raising the question, even
indirectly, of whether ‘and’ or ‘or’ carves at the joints.
But as argued in chapter 8, structure should not be located at the sentence
level. Intuitively put, if “a is F and b is G” carves at the joints, that is because
of something about F-ness, G-ness, and conjunction. Better: “c is F and d is
G”, “e is F and f is G”, and so on, also carve at the joints; this pattern is
surely explained by the recurrence of the joint-carving words ‘F’, ‘G’, and
‘and’.
Further, the motivation for adopting this picture—propositional joint-
carving plus coarse-grained propositions—should be questioned. Is it to
avoid having to answer a hard question about propositional connectives? But
there would remain analogous hard questions in other cases, perhaps with
‘earlier’ and ‘later’, or with ‘part’ and ‘overlap’, depending on how coarse-
grained propositions are taken to be. And there would also remain the hard
question of whether the propositional picture itself is correct, and if so, of
how coarse the grain is. (Propositions must be coarse-grained enough to
avoid the choices we don’t want, but not so coarsegrained to avoid those we
do want, such as the choice between green + blue and grue + bleen.)
There remains a nagging doubt. If questions about joint-carving can be
asked even about logic, then where does it all end? Can we also ask whether
writing in Spanish, or in red ink, or in boldface, carves at the joints better
than the alternatives?8 Surely some differences are just notational!
There are a few different worries one might have here. Consider:
(B) Charge carves at the joints whereas charge does not.
The simplest worry is that I would be forced to admit that (B), or something
like it, is true. But there is just no reason to admit this. ‘Charge’ has no more
explanatory value than ‘charge’ in fundamental physics, for example.
The worrier will regroup: “I see that you needn’t admit that (B) is in fact
true, but you do need to admit that (B) is an epistemically open possibility
(just as you admit as open the possibility that conjunction but not disjunction
carves at the joints).” But do I need to admit this? It would be natural to say
that (B) is the same sentence as the logically contradictory sentence ‘Charge
carves at the joints whereas charge does not’, and therefore is not an open
possibility.
The worry may be put in a slightly different way. The epistemology
proposed in section 2.3 instructs us to search for the most explanatory pair 〈I,
TI〉 of ideology I, and theory TI in terms of I. A natural concern is that there
will be no unique most explanatory pair. My response to the concern is that
when pairs are tied, we should be agnostic which pair is correct (i.e., which is
the pair of the joint-carving ideology and true theory in that ideology). But—
so the worry goes—this will lead to misplaced agnosticism if a new ideology-
theory pair can be produced by changing to Spanish, red ink, or boldface. The
answer to the worry is that these changes needn’t produce new pairs. In a
properly developed version of this epistemology, the ideology-theory pairs
must be formulated in some language, and the sentences of this language will
not be individuated so finely as by language family, font, or ink color.
“Your response to the worry appeals to sentential identity; but that is not a
fundamental notion.” Agreed; but what is the problem? The worry I am
facing is that I must admit that (B) is an open possibility. The crucial notions
in this worry are epistemic, and epistemic notions are not fundamental. And it
is quite natural to think that the nonfundamental matter of how we ought to
form beliefs is intimately tied up with the nonfundamental matter of
sentential identity (or propositional identity if one wanted to put it less
linguistically).
“What constrains the individuation of sentences?” There is no once-and-
for-all answer. We introduce different notions of sentence—with different
associated identity conditions—for different theoretical purposes. For the
purpose of formulating an epistemology, the sentences must be fine-grained
enough to draw the epistemically relevant distinctions, but no finer.
Distinctions based on language family, font, or ink color are epistemically
irrelevant; thus the language for epistemology will identify sentences
differing solely in those respects.
“Why are these differences epistemically irrelevant?” A precise answer
would require a general theory that I do not have about the nature of
epistemic facts; but I can say that the reason is externalist. There are in fact
no joints in nature corresponding to language family, font, or ink color. More
generally, language family, font, and ink color belong to a larger, reasonably
unified class of features, none of which correspond to joints in nature. There
are no joints of nature corresponding to font size, font shape, font weight,
writing versus speaking, volume, timbre, pace, accent, … Given the lack of
such joints in nature (and given the not unrelated fact that none of these
features affects truth-conditional content), there is no point to a conception of
epistemic value that makes distinctions based on factors like these. If
boldface marked a joint in nature, then we would need to individuate
sentences on the basis of font weight; but it doesn’t.
“You appeal to facts about sentence-identity to dismiss questions of joint-
carving for language family, font, and ink color; but couldn’t one do the same
for the hard questions about logic?” I don’t dismiss questions about joint-
carving for language family, font, and ink color. The identification of (B)
with a logically contradictory sentence was based on a substantive
metaphysical claim (albeit an obvious one): that there are no joints
corresponding to “representationally inert” features like font weight. (Given
this substantive claim, I said, a sensible conception of sentence-identity for
epistemology abstracts away from differences in such features.) In the case of
logic, by contrast, it’s plausible to think that there are joints in nature.

10.3 Nonfundamental metalogic

Let us turn from logical to metalogical concepts. Consider first the concept of
a logical constant. The question of what counts as a logical constant is tied up
with the question of the scope of logic. Both the inference from P∧Q to P,
and the inference from ‘a is a bachelor’ to ‘a is male’, are truth-preserving,
and necessarily so (in any reasonable sense of ‘necessary’). But we typically
count only the first inference as being logical. The first inference holds in
virtue of distinctive features of ∧, the second in virtue of distinctive features
of ‘bachelor’, and we typically think of ∧, not ‘bachelor’, as a logical
constant.
This distinction between logical constants and other expressions: is it
fundamental? Is there a corresponding joint in nature? If so, then questions
about which expressions are logical constants—or, alternatively, questions
about the scope of logic—are substantive; and in cases where our concept of
a logical constant—or of inferences that are logical—yields no determinate
verdict, there might still be a fact of the matter. But surely there is no such
joint in nature; the concept of a logical constant is not required in our most
fundamental theories. Some say that logical constants are those expressions
that are defined by their proof-theoretic roles, others that they are the
expressions whose semantic values are permutation-invariant, and still others
that they are the topic-neutral expressions.9 Given that ‘logical constant’ does
not carve at the joints, these proposals should be regarded as conceptual
analyses (or perhaps Carnapian explications) of our conception of the scope
of logic.10 If our conception is determinate enough so that one proposal
uniquely fits it, then there will be a fact of the matter that that proposal is
correct (though the fact might not be substantive). But if our conception is
indeterminate, there may be no fact of the matter which proposal is correct.
Consider, next, the notion of logical consequence. Here, too, there is a
question of whether we have a joint in nature, and there are closely related
questions about substantivity, objectivity, and determinacy
How are claims of objectivity (substantivity, determinacy) regarding
logical consequence related to claims of objectivity regarding the scope of
logic? One might uphold a kind of objectivity for the former but not the
latter: one might think that ‘logical constant’ does not carve at the joints, but
that the relational predicate ‘is a consequence in virtue of the meanings of
expressions e1…’, for variable e1…, does carve at the joints. According to
this view, even if there is no objective fact as to whether ‘=’ is a logical
constant, there is nevertheless an objective fact as to whether ‘x = y’ and ‘y =
z’ imply, in virtue of the meaning of ‘=’, ‘x = z’. But in the other direction, if
‘is a logical consequence of’ (simpliciter) carves at the joints, so that
questions about logical consequence are objective, then one could define a
logical constant as an expression that figures essentially in some logical
implications, in which case questions about the scope of logic would be
objective (even if ‘logical constant’ didn’t carve at the joints).
 As with the concept of a logical constant, there is reason to doubt that
logical truth or logical consequence carve at the joints. Surely these notions
do not improve our fundamental understanding of the world.11
If logical truth and logical consequence are not metaphysically basic, then
they must be metaphysically reduced in some way. I don’t have any
distinctive view to advocate here, but I would like, in a speculative spirit, to
think a bit about whether my extended version of the Ramsey/Lewis “best-
system” account from section 3.1 could be taken as a theory of logical truth.
According to that account, laws are members12 of the system that best
balances simplicity and strength, where “balancing” can be taken in different
ways depending on how much complexity is taken to “cost”. The more it
costs, the more restrictive our account of laws becomes. If complexity is
cheap, then special-science generalizations count as laws; if it has middling
cost then they do not but generalizations of physics still do; and if complexity
is made expensive then even the laws of physics drop away, and all that
remain are “laws of logic”.13
For Lewis, a “system” is a set of true sentences that is closed under some
relation of implication. That relation can’t here be logical consequence, on
pain of circularity. Nor can I take it to be modally strict implication, again on
pain of circularity, since I will define necessity in terms of logical
consequence in chapter 12. Instead, let it be deducibility in an axiomatic
system whose logical axioms and rules of inference are part of what we’re
choosing when we’re choosing the best system. That is: let a system now be
understood as a set of true sentences, any set of true sentences. (As for Lewis,
the language must be constrained; it must contain only joint-carving
expressions, only now this is construed more expansively; all expressions—
including quantifiers and sentential connectives—must carve at the joints, not
just the predicates.) One system is simpler than another depending on how
simply it may be axiomatized. (Like Lewis, I won’t try to say what makes an
axiomatization simple.) An axiomatization of a set, S, is a pair 〈A,R〉, where i)
A, the set of “axioms”, is any subset of S; ii) R, the set of “rules”, is a set of 2
or more place relations on S; and iii) S is the closure of A under R; i.e., S is
the smallest superset of A such that for any n + 1-place r ∈ R and any s1… sn
∈ S and any sentence s, if r(s1… sn, s) then s ∈ S. Given these definitions,
the best-system account doesn’t rely on an antecedently understood notion of
implication.
 Nor does it rely on an antecedent division of terms into logical and
nonlogical. When evaluating a given set for simplicity, we simply look for
the simplest axioms and rules that generate the set. The thought is that in a
competition for the simplest axiomatizations, logical terms will be bound to
figure in the axioms, because of their pervasiveness and topic neutrality.
Indeed, we might define the logical terms as those terms that figure in the
axioms.14
The question, then, is whether the laws of logic, thus understood, could be
thought of as the logical truths. How close is this notion of a law of logic to
the ordinary notion of logical truth? If the language in question is first-order,
the notions might well coincide. For, one could argue, standard axioms and
rules for first-order logic are bound to emerge in the competition as the
axioms and rules in one of the simplest axiomatizations of the winning
system. This system must be exceedingly simple, remember; increasing
strength by adding extra-logical information will require too much
complexity. Since these axioms and rules are sound and complete, the laws of
logic will be exactly the logical truths.
(A potential problem for this argument is that some axiomatizations of the
first-order logical truths have infinitely many axioms. But how can systems
that require infinitely many axioms be simple? Relatedly, suppose one of the
axiom schemas is “If ϕ then ϕ.” Then for each sentence ϕ in the language,
there will be a corresponding axiom. ‘If grass is green then grass is green’, ‘If
something is grue then something is grue’, and so on, will all be axioms.
Thus each notion that can be expressed in the language—grass, green, grue,
and so on—will be involved in some axiom. How then can the axiomatization
be simple? An answer might be that even if the axioms are infinite in number,
and involve many different notions (many of which fail to carve at the joints),
each axiom falls under one of a small number of simple axiom schemas; and
the schemas involve only logical notions. But notice that this turns the
simplicity of the axiomatization into a fact that’s “external” to the language in
question; it depends on facts that cannot be in general articulated from within
the language.)
Logical lawhood will not coincide with logical truth, as usually
understood, if the language includes more expressive resources, such as
second-order variables and quantifiers. For since there are no sound and
complete axiom systems for second-order logic, the best systems will not
include all the semantically valid second-order sentences. This is a clash with
typical usage, since such sentences are usually taken to be logical truths. (I
doubt, however, that the second-order quantifiers carve at the joints; so I
doubt that this mismatch is a problem when it comes to fundamental
languages.)
Also, it’s telling that the best-system account of the metaphysics of logical
truth doesn’t seem be what inspires formal work on metalogic, in particular
on model theory. The model-theoretic definition of logical truth in formal
languages as truth-in-all-models seems to be inspired, rather, by the vague
conception of logical truth as truth “in virtue of” the meanings of logical
constants.15. Further, this vague conception seems to be our ordinary
conception of logical truth. It’s doubtful, then, that the best-systems account
captures our ordinary concept, or the logician’s concept, of logical truth.

10.4 Logical pluralism

JC Beall and Greg Restall (2006) defend “logical pluralism”, the claim that
there is no single correct notion of logical consequence. Classical
consequence, intuitionistic consequence, relevant consequence, and other
such competitors are all on a par. According to Beall and Restall, there is no
sensible question as to whether the argument “P; ~P; therefore, Q” is really
valid. It’s classically and intuitionistically valid, but not relevantly valid, and
that’s all that can be said. This meshes with the previous section’s claim that
metalogical notions do not carve at the joints. But various theses in the
vicinity should be distinguished.
First, one might be a pluralist about logical consequence simply because
one is a pluralist about the notion of a logical constant. Suppose i) logical
consequences are defined as consequences in virtue of the meanings of the
logical constants; ii) pluralism about the notion of a logical constant is true;
but iii) antipluralism about the notion of a “consequence in virtue of the
meanings of expressions e1…” is true. This would result in a sort of pluralism
about logical consequence. But it would not be Beall and Restall’s pluralism,
since they would continue to uphold pluralism even after the logical constants
have been fixed (by convention, say).
Second, one might be a pluralist only at the level of the family of logical
consequence, so to speak, rather than at the level of its genera. One might, for
example, be an antipluralist about both provability and semantic
consequence, but hold that provability and semantic consequence are equally
good notions of logical consequence. Again, this isn’t Beall and Restall’s
pluralism. The pluralism they are most concerned to defend is at the genus
level: that there is no uniquely correct notion of semantic consequence.16
Third, and most importantly, the denial of a single “correct” notion of
logical consequence might be taken either in a metaphysical sense, as the
denial that any one of various candidates is metaphysically privileged (none
carves at the joints better than the others), or in a conceptual sense, as the
denial that any one of the candidates is conceptually privileged (none is
uniquely our concept). Beall and Restall primarily defend the conceptual
thesis. Classical, intuitionistic, and relevant consequence, they in effect argue,
are all legitimate notions—perhaps useful for different purposes—and no one
of them uniquely captures our concept of logical consequence. But they had
better defend the metaphysical claim too, since the conceptual claim does not
on its own rule out the existence of further facts about logical consequence
that outstrip our conception. In a case where classical and relevant
consequence (say) come apart, and our conception is silent, there may yet be
an objective fact of the matter as to whether logical consequence really holds.
Compare: an ethical pluralist will not stop with the claim that utilitarian and
deontological conceptions both satisfy, well enough, the way we think about
morality. She will also want to reject the idea that in cases where utilitarian
and deontological conceptions come apart, there is a further fact of the matter
as to what is really morally right to do.
Beall and Restall’s pluralism, then, should be augmented with a
metaphysical claim: none of the candidate notions of logical consequence
carves at the joints better than the others.17 Thus augmented it implies that
questions about logical consequence are nonsubstantive in the sense of
chapter 4—a result that seems to mesh with their picture. I suspect that this
view is correct.
Even thus augmented, though, notice that their pluralism is a claim about
the metalogical notion of logical consequence, not about logical notions like
disjunction and negation. (Thus it might better be called metalogical
pluralism.) It is therefore compatible with the claim that logical notions carve
at the joints. But if logical notions carve at the joints, then questions stated in
terms of those notions—for instance, questions of the form “ϕ or not-ϕ?”—
are substantive.18 As a result, their pluralism is not as deflationary as it might
first seem—not even in the realm of metalogic, in fact.19 For many questions
about logical consequence have ramifications for purely logical questions.
Consider, for example, the law of the excluded middle, “ϕ or not-ϕ”,
whose instances are logical truths (i.e., logical consequences of the empty set
of sentences) in classical but not intuitionistic propositional logic. If the
dispute over excluded middle were only about whether those instances are
logical truths—if, that is, all disputants accepted the instances but merely
disagreed over whether they are logical truths—then Beall and Restall’s
pluralism would imply that the dispute is nonsubstantive. But the dispute is
also about those instances themselves. Mathematical intuitionists reject
classical logic in favor of intuitionistic propositional logic precisely because
they do not accept certain instances of excluded middle in which neither
disjunct has been proven, such as:

(*) The string “666” either does or doesn’t occur at least 666 times in the
decimal expansion of π.
Since (*) contains no metalogical notions, (meta)logical pluralism allows a
dispute over it to be substantive.20 Thus (meta)logical pluralism allows this
aspect of the dispute over classical logic to be substantive. Further,
intuitionists who refuse to accept (*) cannot even recognize classical logical
truth as a legitimate notion of logical truth. For under any legitimate notion
of logical truth, surely one can infer ϕ from “ϕ is a logical truth”.
Relatedly, certain dialetheists respond to the semantic paradoxes by
asserting contradictions, sentences of the form “ϕ and not-ϕ” (Priest, 1987).
Again, the dispute isn’t just over the metalogical status of contradictions; it’s
also over the contradictions themselves. Here again we have an aspect of the
dispute over classical logic that (meta)logical pluralism allows to be
substantive, since the relevant notions are logical, not metalogical. Further,
dialetheists who accept contradictions cannot even recognize classical
consequence as a legitimate notion of logical consequence. For under any
legitimate notion of logical consequence, surely one can infer ϕ from “ϕ is a
logical consequence of ϕ” and ϕ; but in classical logic, all sentences are
logical consequences of contradictions, which would commit the dialetheist
to accepting all sentences.

10.5 Objectivity in model theory

Beall and Restall are pluralists about the genus semantic consequence, but do
not question the objectivity of its species—classical semantic consequence,
intuitionistic semantic consequence, and so on. Objectivity might fail even
here, for a quite different reason: the vocabulary used to make statements
about models might admit multiple equally good interpretations—even
holding fixed what sorts of models are at issue (classical, intuitionist…).
For example, suppose statements about models are to be taken at face
value, as quantifying over abstract entities; suppose that the quantifiers fail to
carve at the joints; and suppose in particular that various equally joint-carving
quantifier-meanings differ in what truth-values they assign to statements
about infinite totalities of abstracta. Then there would correspond various
equally joint-carving meanings for statements about what is true in all models
of a given type.
Or suppose again that statements about models are to be taken at face
value; but now suppose that although quantifiers carve at the joints, the set-
theoretic predicate ∈ does not. Failure of objectivity in statements about
models might then result from multiple candidate meanings for ∈.
Or suppose that there are no abstract entities, and that statements
quantifying over abstracta must be reinterpreted in some nominalistic way.
Objectivity might again fail, if the ideology used in the reinterpretation fails
to carve at the joints. For example, if the nominalist claimed that Г
semantically implies ϕ iff it’s impossible for there to exist a model of Г in
which ϕ is false, then if the modal notion of possibility fails to carve at the
joints, statements about semantic consequence could fail to be objective.21
Alternatively, instead of using modal ideology in the reinterpretation, the
nominalist might use higher-order quantifiers claimed to be “innocent”—not
to be understood as first-order quantifiers over sets (nor for that matter as
plural quantifiers over the same domain as the first-order quantifiers).22 If the
higher-order quantifiers fail to carve at the joints, then statements about
models understood in the higher-order terms might fail to be objective.
Another nominalist possibility: Stephen Yablo (2000) defends the practice
of quantifying over abstract entities such as models without thereby
committing to the literal existence of abstracta, since one can regard such
quantification as taking place within a pretense. An advocate of this approach
might go further than merely defending the propriety of this practice, and
offer a pretense theory of the metaphysical nature of semantic consequence:
for ϕ to semantically imply ϕ is for the metaphor of set theory to render
pretense-worthy the following statement: “ϕ is true in every model (of some
specified sort) in which every member of ϕ is true.” But if—as is surely the
case—the relation of rendering pretense-worthy doesn’t carve at the joints,
facts about semantic consequence would be in danger of being nonobjective.
I say “in danger of because the failure of ‘renders pretense-worthy’ to
carve at the joints wouldn’t, on its own, undermine objectivity. What is
pretense-worthy depends on how the world really is, and the facts about how
the world really is might include facts about other joint-carving notions that
secure objectivity.23 Suppose, for example, that an appropriate modal notion
carves at the joints, so that facts about which models might have existed are
objective. Facts about which models exist according to pretenses might then
be objective, even if ‘renders pretense-worthy’ doesn’t itself carve at the
joints. Compare: some children are pretending that a certain phone is a bomb,
and the ringing of the phone represents the explosion of the bomb. In fact,
unbeknownst to the children, the phone is about to ring. It’s then objectively
true that “The bomb is about to go off” is true in the pretense. What’s doing
the heavy lifting here is the objectivity of statements about the future, not the
relation of rendering-pretense-worthy. (If the future were “open”, then there
would be no determinate fact of the matter of whether this is true in the
pretense.) Similarly, if higher-order quantifiers carved at the joints, then
objective higher-order facts could ground facts about truth under the pretense
of set-theory, thus resulting in objective facts about consequence, even if
rendering-pretense-worthy does not itself carve at the joints. So Yablo’s
account is consistent with objectivity in facts about consequence. But
assuming, as we surely should, that ‘renders pretense-worthy’ does not carve
at the joints, the objectivity would derive from modal or higher-order facts
that are external to the pretense-theoretic account—the world would contain
real facts that are nearly isomorphic to the facts that are being pretended to
exist. So why not appeal directly to those real facts, and defend a modal or
higher-order theory? (This is not a criticism of Yablo, since his aim is not, I
think, to provide a metaphysics of logical consequence.)
Incidentally, we can now see a weakness in a popular argument against
platonist conceptions of logical consequence. Yablo (2000, §IX) argues
against the model-theoretic account thus:24
An argument’s validity-status would seem to be a conceptually necessary fact about it. Surely we don’t
want the validity of arguments to be held hostage to a brute logical contingency like what model-like
entities happen to exist!
The charge is that the model-theoretic account gives away a hostage to
fortune: it makes the facts about consequence hostage to the existence of
abstracta. If the world doesn’t cooperate, facts about logical consequence will
fail to be objective. But nominalistic metaphysical accounts also give
hostages. If the world doesn’t cooperate—if truth-in-pretense, modality,
higher-order quantification, and so on fail to carve at the joints—then facts
about consequence will fail to be objective. Even the primitivist about logical
consequence is hostage to the unhappy possibility that the primitive
expression ‘logically implies’ (or, ‘it is logically necessary that’) doesn’t,
after all, carve at the joints. The nonplatonist’s hostages are ideological, not
ontological, but they are hostages all the same. If we want to believe in
objective facts about consequence, we must make a commitment, ideological
or ontological. And the attraction of making this commitment in the
platonistic, ontological, way is that it keeps our ideology simple.25
The fact that nominalist approaches to metalogic also give hostages is
often missed. I wonder whether this is yet another wage of Quine’s sin of
terminology. “Ideological” commitments, far from being merely
psychological, are as much commitments to metaphysics as are ontological
commitments.

10.6 Classical logic and fundamentality

Despite its brilliant success in mathematics and fundamental science,
classical logic applies uneasily to natural language. The mismatch between
natural language conditionals and the material conditional is well known.
Ordinary language philosophers challenged even classical logic’s account of
‘and’, ‘or’, and ‘not’26 (though Grice’s (1975) powerful reply won the
classical logician a respite). Free logicians argue that natural language names
—which may be empty, like ‘Sherlock Holmes’—and quantifiers are not
related as they are in classical logic. More significantly, linguists have
identified numerous ways in which the syntax and semantics of natural
language logical words is richer than in classical logic. ‘And’ and ‘or’ have a
richer function than expressing sentential conjunction and disjunction (they
connect nonsentences, for example); ‘not’ does more than express sentential
negation (Horn, 1989); natural language quantifiers are syntactically unlike
classical logic’s (Montague, 1973), and can bind anaphora beyond sentence
boundaries (see King (2005) for an overview); and so on.
 It is overwhelmingly tempting to respond to this combination of success
and failure by saying that, although classical logic is an inadequate model of
all the complexities of natural language, it is adequate for the simpler
languages of mathematics and fundamental science. And it is also tempting to
privilege the latter languages in some way. Logic is ultimately classical; it’s
just that natural languages hook up to reality in complex and indirect ways.
A great advantage of realism about structure is that it lets us succumb to
this temptation. Classical logic, we can say, holds in fundamental languages.
That is, in fundamental languages: i) the logical ideology includes that of
classical first-order logic; and ii) the consequence relation (over the first-
order fragment anyway) is classical.27 This view is unthreatened by the
mismatch between natural language and classical logic, it preserves the role
of classical logic in more fundamental endeavors, and it privileges classical
logic: languages whose logic is nonclassical are metaphysically second-rate.
The semantic paradoxes and the paradoxes ofvagueness also challenge
classical logic. Some have argued that the right response is to renounce
classical logic; suppose for the sake of argument that they are right. Should
we continue to uphold the tempting view, and say that the renouncement is
only for nonfundamental languages? Or should we give up classical logic at
the last, and adopt a nonclassical logic at the fundamental level (perhaps a
logic that reduces to classical logic in special cases like mathematics and
fundamental science)? I will argue for the first choice, and draw some morals
about the relationship between fundamental and nonfundamental languages.
Suppose our language is powerful enough to speak of elementary
arithmetic. Then we can speak of linguistic items by Gödel-numbering. For
any sentence, A, let 〈A〉 be a name of A (i.e., of A’s Gödel number; I’ll
suppress this from now on). Suppose further that our language includes a
truth predicate, ‘True’, which obeys the following inference rules:28
Truth-introduction S True (〈S〉)
Truth-elimination True(〈S〈) S
Provided we accept a minimal amount of arithmetic, we can use techniques
from Gödel to construct, for any formula B(y), a sentence G that “says” I am
B, in the sense that G B(〈G〉) and B(〈G〉) G. Applied to the formula
~True(y), these techniques yield a sentence L such that:
(Ia) L —True(〈L〉)
(Ib) ~True(〈L〉) L
L says, in effect, I am not true. We can now derive a contradiction using
classical logic:
1. True(〈L〉) assume for reductio
2. L 1, Truth-elimination
3. ~True(〈L〉) 2, (Ia)
4. ~True(〈L〉) reductio, 1-3
5. L 4, (Ib)
6. True(〈L〉) 5, Truth-introduction. This contradicts 4.

Something has to give. There are many choices for what gives, some of
which involve retaining classical logic. But there are attractive ways to block
the paradox by revising classical logic.29 These nonclassical approaches to
the paradox are attractive in part because they (or some of them, at any rate)
let us retain truth-introduction and truth-elimination, thus preserving the
inferential role that we naïvely take the truth predicate to have. Indeed, the
core inference rule underwriting our naïve use of the truth predicate is
stronger:
Truth-transparency S S′ whenever S and S′ differ only in that one of
them contains one or more occurrences of some sentence A where the
other contains True(〈A〉).
One of the main purposes of having a truth predicate in natural language is to
give access to the content of what someone said, when one cannot reproduce
her words (whether because one has forgotten exactly what she said, or
because one is making a general statement that is supposed to hold good
regardless of what she said).30 We need truth-transparency to move from the
premise “Everything Jones said is true” to such conclusions as “If Jones said
‘snow is white’ then snow is white”, “If Jones said ‘grass is green’ then grass
is green”, and so on. So securing truth-transparency is a high priority.
One very simple nonclassical account of the liar runs as follows. We claim
that the correct propositional logic is the nonclassical logic K3—the logic
generated by Kleene’s (strong) three-valued tables:
An argument is K3-valid iff its conclusion gets value 1 whenever its premises
get value 1, as evaluated under the Kleene tables, for every assignment of
values drawn from {1,0, #} to its atomic sentences. The adoption of K3
blocks the argument given above from (Ia) and (Ib) to contradiction, since it
invalidates31 reductio ad absurdum (“if A B∧~B then ~ A”), which was
used in that argument. More generally: suppose we begin with a first-order
theory, T, which includes enough arithmetic to do syntax, but which does not
include a truth predicate. Kripke’s (1975) paper on truth shows how to
construct, for any initial classical (two-valued) model, M0, of T, a three-
valued model M for the language of T plus a one-place predicate ‘True’, in
which:
• propositional connectives are valuated using the Kleene tables; quantifiers are valuated in the
obvious analogous way;32
• nonlogical expressions in the original language have the same denotations (extensions, etc.) in M
as in M0;
• sentences not containing ‘True’ get the same values in M and M0;
• each sentence S gets the same value in M as does True(〈S〉). Thus, Truth-transparency—and so
Truth-introduction and Truth-elimination as well— are validated in M (each preserves value 1).

So if we begin with a consistent theory (in the classical model-theoretic

sense), there is a way to add a truth predicate obeying Truth-transparency to
obtain a consistent (in the K3 model-theoretic sense) extension ofthat theory.
The K3 solution to the liar paradox is imperfect. Most notably, K3 contains
no reasonable conditional: one cannot define a conditional → using ∧, ∨, ~
such that A→A33 There are ways to add a reasonable conditional to K3, but
they add complexity that isn’t relevant here.34 So let’s stick to the K3 solution
as our working example of a nonclassical solution to the liar paradox.
Classical logic is also threatened by vagueness. The sentence:
(J) Jones is bald ∨~ Jones is bald
is a classical logical truth, but seems inappropriate to accept if Jones is a
“borderline case” of baldness. Anyone who accepted that Jones is either bald
or not bald could reasonably, it would seem, wonder which it is—is Jones
bald, or is he not bald? Since such wondering seems inappropriate,
acceptance of (J) seems illegitimate.35
There are various proposals for how to modify classical logic to deal with
vagueness. One of them is simply to move to K3. Since the law of the
excluded middle does not hold in K3, the defender of K3 need not accept (J).
As mentioned earlier, the K3 solution has warts (most notably the absence of
a decent conditional), but again, it will do as an example of a nonclassical
solution.
Whether or not the K3 solution to the paradoxes is really correct, we can
ask what it would show about fundamental reality if it were correct. What do
nonclassical approaches to paradox teach us about reality’s fundamental
logical structure?
Nothing, one wants to answer. In our most secure and successful domains
of inquiry, fundamental science and mathematics, classical logic works like a
charm. “Works” understates it: classical logic is an integral part of our
theories in those domains. So the view that reality’s fundamental logical
structure is classical, and that classical logic breaks down only in
nonfundamental languages, is initially compelling.
The opposing view would be that logical reality is fundamentally
nonclassical, but that the classical laws hold within certain restricted
domains, for example within fundamental science and mathematics. The
restricted success of classical logic might be compared to the approximate
success of classical physics in macroscopic domains.
But the comparison is imperfect. It’s attractive generally to keep
fundamental theories simple. That’s why it’s attractive to keep logic classical,
at the most fundamental level. But keeping fundamental physics simple, by
retaining (say) Newtonian physics, is simply not an option. The asymmetry
between the two cases is this: in physics, the simple theory breaks down at
the fundamental level, though it holds (at least approximately) for certain less
fundamental domains, whereas in logic, the simple theory looks good at the
fundamental level, and only apparently breaks down in nonfundamental
domains.
So it would be nice to keep logic fundamentally classical, to say that
classical logic holds in fundamental languages. And we can say this even
while embracing the K3 solution to the paradoxes if we say that the
paradoxes arises only in non-fundamental languages. Fundamental languages
lack the features that threaten paradox: there is no vagueness at the
fundamental level (section 7.12), and truth-theoretic ideology is not
fundamental.36 These are substantive claims of metaphysics: vagueness and
truth are nonfundamental.
 Adopting this strategy, however, requires saying some striking things
about the relationship between fundamental and nonfundamental languages.
In the terminology of section 7.4, it requires a distinctive form of
metaphysical semantics. The remainder of this section will explore this issue.
Let be a nonfundamental language in which we wish to make the K3
response to the liar and vagueness paradoxes. Assume that the form of
metaphysical semantics for is truth-conditional (rather than expressivist,
say). Let be a fundamental language in which metaphysical truth-
conditions for are given. Assume that contains the notions of classical
first-order predicate logic; and assume further that we can identify in
analogs of ’s logical constants— connectives whose inferential roles in
are analogous to the roles of their fundamental counterparts in . (When
necessary, we will write ’s logical constants in boldface to distinguish
them.) Let μ be the function that maps sentences of to their metaphysical
truth-conditions in . Assume that we have a truth predicate for : ‘true in
. ( does not itself contain a truth predicate, but we can introduce a truth
predicate for it.) Assume that this truth predicate behaves normally with
respect to the connectives—that, for example, a disjunction is true in iff
one of its disjuncts is true in . (Given the lack of vagueness or semantic
vocabulary in , there is no reason to doubt this.) And assume, finally, the
following connection between metaphysical truth-conditions and acceptance:
if μ(A) is true in , then A is to be accepted in . I intend “to be accepted” in
an objective sense; whether a sentence is to be accepted does not depend on
what evidence one has.
Question: what form must the metaphysical truth-conditions for take?
In particular, how will those truth-conditions treat the logical constants of ?
The first thing to note is that those metaphysical truth-conditions cannot
be “logical form preserving”, in that for any formulas A, B, and variable x of
, μ(~A) = ∼μ(A)μ(A ∨ B) = μ(A)∧μ(B), μ(∃xA) = ∃x μ (A), and so on.
Given logical form preservation, the metaphysical truth-condition of sentence
(J) would have the form A∨~A Given classical logic in the latter is true in
Thus (J) is to be accepted in But the defender of the K3 approach to
vagueness instructs speakers of not to accept (J).
Logical form-preserving metaphysical truth-conditions were in jeopardy
anyway. For as we saw in section 7.7, given a very minimal fundamental
ontology, the metaphysical truth-conditions for existential sentences might
not be existential in form. But in fact, the K3 approach to the paradoxes rules
out even negation-preserving metaphysical truth-conditions—truth-
conditions under which μ(∼A) = ∼(A), for all A. The reason is that for certain
sentences, the approach requires us to reject both those sentences and their
negations.
Return to the K3 account of vagueness. According to that account, we
must not accept (J). But nor may we accept (J)’s negation:
(~J) ~(Jones is bald ∨ ~ Jones is bald)
If we did accept (~J), we would have to accept B, for every sentence B. For ~
(A∨B) and ~(A∨B) B are both K3 valid, as is ex falso: A,~A B. The
situation with the liar is similar: we must accept neither L nor ~L. For if we
accepted L we could reason to arbitrary B thus:37

1. L by hypothesis
2. ~True(〈L〉) I, (Ia)
3. True(〈L〉) I, Truth-introduction
4. B 2, 3, ex falso

and likewise from ~L:

1. ~L by hypothesis
2. ~True(〈L〉) I, Truth transparency
3. L 2, (Ib)
4. B I, 3, ex falso

So the K3 solution to the paradoxes requires that for certain sentences we

accept neither them nor their negations. But it calls for something more.
When we are ignorant of a sentence’s truth-value, we also accept neither that
sentence nor its negation. I do not accept ‘There are an even number of trees
in North America’, nor do I accept its negation. But I in no sense repudiate
this sentence, whereas some sort of repudiation of L, ~L, (J), and its negation
seems to be called for. For as we saw, acceptance of L, ~L, or (J)’s negation
would require accepting every sentence; and accepting (J) would require
being in a state of ignorance that seems inappropriate. The requisite sort of
repudiation, stronger than mere failure of acceptance but weaker than
accepting the negation, is often called rejection. Rejecting A is not merely
failing to acceptA, but does not require accepting ~A
 The K3 resolution of the paradoxes, then, instructs speakers of to reject
both B and ~B, for various B. But given negation preservation, the
metaphysical truth-conditions of B and ~B are μ.(B) and ~μ(B), respectively.
Given classical logic in , μ(B) ∨ ~μ(B) is true in , and so either μ(B) or
~μ(B) is true in and so either B or ~B is to be accepted in . Since no
sentence is to be both accepted and rejected, the K3 resolution is incompatible
with negation preservation.
So the K3 solution to the paradoxes rules out negation-preserving
metaphysical truth-conditions. But what then will metaphysical truth-
conditions for negations look like? Here is a natural picture. Suppose some
three-valued model M, as described in the discussion of Kripke above, can be
regarded as an intended model for the language . Suppose further that
can speak of model theory. A metaphysical semantics might then assign to
each sentence, A, of the metaphysical truth-condition “Sentence A has
value 1 in model M.” Notice that the metaphysical truth-condition for ~A,
namely “Sentence ~A has value 1 in model M”, is not the negation of the
metaphysical truth-condition for A Indeed, both truth-conditions can fail to
be true in , since M is three-valued. Both the liar sentence L and its
negation, for example, will have value # in M, and so neither has a true-in-
metaphysical truth-condition. It is natural to pair this metaphysical semantics
with the principle that a sentence whose metaphysical truth-condition is not
true in is to be rejected. Thus both the liar sentence and its negation are to
be rejected.
(It is crucial to distinguish “has a true metaphysical truth-condition” from
’s truth predicate ‘True’. L does not have a true metaphysical truth-
condition since it has value # in M; but we must not accept ~True(〈L〉)
because doing so would lead to contradiction—to L by (Ib), and so to
True(〈L〉) by Truth-introduction.38)
So: one can combine classical logic at the fundamental level with
nonclassical resolutions of the paradoxes, but at a price. In the
nonfundamental languages in which the paradoxes are nonclassically
resolved, negation doesn’t mean fundamental negation.
One final point. I have been speaking of metaphysical truth-conditions,
but have not specified in which language this occurs. If it occurs in itself—
as well it might; is supposed to model our own language, after all—then a
problem arises, as Hartry Field pointed out to me. (The problem would also
arise on approaches to the liar other than the K3 approach.) A liar argument
may be formulated in using the predicate ‘has a true metaphysical truth-
condition’.
Let Cxy mean in that x is a metaphysical truth-condition for y, and
make the following assumptions. 1: has no nonfactual sentences, so that
completeness (section 7.5) implies that every sentence of has a
metaphysical truth-condition. 2: The vocabulary of is contained in thus
metaphysical truth-conditions can actually be expressed in Thus for any
sentence A in , there is a sentence A0 in such that C〈A0〉〈A〉 is true in 3:
Metaphysical truth-conditions are unique; for simplicity, construe this as
allowing the following inference: Czy,~ϕ(z) ~∃x(Cxy ∧ ϕ(x)). 4:C〈A0〉
〈A〉,A0 A (recall the claim above that a sentence in with a true
metaphysical truth-condition is to be accepted). Let τ(y) abbreviate ∃x(Cxy
∧True(x)). Gödel’s techniques applied to the formula ~τ(y) yield a
“metaphysical liar sentence” M, which says that its metaphysical truth-
condition is not true; that is: i) M ∼ τ (〈M〉); and ii) ~τ(〈M 〉) M. By
assumption 2, M has a metaphysical truth-condition M0; thus C〈M0〉〈M〉. M0
is a sentence of , so presumably this instance of excluded middle holds:
M0∨~M0. We can then reason by cases to M : from the left by assumption 4,
and from the right as follows: suppose ~M0; then ~True(〈M0〉) by Truth-
transparency; so ~τ(〈M〉) by uniqueness; so M by ii). Then from M we can
conclude ~τ(〈M〉) by i); and so ~True(M0) by K3 -predicate-logic,39 and so
~M0 by Truth-transparency. So for the sentence M, we can prove both it and
the negation of its metaphysical truth-condition!
My tentative reply: the notion of a metaphysical truth-condition is a
theoretical one, introduced for a certain theoretical purpose. Thus certain
approaches to the paradoxes are more attractive when applied to the
paradoxes of metaphysical truth-conditions than to the paradoxes of truth
itself, construed as an ordinary, intuitive notion.40 For example, the situation
with M and M0 might simply be accepted. A metaphysical semantics is
supposed to be an explanatory theory of a certain sort; and for languages like
, perhaps the most explanatory theory involves a notion of metaphysical
truth-condition that has certain anomalous cases. The fact that “having a true
metaphysical truth-condition” does not behave exactly like the ordinary
notion of truth would not, according to this perspective, count against the
explanatory worth of the theory. Alternatively, a Tarskian hierarchical
approach to metaphysical truth-conditions could be taken. On this approach,
no language can speak of metaphysical truth-conditions for its own sentences
(and so we cannot accept an unrestricted thesis of completeness). The usual
objection to the Tarskian approach stresses how we apply the ordinary notion
of truth without regard to Tarskian level, and thus is no objection to a
Tarskian approach to the theoretical concept of metaphysical truth-
conditions.

1 If a logical notion such as V does not carve at the joints, then the question of whether it carves at
the joints is not phrased in purely joint-carving terms. But this does not make the question
nonsubstantive. For every candidate metaphysical semantics for the sentence ‘V carves at the joints’,
i.e., ‘ (V)’ will count it false. A metaphysical semantics that assigns a true truth-condition, such as ‘
(∈)’ or ‘ (∧)’ (assuming that both ‘A’ and ‘E’ carve at the joints) won’t fit the use of ‘ (V)’
well enough to count as a candidate. The candidates will take the form (α) where a is logically
complex—‘ ~(~Pʌ~Q)’, perhaps, in one of the notations introduced in section 6.3—and such
sentences are false (a metaphysical semantics is given in purely joint-carving terms, so if α is logically
complex then it does not carve at the joints).
2 Those who reify relations might instead get off the hook by identifying earlier-than with later-
than; see Fine (2000); Williamson (1985); but see also Dorr (2004).
3 See, for example, Field (1980, chapters 3, 5).
4 I have in mind pairing this view with an acceptance of classical logic (at the fundamental level,
that is; see section 10.6). And I have in mind only the one- and two-place connectives.
5 This is not a definition. If anything is to be defined, it should be ‘more massive than’: x is more
massive than y iff x has some mass property that is greater than some mass property of y.
6 See Eddon (2012).
7 Replacing ‘green’ and ‘blue’ with predicates of fundamental physics would result in an example
with the vice of unfamiliarity but the virtue of correctness.
8 Fine also faces the worry; can we ask whether: in reality, some things are charged, but it’s not the
case that in reality, some things are charged?
9 See MacFarlane (2005) for a survey.
10 This seems to be how the proposals are generally viewed; see Etchemendy (1990, introduction).
11 Field’s (1980; 1989) nominalist account of fundamental physics makes use of a primitive notion
of logical consequence. But this notion plays no role in Field’s official nominalist versions of
fundamental physical theories. It comes into play only in his account of why it’s legitimate to use
mathematics within physics.
12 Lewis restricts ‘law’ to the members that are regularities—universal generalizations. I don’t quite
see the point; but anyway it’s inappropriate in the case of logic: logical truths like ‘If snow is white then
snow is white’ aren’t regularities. (Though if the language in question has propositional quantification,
then it can state regularities like ‘For all p, if p then p.’)
 13 Objection: a system that includes only laws of logic has zero strength. Reply: logical truths do
have content; they concern the world’s most general and abstract aspects.
14 Given the vagueness in the account (e.g., in terms like ‘balance’), there might be vagueness in
which terms count as logical. If mereological terms carve at the joints, for example, it might be vague
whether they count as logical.
15 We can think of truth in virtue of the meanings of the logical constants as truth that remains no
matter how everything else is varied, and truth-in-all-models as a way of formalizing the intuitive
notion of truth-no-matter-how-everything-else-is-varied. Thanks to Agustín Rayo.
16 At the level of family they write as antipluralists, assuming that semantic consequence is the core
concept of logical consequence. But this may be just for the sake of definiteness.
17 Side issue: Beall and Restall seem to be antipluralists about metaphysical modality. But the case
against joint-carving metaphysical necessity seems at least as strong as the case against joint-carving
logical consequence: metaphysical necessity seems at least as theoretically dispensible, “spooky”, and
epistemically intractable as logical consequence.
18 Beall and Restall are at pains to show that their pluralism is consistent with realist metaphysics,
and in this vein point out that it doesn’t imply relativism about truth. The present point goes much
further: their pluralism is compatible with the substantivity of logic.
19 In another respect it is more deflationary: it generates pluralism about epistemic notions that are
constitutively connected to logical consequence. Beall and Restall address this, and say two things
(2006, 94–97). First, they say that everyone already admits a sort of epistemic pluralism: pluralism
about what threshold of epistemic value is required for positive appraisal—for “justification”, say. But
consider: from an intuitionist’s point of view, belief in (*) (see text) is epistemically terrible, no better
than a guess; whereas from the point of view of a classical logician it is epistemically superlative. Beall
and Restall will want to say that neither point of view is objectively mistaken, which requires pluralism
about the underlying scale on which the threshold is defined, and not just pluralism about the threshold.
Second, they say that epistemic value is not determined solely by logical consequence. But in cases
where the only source of epistemic value would be a beliefs status as a logical truth—a belief in (*), for
example—their pluralism would indeed imply pluralism about epistemic value.
20 In a “dispute” over a sentence, one side normally accepts the sentence while the other side
accepts its negation; but not here: intuitionists do not accept the negation of (*) since the negation of
the negation of (*) is a logical truth in intuitionistic logic. (Intuitionists reject double-negation-
elimination in addition to excluded middle.) What the intuitionists want is to resist classical logic’s
demand to assert (*), not to accept its negation.
21 As Stephen Yablo pointed out, similar remarks apply to related proposals that replace
metaphysical with epistemic possibility
22 The approach I have in mind uses formal strategies from Rayo and Yablo (2001), Rayo and
Uzquiano (1999), and Williamson (2003). In outline: first think of the domain of a model as containing
properties, of predicate-extensions as higher-level properties and relations, and of the model itself as a
still higher-level relation between linguistic items and their extensions; then reconstrue all this talk of
properties and relations in terms of “innocent” higher-order quantification. The required ideology is
obviously very powerful; it’s hard to see why such ideological commitment would be preferable to
ontological commitment to sets.
23 Compare the related points in Yablo (2001, sections 4 and 5).
24 See also Etchemendy (1990).
 25 There is a quite different way of taking Yablo’s anti-platonist argument (perhaps hinted at by his
mention of “conceptual necessity”). He might be insisting that the facts of logical consequence are
somehow intrinsic to the sentences (or propositions) involved, whereas for the platonist, facts of logical
consequence inherently involve further entities: models. (Compare Field’s (1980, chapter 5) objection
to Platonist accounts of measurement.)
26 See in particular Strawson (1952, chapter 3, section 2).
27 Requirement ii) is not meant to imply that ‘consequence’ is a predicate of the fundamental
language; that predicate is one of my language.
28 I’ll be a bit loose in my use of the symbol. A …A
1 n B means that sentences A1…An, perhaps
with a background mathematical theory we accept, logically imply sentence B. The “logic” may include
the logic of the truth-predicate, and the underlying propositional logic in question will vary from
context to context. Here at the outset the assumed propositional logic is classical; later it won’t be. I’ll
sometimes explicitly indicate the underlying propositional logic with a subscript on
29 See Field (2008) for a survey.
30 See Horwich (1990).
31 For example, p ∧ ∼ p
K3 p∧∼p, but K3 ∼(p ∧ ∼ p) since ∼(p∧∼p) is # when p is #.
32 e.g. ∃vA gets value 1 if A gets value 1 under some assignment to v, value 0 if A gets value 0
under all assignments to v, and gets value # otherwise.
33 It’s easy to see that for no sentence B is h B, since whenever every atomic sentence is #, so is
every complex sentence.
34 Field (2003; 2008) adds a conditional to K and uses the resulting logic to give a solution to the
3
semantic and vagueness paradoxes.
35 See Field (2008, pp. 152–3). The sorites paradox also challenges classical logic.
36 There are other paradox-threatening features, such as property- and set-theoretic quantification
and ideology. The friend of fundamental classical logic can either reject a given feature at the
fundamental level, or else accept it but adopt some classical resolution of the corresponding paradoxes.
(My own choice is rejection for property-theory and a classical approach—ZF, say—to fundamental set
theory.)
37 The arguments for the inferences (Ia) and (Ib) go through provided classical logic may be
assumed for the purely mathematical part of the language.
38 In essence, having-a-true-metaphysical-truth-condition is the non-disquotational, “external”,
notion of having value 1 in Kripke’s minimal fixed point, whereas the K3 solution to the liar identifies
truth with what is expressed by ‘True’—the “internal”, disquotational notion. See Field (2008, section
3.4).
29 See Field (2008, chapter 3, Appendix) for the inference rules.
40 Compare those who distinguish between correspondence and disquotational truth.
11 Time
A cloud of suspicion hangs over the philosophy of time. Although the
pictures are vivid—the static manifold, the growing block, the policeman’s
bullseye—many doubt that coherent positions and genuine contrasts lie
behind them.
I think, however, that the traditional questions about time are good ones.
Let us view them through the prism of realism about structure.

11.1 Presentism
Our focus will be “A-theories” of time, theories according to which “time
passes”. Let us begin with (Priorian1) presentism.
What defenders of time’s passage share is opposition to the spatializing
philosophy of time familiar from Quine, Russell, Smart, and many others.2
The central question in the philosophy of time is: How alike are time and
space? Russell, Quine, and Smart answer: Very alike indeed. Reality consists
of a four-dimensional “block universe”; time is “just another dimension”;
past, present, and future all have the same “status”; change is dissimilarity
between earlier and later parts of the block universe; and so on.
A-theorists complain that this grants no special status to the present, and
that it does not admit genuine change since reality as a whole, on this picture
—the block universe—does not change. And presentists—a subgroup of the
A-theorists— complain in addition that it grants too much reality to the past
and future.
The presentist’s rival conception is that only the present is real: the truth
is from the perspective of the present moment. Part of this conception is
ontic: if an object does not exist now, then it does not exist simpliciter. For
the presentist, there simply are no such things as dinosaurs or human colonies
on Mars (though there were and perhaps will be such things). In contrast,
dinosaurs, computers, and perhaps colonies on Mars are all equally real in the
spatializer’s block universe. The pastness of dinosaurs and futurity of
colonies on Mars is locational, not ontic: the dinosaurs are located before us
—before the portion of the block universe that we occupy—and the human
colonies are located after us; in this way dinosaurs and colonies on Mars are
analogous to spatially distant objects. The presentist’s conception also has a
qualitative part: if an object does not have a feature now, then it does not
have the feature simpliciter. Since I am not in pain now, I am not in pain
simpliciter (though I was once in pain and will again be in pain). In contrast,
for the spatializer, my being in pain is equally real whether it’s located in the
past, present, or future.
Thus the presentist privileges the present, and claims to admit genuine
change since presentist reality as a whole is current reality, which changes.
This all is pretty impressionistic. To get a firmer grip on what presentism
amounts to, let’s take a closer look at the presentist’s fundamental ideology.

11.2 Presentist ideology: quantified tense logic

Tense logic is for me, if I may use the phrase, metaphysically fundamental,
and not just an artificially torn-off fragment of the first-order theory of the
earlier-later relation. —Arthur Prior3
The presentist’s fundamental ideology is that of quantified tense logic, a
logic that results from adding tense operators to predicate logic. The usual
tense operators are one-place sentence operators that are to time what modal
operators are to modality:4
operator meaning
P it was the case in the past that
F it will be the case in the future that
G it is always going to be the case that
H it always has been the case that
S it is sometimes the case that
A it is always the case that
N it is now the case that

(Some can be defined in terms of others: Sϕ↔(Pϕ∨ϕ∨Fϕ), Fϕ↔∼G∼ϕ, etc.)

The presentist’s fundamental sentences describe reality from the current
perspective, so to speak, rather than from an atemporal perspective on the
whole of temporal reality. Thus the presentist accepts ϕ↔Nϕ, for all ϕ.
“Jones sits”, Sj, is true iff Jones now sits; “Everything is sitting”, ∀xSx, is
true iff all current entities are currently sitting. When the presentist wishes to
speak of the past, she prefixes her sentences with tense operators, as in: “It
was the case that there are dinosaurs”, P∃xDx. This can still be regarded as
being from the current perspective since from that perspective, dinosaurs are
in the past. But this is not to say that the presentist regards their pastness as a
relative matter. Their pastness is absolute, since the current perspective is the
only perspective. It is true simpliciter—not relative to a “perspective”—that
P∃xDx.
Contrast this present-centric picture with the atemporal picture of the
spatializer. The spatializer does not regard his sentence ϕ as meaning that ϕ is
now the case. He regards it, rather, as a description of reality from the
atemporal perspective, so to speak. Thus he accepts “There are dinosaurs”,
∃xDx.5 If he wants to make a statement that is specific to the present, he will
restrict quantifiers or relativize predicates. Where t0 is the moment at which
he is speaking, he might say “Nothing located at t0 is a dinosaur”,
or “Jones sits at t0”, Sj t0. (Notice the symmetry between Sj
t0 and Sj t1, where t1 is some past or future time—neither is privileged.) For
the spatializer, the presentist’s statement “Jones sits”, Sj, without a parameter
for the time, makes no sense, because from the atemporal perspective, sitting
isn’t something you just do; it’s something you do at times. (This is what
those who reject temporal parts say, anyway.6) For him, the pastness of
dinosaurs is a relative matter: although they are located before his own
perspective in time, they are simultaneous with their own perspective and are
located after perspectives in the Paleozoic era; and no perspective is
privileged.
Spatializers do not admit tense operators into their fundamental ideology,
since they can describe temporal reality without them—by quantifying over
past and future entities and predicating features of them relative to times.
Spatializers may use tense operators in their nonfundamental languages, since
they can give a metaphysical semantics for the language of quantified tense
logic in their tense-operator-free fundamental language. Such a semantics
will, for example, count an utterance of P∃xDx that takes place at t0 as being
true iff some dinosaur is located before t0. Since the presentist rejects the
existence of dinosaurs, she rejects this semantics. She instead regards the
tense operators as metaphysically ultimate, as carving at the joints, and so as
not needing a metaphysical semantics. If asked what makes P∃xDx true, she
will either refuse to answer, or shrug and say: Well, it’s true because there
once existed a dinosaur! We all think something like this about negation,
after all. We do not believe in a realm or point or index, u (for “unreal”), such
that ∼ϕ iff ϕ in u.7 We think that there is not an F iff, well, there is not an F.
‘Not’ is primitive ideology (a primitive part of the best fundamental ideology,
in my view). Same for the tense operators, according to the presentist.
Presentists need their primitive tense operators. Without them, their denial
of the existence of dinosaurs would land them in the looney bin along with
the young-earth creationists. With them, they can stand tall and say that there
were dinosaurs: P∃xDx.

11.3 Is the dispute over presentism substantive?

Part of the dispute between presentists and spatializers is purely ontological.
Spatializers say, for example, that there are dinosaurs, ∃xDx, whereas
presentists reject this claim.8 Now, some suspect that there is less to this
dispute than meets the eye.9 After all, they say, everyone believes that there
are dinosaurs in the sense that there were dinosaurs; and everyone denies that
there are dinosaurs in the sense of dinosaurs located in the present; so what
proposition is under dispute?
This suspicion is an instance of the sort of metaphysical deflationism
considered in section 5.1. A sentence under dispute (“There are dinosaurs”) is
given an obviously true interpretation (“There were dinosaurs”) and an
obviously false interpretation (“There are dinosaurs located in the present”);
these are claimed to be on a par with other interpretations; and so it is
concluded that, in one way or another, the dispute is nonsubstantive. But such
deflationist challenges are mistaken if the crucial terms in the dispute carve at
the joints. The crucial term in the dispute over whether ∃xDx is ∃, the
existential quantifier; and I argued in chapter 9 for the thesis of ontological
realism: quantifiers do indeed carve at the joints. Given this thesis, presentists
and spatializers can stamp their feet and say: Our dispute is not over whether
there were dinosaurs, or whether there are dinosaurs located in the present,
but rather over whether there are dinosaurs! Since the existential quantifier
carves at the joints, this latter question is a substantive one, a question about
reality’s fundamental structure.
 Someone determined to debunk the dispute over presentism could simply
reject ontological realism. But might the debunker instead accept ontological
realism, in some form anyway, and nevertheless maintain that the dispute
over presentism is nonsubstantive?
The debunker might accept ontological realism in the sense of accepting
some sort of fundamental quantifier, but deny that the “absolute” quantifiers
assumed by the disputants carve at the joints. The dispute is allegedly over
whether there are dinosaurs. But, the debunker might say, the absolute
quantified claim “There are dinosaurs” makes no sense. The only intelligible
claims in the vicinity are temporally relativized quantified claims, claims of
the form “At time t, there are dinosaurs”. But no claim of this form (it might
be thought) is in serious doubt. Everyone agrees, for example, that in 2011
A.D. there are no dinosaurs, and that at certain earlier times, there are
dinosaurs.
The crucial claim here is that “It makes no sense to say ‘There are
dinosaurs’”, that all quantified statements require temporal qualifiers. What
does this mean and why think that it is true?
One might argue that in natural language, one cannot make absolute
claims about what there is; one must always relativize to a time. At some
deep level of syntactic analysis, a temporal parameter is always required.
This claim about natural language may be true, but it is irrelevant. The
question is whether quantification is metaphysically absolute (whether
quantified claims in fundamental languages are absolute), not whether it is
linguistically absolute (whether quantified claims in natural language are
absolute). Given metaphysical absoluteness, the dispute over presentism is
substantive, even given linguistic relativity, since the presentist and her
opponent may simply enter the philosophy room and resume their debate
using quantified claims that are stipulated to be absolute. And if quantified
claims are metaphysically relative to time, then the debate will be
nonsubstantive, even if such claims are linguistically absolute. For natural
language quantificational claims would then be analogous to actual natural
language claims about simultaneity, if spacetime is Minkowskian. The deep
syntax of such claims requires no parameter for a frame of reference, but
Minkowski spacetime demands such a parameter. As a result, natural
language disputes over simultaneity can be nonsubstantive if the answer turns
on which frame of reference one adopts. Similarly, natural language disputes
over whether there are dinosaurs would be nonsubstantive when the answer
turns on which temporal parameter is adopted.
So the attack on the substantivity of the dispute over presentism must be
metaphysical; it must claim that in fundamental languages, quantificational
claims are always relative to time.10 Leaving out the time in a claim like
‘There are dinosaurs’ would be like leaving out the orientation when telling
someone to turn left. Just as there is no such thing as just plain left, there is no
such thing as there just plain being a dinosaur. Any attempt to stipulate
unrelativized fundamental quantification in the metaphysics room would fail
because no joint in nature answers to the attempted stipulation. Given this
metaphysics, there simply is no fundamental question of whether there are
dinosaurs. The only fundamental questions are whether there are dinosaurs
relative to 2011 A.D., whether there are dinosaurs relative to times in the
Triassic era, and so on; and over such questions (it may be thought) there is
no disagreement.
There are different ways to develop this idea, since the demand for a
temporal parameter could be pinned on different sources. One idea would be
that every assertive utterance somehow requires completion by a time, in
order to result in a truth-evaluable content. Another idea—the one I will
focus on—is that fundamental quantifiers have “slots” for times: ∃t, ∀ t.
Using a quantifier without filling in the time slot would be like saying that a
certain direction is left simpliciter.
The mere fact that what there is varies in some sense over time does not
on its own favor the relativizer, since spatializers and presentists both accept
absolute quantifiers and can nevertheless account for variation. Spatializers
can say that dinosaurs are located at some times but not others, and
presentists can say that there were dinosaurs but now there are none: P∃xDx
Λ ∼∃xDx. The need for temporal supplementation is not forced on us by the
“data”; it must be a substantial metaphysical hypothesis.
And it’s an unattractive hypothesis. The simple logical ideology of
standard predicate logic—with its absolute quantifiers—systematizes
discourse in diverse domains across the sciences and mathematics.
Incorporating time into the quantifiers seems inappropriate in mathematics,
so the relativizer needs a bare quantifier in mathematics, ∃, in addition to the
temporally relativized ∃ t. His overall ideology is thus fractured. Overall
ideology is further fractured by the need to quantify into the time-slot of the
quantifiers. One might need to say, for example, that for some time, t, ∃
txDx; but how should the outer quantifier, ‘for some time’, be understood? It
seems forced (though barely consistent, I suppose) to regard it as the
temporally relativized quantifier (relativized, say, to any chosen time
It likewise seems forced to regard it as the bare quantifier
from mathematics: ∃t ∃ txDx. Least forced would be to further complicate
ideology by introducing a third sui generis quantifier, whose sole purpose is
to quantify over times; but then the ideology’s ugliness is manifest.11
These considerations of ideological simplicity, incidentally, favor the
spatializer against the presentist. The presentist adds the tense operators to
fundamental ideology in order to account for temporal variation, whereas the
spatializer needs no special logical addition. (Spatializers do need temporal,
or spatiotemporal, predicates over points of spacetime, but these are of a
piece with the spatial predicates over space that presentists need.) Presentists
will reply that the extra required ontology of past and future objects, and the
failure to privilege the present, are too high a price to pay for the
simplification in ideology. Whether the price is worth it is a question about,
in Mark Johnston’s phrase, the endgame.12
I have argued against the relativizing ideology, but now set this aside.
Even if it is correct, so that the ontological aspect of the dispute between
presentists and their foes is nonsubstantive, another aspect remains
substantive: the dispute over the correct temporal ideology. And the
relativizer has given no reason to doubt that this dispute is substantive.
Indeed, he has himself entered into this dispute. A metaphysical theory is
a theory about reality at the most fundamental level; and the very first choice-
point in putting forward such a theory is that of what ideology to use. This
choice determines the “shape of the facts” according to that theory. The
relativizer, the presentist, and the spatializer each put forward a distinctive
position on the shape of the fundamental facts since each puts forward a
distinctive fundamental ideology: temporally relativized quantifiers for the
relativizer; absolute quantifiers and tense operators for the presentist;
absolute quantifiers and no tense operators for the spatializer.
Granted, if the relativizer is right then the further dispute over whether
there are dinosaurs is nonsubstantive. But that’s not to say that presentists and
spatializers are being irrational or methodologically confused in pursuing that
further dispute—they reject the relativizer’s ideology, after all. Nor can they
be faulted for pursuing this further “doctrinal” dispute before first settling the
ideological dispute. The disputes must be addressed simultaneously since a
metaphysical theory is a “package deal” of ideology plus doctrine;
ideological and doctrinal choices interact.
Note also that the relativizer does not occupy the epistemic high ground,
for he presumably supports his relativizing ideology by the same sort of
metaphysical argumentation that presentists and spatializers use to support
their theories. Granted, he can say that if his ideology is correct, then we do
not face another hard question, that of whether presentism is correct; indeed,
we avoid a choice between two alternatives, presentism and spatializing,
which may each be regarded as unpalatable.13 These claims may or may not
prevail in the endgame. But at any rate, the fact remains that the availability
of the relativizing ideology does not provide a detour around the first-order
debate over presentism, for the road to this ideology leads straight through
the first-order debate.
Our discussion of relativization has reprised a now-familiar theme:
metametaphysical critiques tend to substitute one substantive dispute for
another. The relativizer rejects a substantive dispute over whether there are
dinosaurs; but in doing so he takes a substantive stand on fundamental
temporal ideology. Moreover, doesn’t his own ideology still allow
substantive ontological questions about dinosaurs? He denies, I have been
assuming, that ∃2011ADxDx; but couldn’t a rogue relativizer coherently think
that for all t, ∃ txDx? There are dinosaurs at all times; it’s just that for all t,
not dinosaurs are located in 2011 A.D. Given the relativizer’s stand on
fundamental ideology, the dispute with the rogue is a substantive one.

11.4 Passage
Consider next that most infamous of doctrines, “time’s passage”. Presentism
is one version of this doctrine, but let us now focus on its combination with
the denial of presentism. Defenders of this combination accept (in some
sense) the spatializer’s ontology of past, present, and future objects. But the
spatializer’s block universe, they say, is missing a crucial element: time’s
“passage”, “genuine change”. Change, they say, isn’t just variation across the
spatiotemporal manifold. There is also time’s “flow”. The block universe is
“static”.14
At first glance anyway, this view richly deserves the ridicule heaped on it
by Donald Williams—“prophets of passage” and the like—and so many
others. What does it mean to say that time “flows”? Absurdity threatens on
one hand, triviality on the other. The suggestion seems to be that time itself
moves. Does that mean that time moves with respect to some “hyper-time”?
Or merely that in one second, a time one second from now will be present?15
When time passes and things change, reality is temporally heterogeneous.
The prophet of passage insists that this is different from heterogeneity across
space. (The alleged problem with spatializing is that it collapses the two.) But
what exactly is the difference? Beware of the temptation to use temporal
vocabulary to get at the difference: “Space is static, whereas things move
through time; things genuinely change.” ‘Static’, ‘move’ and ‘change’
usually have purely temporal significance, but also have spatial senses. A
stationary sine wave spatially “moves” up and down; unlike a “static” straight
line, its height “changes” as one progresses from left to right. The italics
bespeak conviction that the temporal senses of these words are special; but in
what way?
So on the face of it, there are questions about whether time’s passage is a
coherent, distinctive, position. We can begin to answer them by focusing on
fundamental ideology. A prophet could formulate a coherent, distinctive
philosophy of time by advocating a distinctive fundamental ideology. For
example, whereas the spatializer’s ideology is that of standard predicate logic
(including spatiotemporal predicates), the prophet might advocate the
ideology of predicate tense logic. If “time passes” just means that the tense
operators carve at the joints, then the view is coherent, nontrivial, does not
imply the existence of hyper-time, and sharply distinguishes temporal from
spatial change: the former involves irreducibly tensed claims, whereas the
latter involves the instantiation of properties in distinct parts of a spatial
manifold.16
Thus understood, the dispute over time’s passage is simply the ideological
half of the dispute over presentism. Recent philosophy of time has been
kinder to the dispute over presentism than to the dispute over time’s passage,
probably because the former has an ontological aspect as well, and
ontological disputes are regarded with less suspicion than purely ideological
disputes. But this attitude is misguided. Many of the most important
metaphysical disputes—over, for example, whether reality is ultimately
modal, causal, nomic …—are ideological, not ontological. So there is no
reason to distrust the dispute over time’s passage if it, too, is ideological.
 Nevertheless, questions remain. For one thing, why does the prophet
regard the tense operators as fundamental? Since the prophet does not follow
the presentist in rejecting past and future objects, he agrees—one might think
—that “It was the case that ϕ” is true if and only if ϕ is true before the
utterance. So why not regard such biconditionals as providing a metaphysical
reduction of tense operators?17
Moreover, there is a question of how primitivism about tense operators
fits the intuitive picture of time’s passage. Why does this primitivism supply
a better account of “genuine change”? And how does it fit the traditional
rhetoric of the prophets? Think, for example, of McTaggart’s (1908) criticism
of Russell’s (1903, section 442) spatializing conception of change. For
Russell, a poker changes from hot to cold by being hot at one time and cold
at another. But, McTaggart complains, on Russell’s view the facts don’t
change, since it’s always the case that the poker is hot at the first time and
cold at the second. How is this complaint underwritten by tense-operator-
primitivism? The complaint, moreover, is extremely perplexing. If the poker
is hot at one time but not at another, doesn’t it follow that “the facts change”?
If not, what is this notion of fact on which McTaggart rests so much weight?
To answer these questions we must delve more deeply into the mindset of
the prophet of passage, and into further questions about fundamental
ideology. Doing so will reveal the existence of a number of different versions
of the doctrine of time’s passage.

11.5 Williamson on saturation and contingency

Let’s begin by shifting focus to a related topic: modality. Consider this
argument by Timothy Williamson (2002) against David Lewis’s (1986b)
modal realism:
There is genuine contingency in how things are only if, once values have been assigned to all variables,
the resulting proposition could still have differed in truth-value. It is not contingent that Blair was
Prime Minister in 2000 in @ and that he was not Prime Minister in 2000 in w. What is contingent is
simply that Blair was Prime Minister in 2000. Its contingency requires it not to have a variable waiting
to be assigned a world. The reply ‘But contingency just is variation in truth-value with variation in the
value of the world variable’ betrays a failure to grasp what contingency is.
According to David Lewis’s modal realism, contingency consists in differences between possible
worlds, which are conceived as equally real, mutually disconnected spatiotemporal systems. Consider
the common sense claim ‘It is contingent that there are no talking donkeys’
If one interprets the quantifier as unrestricted, modal realism
makes the claim false by making its first conjunct false: the modal realist holds that there really are
talking donkeys, in spatiotemporal systems other than ours. For modal realism to make the claim true as
uttered in the actual world, one must interpret the quantifier as implicitly restricted to the objects in a
world … The restricted quantifier is given an implicit argument place for a world… Even if there are
mutually disconnected spatiotemporal systems such as Lewis postulates, they are not the distinctive
subject matter of modal discourse. They are simply more of what there is, about which we can ask
genuinely modal questions: for instance, whether there could have been more or fewer spatiotemporal
systems than there actually are.

As I read him, Williamson isn’t just offering the familiar complaint that if
Lewis’s worlds existed, they’d be part of actuality. (Or maybe he’s offering a
deeper version of that complaint.) The argument’s engine is a notion that I’ll
call saturation— the notion of “all variables and parameters being filled in”.
Williamson’s idea is that the possible-worlds theory of modality is
incompatible with the evident fact that saturated statements are capable of
contingency. According to Williamson, the possible-worlds theory delivers
contingency only for something like an open formula: ‘In w, ϕ’, where w is a
free variable. This is said to be contingent because it is satisfied by some
worlds and not others. But this formula is unsaturated. What are saturated, for
the possible-worlds theorist, are statements like ‘In world a, ϕ’, where a is a
name of a particular world. But such statements are not contingent. More
fully, the argument maybe interpreted thus:18
1. For some ϕ concerning the nonexistence of talking donkeys, It is contingent that ϕ is true.
2. It is contingent that ϕ is true only if: ϕ is true and saturated and possibly, not: ϕ is true.
3. If modal realism is true, then ‘There are no talking donkeys’ (unrestricted quantifier) is not true.
4. If modal realism is true, then ‘Possibly, not: there are no talking donkeys in @’ is not true.
5. ‘There are no talking donkeys in w’ (where w is a variable) is not saturated.
6. If modal realism is true, then no other ϕ concerning the nonexistence of talking donkeys is true,
saturated, and such that possibly not: ϕ is true.
7. Therefore, modal realism is not true.

If modal realism doesn’t deliver genuine contingency, then what would?

A natural answer for Williamson would be the conjunction of “actualism”
and “modalism”. According to actualism, the facts are all from the
perspective of the actual world. Below I will cash “facts” out as saturated
statements; thus the claim is that actual-world-perspective statements like
“There are no talking donkeys” are saturated. According to modalism, the
modal operators carve at the joints. These operators can be used to
characterize the contingency of the nonexistence of talking donkeys: “There
are no talking donkeys and ♢(there are talking donkeys).”
Williamson has hit on a deep clash between modal realism and our
ordinary notion of contingency. Intuitively, genuine contingency requires
contingency in the facts, not just contingency in unsaturated fact-parts. As
we’ll see next, one can construe the complaint that the block universe is
“static” in an analogous way.

11.6 Change and saturation

According to the spatializer, the “atemporal perspective” is fundamental. The
statements from this perspective are, intuitively, changeless: ‘Ted is sitting at
12:00 AM EST on 1/1/10’, ‘World War 2 is occurring in 1943’, ‘Humans
learned sea travel before they learned space travel’, and so on. The prophets
of passage then complain that this makes reality static. I propose to
understand this complaint as being analogous to Williamson’s complaint
about modal realism. The complaint, thus understood, is that the spatializer
cannot allow for genuine change in the sense of change in saturated
statements. The spatializer’s saturated statements are those with “filled in
temporal parameters”, such as ‘Ted is sitting at 12:00 AM EST on 1/1/10’,
and these statements do not change. Only statements that omit temporal
parameters, such as ‘Ted is sitting’, change; but they’re not saturated. So
there is no genuine change in the block universe.
As a spatializer, I reject this argument. My metaphysical crystal ball
doesn’t reveal this sort of genuine change. (Nor does it reveal Williamson’s
genuine contingency, I have to say.) But my purpose here isn’t to take sides;
it’s rather to dig more deeply into this notion of genuine change, and into the
mindset of the prophets of passage.
If spatializing doesn’t provide genuine change, then what would? The
defenders of the Williamsonian argument should say: the doctrine of passage.
So far I’ve been construing this doctrine as the claim that the tense operators
are fundamental. This is analogous to modalism in the previous section. But
securing genuine contingency also involved actualism, the claim that the facts
are those from the perspective of the actual world—that statements lacking a
parameter for a possible world, such as ‘There are no talking donkeys’, are
nevertheless saturated and hence capable of genuine contingency. This
suggests that in addition to holding that the tense operators are fundamental,
the defenders of passage should embrace “temporalism”, the temporal analog
of actualism, according to which the facts are those from the perspective of
the present moment—that statements lacking a parameter for a time, such as
‘Ted is sitting’ and ‘World War 2 is occurring’, are nevertheless saturated
and hence capable of genuine change. Genuine change, on this view, amounts
to the truth of tensed claims of the form PϕΛ∼ϕ, ϕΛF∼ϕ, and so on.
If we think of saturated statements as McTaggart’s “facts”, then this view
vindicates his claim that the facts themselves change—we have our
connection to the traditional rhetoric.
As we will see below, there is more than one view along these lines. But
notice how much weight is being put on the notion of saturation, and related
notions like that of a “perspective”. Before going further we must satisfy
ourselves that talk of saturation is legitimate, and that there are genuine issues
about which statements are saturated.

11.7 Talk of saturation is legitimate

The spatializer and the defender of passage disagree, I have said, over which
statements are saturated. Consider, for example:

(T) Ted is sitting

(T-rel) Ted is sitting at 12:00 AM EST on 1/1/10.

The defender of passage thinks that (T) is saturated (and can nevertheless
change), whereas the spatializer thinks that only (T-rel) is saturated (and he
supplies no notion of change for such statements). But in what sense must a
spatializer think that (T) is “unsaturated”, “missing a parameter”?
Is it because (T) doesn’t correspond to a fact? Earlier I associated one
sense of ‘fact’ with saturation in order to formulate actualism and
temporalism. But my intent was to use saturation to explain that sense of
‘fact’; it’s doubtful that there is any notion of facthood that can be used to
explain saturation. The ordinary notion of facthood cannot do the job. It does
little violence to ordinary usage if we define a ‘fact’ as a property of times;
but under this definition, (T) does correspond to a fact, namely the property
of being a time at which Ted is sitting. Under this definition, facts obtain
relative to times; fact F obtains relative to t iff t instantiates F. Of course, one
could introduce a different sense of ‘fact’ on which obtaining is nonrelative;
but the question of whether (T) is saturated had better not turn on the merely
terminological question of what we mean by ‘fact’. One might reply that the
question isn’t terminological if we take ‘fact’ in a joint-carving sense, but it is
doubtful that there is any such sense; and anyway, the debate over passage
surely doesn’t require its acceptance. (I would reply similarly to the claim
that (T) is unsaturated because it doesn’t express a proposition: we could
define up propositions that obtain relative to times; the existence of a joint-
carving sense of ‘proposition’ is doubtful; and anyway the debate over
passage surely doesn’t turn on its existence. See also section 6.3.)
One might say instead that (T) is unsaturated because it isn’t true—true
simpliciter that is. Only saturated statements like (T-rel) are true. But now we
have defined saturation in terms of the notion of a sentence being true
simpliciter, and the latter notion seems to be in the same boat as the notion of
saturation. Furthermore, deflationists about truth cannot take this line. For
even spatializers must admit that one can assert sentences like (T) without
specifying a time (the time is somehow “supplied by context”); and given
deflationism, this requires asserting that (T) is true.
Assume that spatializers will indeed admit that (T) is true (perhaps
because they accept deflationism about truth). This introduces a further
problem with the notion of saturation: how do we single out an appropriately
metaphysical sense of saturation? The fact that (T) is true means that it is
semantically complete, so to speak. A spatializer might also allow that (T) is
grammatically and psychologically complete: (T) is a grammatical sentence
given our best syntactic theory (unlike ‘loves Sally’) despite the absence of a
syntactic element for a temporal parameter, and one can think (T) without
thinking a temporal parameter (again unlike ‘loves Sally’). It’s only in a
metaphysical sense that the spatializer thinks (T) is incomplete: it is the
spatializer’s metaphysics of time—not psychology or syntax or semantics—
that demands a temporal parameter. But what is this distinctively
metaphysical notion of saturation?19
There are, therefore, questions about what saturation might amount to.
Nevertheless, reflection on some further examples will make it intuitively
clear that there is indeed a distinctively metaphysical notion of saturation. As
we will see in the next section, there are difficult questions about the
underlying metaphysics of saturation, but that should not lead us to doubt that
some such notion is intelligible and important.
Suppose you’re facing the Empire State Building from the south. To the
left of the ESB is the Hudson River. Now, the fact in question is really that
the Hudson is to the left of the ESB relative to your orientation. It’s not to the
left of the ESB, full stop; there’s no such thing as being to the left, full stop.
That’s not to deny that the Hudson is to the left of the ESB; it’s just to say
that “The Hudson is to the left of the ESB” isn’t “complete” until an
orientation is filled in.
According to me, God does not exist. According to Dean Zimmerman,
God does exist. But that doesn’t exhaust the facts in this case. There’s
another fact: that God doesn’t exist. (Were Dean writing this paragraph, of
course, the previous sentence would read differently.) When I say this, I’m
not merely reporting the claim with which this paragraph began, keeping the
relativization to my beliefs implicit. The claim that God doesn’t exist needs
no completion concerning my beliefs. God doesn’t exist, full stop.
Contrast this with the case of the Empire State Building. We don’t want to
say that, in addition to the relativized facts: the Hudson is to the left of the
ESB relative to this orientation, and the Hudson is to the right of the ESB
relative to that orientation, there is an unrelativized fact: The Hudson is to
the left of the ESB, full stop. There simply is no such fact. Whereas, in
addition to the relativized facts God exists according to Dean and God does
not exist according to Ted, there is an unrelativized fact about whether God
exists, full stop.
Few would accept unrelativized facts of directions or reject an
unrelativized fact about God’s existence; but in other cases there is room for
disagreement. In addition to facts about what is morally wrong according to
various cultures, are there further facts about what is wrong, full stop? Moral
realists will answer yes; moral relativists no (though they might grant that
nonrelativized utterances of the form ‘X is wrong’ can be true, since they
attribute wrongness-in-the-speaker’s culture.)
It seems clear from these examples that the notion of saturation makes
sense. There is no need for a further parameter to get a truth-value out of
‘God exists’; there is need for a further parameter with ‘The Hudson is to the
left of the ESB’; ‘X is wrong’ may or may not need a further parameter.
Also, the need for a parameter is metaphysical, not psychological or
linguistic. I don’t know whether there’s a syntactic or semantic or
psychological demand for a parameter of orientation in ‘the Hudson is to the
left of the ESB’. But at any rate, think of other cases. A moral relativist
needn’t think that there’s a linguistic or psychological need for a parameter
specifying a culture in ‘X is wrong’. Given the special theory of relativity, we
need to fill in a frame of reference to get a truth-value out of ‘e1 and e2 are
simultaneous’, but neither the mind nor the language organ demands it.
There’s something metaphysically distinctive, then, about all parameters
being filled. When all parameters are filled, we can call the result a fact. And
this is a metaphysical—not linguistic or psychological—notion of fact.
We can think of all this in terms of “perspective”. Among all the belief-
perspectives, one is distinguished (the one corresponding to someone whose
beliefs are all true).20 Concerning left–right, there is no distinguished
perspective. The Hudson is to the left of the ESB with respect to one
orientation, to the right of the ESB with respect to another; neither orientation
(perspective) is privileged.

11.8 The metaphysics of saturation

The examples of the previous section make a compelling case that there’s
some metaphysically distinctive notion of saturation—and relatedly, some
metaphysically distinctive notion of fact, and some metaphysically distinctive
notion of the correct perspective. But the underlying metaphysics of these
notions is far from clear.
Begin with a three-place relation like between, and start filling in its slots:

— is between — and —
a is between — and —
a is between b and —
a is between b and c.

The idea of saturation is that the final case differs importantly from the first
three. There’s something metaphysically distinctive about all of the argument
places of a relation being filled in. In my terms, the distinction between the
final case and all the previous cases is some sort of joint in nature. Or again,
where F is a one-place predicate, R is a two-place predicate, and a and b are
particulars, there is something in common between a case where something is
F and a case where a bears R to b; and this something in common is not
shared by the property F, the relation R, the property of bearing R to b, and so
on. Some kind of joint in nature separates (saturated) facts, on the one hand,
from (unsaturated) properties and relations on the other. But what kind of
joint?
We might appeal to the existence of facts or propositions to make the
distinction, but the existence of facts or propositions at the fundamental level
is doubtful. And anyway, the intuitive idea requires no such reification: it is
that utterances of “New York is between Washington and Boston”, “The
Phillies play in the National League”, and “Ted is a philosopher” are to be
grouped together (at a very high level of abstraction!), and are to be
distinguished from “New York is between”, “in the National League”, “Ted”,
and so forth; and that this grouping is no arbitrary choice on our part, but
rather reflects the world’s nature.
It’s initially tempting to invoke a joint-carving notion of “argument
place”. The idea would be that a distinguished description of the world will
include a specification, for each relation, of all of its argument places. Then
we could define a saturated sentence as one that specifies a value for each
argument place of each relation that it concerns.21 But this approach requires
reifying relations. Worse, it apparently requires reifying argument places.22
Also the approach recognizes only one potential source of unsaturation: a
relation with an unfilled argument place. But there are other potential
sources. For example, some claim that quantifiers have argument places
(places for times as with the relativizer of section 11.3, or for sortal
properties).
We are searching for a joint in nature that unifies “complete” sentences
like “∃xFx” and “Rab”, as against fragments like “F” and “Ra”. The
problem is that there seems to be no particular linguistic expression to which
this joint corresponds. My solution is to think of the joint as not
corresponding to any particular linguistic expression at all, but rather to a
linguistic category: that of the complete sentence in a fundamental language.
In a fundamental language, a language in which the category of sentence
carves at the joints, sentences are always “metaphysically complete”—
saturated. In languages like English, which includes metaphysically
incomplete as well as metaphysically complete sentences (‘Ted is sitting’ as
well as ‘Ted is sitting at noon on May 17, 2010’), the category of sentence
does not carve at the joints.
Think of it this way. Sentences play a distinctive and central role in our
cognitive and linguistic lives, a role that Frege explored in “Thought”. It is
with sentences that we make assertions, express our beliefs, desires,
conjectures, and so forth. (We do these things with complete sentences, not
fragments like “between”, “ a is between”, and so on.) It is natural to think
that there is an aspect of the world’s structure that we are getting at with this
role. We would get at this aspect perfectly only if we eschewed all implicit
parameters; and we do not do this in English (sentences like ‘Ted is sitting’
include an implicit temporal parameter). Nevertheless, we get at this aspect
reasonably well in English. Parameters in English must normally be explicit;
we cannot—without ellipsis—assert ‘between’, ‘a is between’, and so forth.
Such exceptions as there are, are in the minority, and they are rule-governed.
Asking whether the category of sentence carves at the joints is less
familiar than asking whether a given predicate carves at the joints, but the
underlying idea is the same. When we ask whether quantifiers, modal
operators, or causal predicates carve at the joints, we are asking whether
certain facets of our conceptual scheme latch onto reality’s distinguished
structure. Whether our discourse using the facet is objective, semantically
determinate, worthy of attention, and so on, turns on this matter of joint-
carving. These questions do not go away when we turn to more ubiquitous,
abstract, and pervasive facets of our conceptual scheme. Just as we can ask
how well we are getting at the nature of things when we quantify, modalize,
or speak of causation, we can ask how well we are getting at the nature of
things when we accord a cognitively central role to the category of sentence.
Might there be no joint in nature in the neighborhood of our use of
sentences? Metaphysicians must always live with the possibility of a bad
mismatch between how we think and how the world really is. But in this case
the realization of this possibility would be dire. Our cognitive lives revolve
around sentences. To claim that all of our asserting, believing, desiring, and
so forth is just badly on the wrong track—not just the small imperfection that
results from our use of implicit parameters, but a massive mismatch with
reality resulting from there simply being no joint in the vicinity—would be
far more of a skeptical position than claiming that we are on the wrong track
with quantification (say), or with some particular predicate, or even with
predication itself. I can’t even imagine what it would be like to reorient my
cognitive life so as not to revolve around sentences.
Hitherto I have defined a “fundamental language” as one in which all the
primitive expressions carve at the joints. Let us now expand this formulation,
to require that the category of sentence also carves at the joints.23 We may
then introduce a distinction between saturated and unsaturated sentences in
nonfundamental languages—at least when those languages have an
appropriate sort of metaphysical semantics. Suppose, for example, that the
sentences of some nonfundamental language divide into two sorts, depending
on the relationship between the sentence S and its metaphysical truth-
condition µ(S). If S is of the first sort, then S and µ(S) share logical form. But
if S is of the second sort then µ(S) is logically richer than S; in particular, µ(S)
contains a part that is syntactically like S, but also a further part that
systematically depends on some feature of the context in which S is uttered.
(This further part of µ(S) has no syntactic counterpart in S.) We may then call
the first sort of sentences saturated and the second sort unsaturated.24
For example, suppose that the spatializing metaphysics of time is true, and
suppose the metaphysical truth-condition for ‘Ted is sitting at noon on May
17, 2010’ has the form Ted is ϕ at α (ϕ and α are translations into a
fundamental language of ‘is sitting’ and ‘May 17, 2010’, respectively). This
metaphysical truth-condition is syntactically parallel to ‘Ted is sitting at noon
on May 17, 2010’, and so that sentence is saturated. Suppose further that the
metaphysical semantics for an utterance u of ‘Ted is sitting’ has the form
Ted is ϕ at β u , where β u names the time of u. Then ‘Ted is sitting’ is
unsaturated: its metaphysical truth-condition consists of a portion that is
syntactically like it, but contains in addition a further systematic enrichment:
at β u .25
The failure of the category of sentence in English to carve at the joints is
due, intuitively, to the fact that this category is metaphysically heterogeneous.
The category includes both dated sentences, which match their metaphysical
truth-conditions in syntactic structure, as well as present-tense sentences
which do not. (Other kinds of context-sensitivity also contribute to the
heterogeneity.) This metaphysical imperfection of English is of an unfamiliar
sort. More familiar sorts result from the English lexicon containing particular
words that fail to carve at the joints, whereas this less familiar sort results
from a structural feature of the metaphysical semantics of English.
Applying the notion of joint-carving to linguistic categories like the
category of sentence requires us to revisit the question of how to regiment
structure-talk. Section 6.3 distinguished two regimentations. On one, the
primitive operator attaches to individual expressions we are querying for
joint-carving: (∃) (is negatively charged), and so on. This regimentation
apparently cannot formulate the claim that the category of sentence carves at
the joints, since there is no particular expression to be queried. On a second
regimentation, attaches to full sentences, but the sentences contain dummy
variables taking the place of sentence-parts we are not querying. Thus “
” is how we regiment the claim that existential quantification carves
at the joints. Given this regimentation we may formulate the idea that the
category of sentence carves at the joints by attaching to a dummy
sentence-variable: (P) The intuitive idea is that we are querying the
category of the dummy variable P—the category of the sentence.
Since this approach to regimentation seems a bit artificial, it’s worth
mentioning an alternative, “quietist” approach to introducing the idea of a
fundamental language. On the quietist approach, we no longer speak of the
category of sentences carving at the joints. Instead, we say that a fundamental
language is one that results from raising the standards—in a distinctively
metaphysical way—for making assertions and doing other things with
complete sentences. When we raise the standards in this way, we disavow
various typical mechanisms for tacitly supplying parameters, such as the
current time for tensed sentences, quantifier domain restriction, comparison
classes for gradable adjectives, and so on. Whereas normally we are willing
to say things like “It is now raining”, “everyone is at the meeting”, “Allen
Iverson is short”, and so forth, we refuse to say such things when attempting
to speak a fundamental language. We say only fully articulated things like “It
is raining in New York City at noon on May 17, 2010”, “Every member of
the NYU Philosophy Department was at the philosophy faculty meeting on
May 6, 2010”, “Allen Iverson is short for a basketball player” and so on. This
raising of standards is parallel to other sorts of standard-raising in
metaphysics room. Ordinary talk of properties and similarity, for example, is
flexible and context-dependent: we are apt to speak of things as sharing some
property in common when they share nearly any salient feature, never mind
whether that feature carves at the joints. In the metaphysics room, on the
other hand, one speaks of shared properties only when the shared feature
carves at the joints. Relatedly, ordinary quantification might plausibly be
regarded as not being joint-carving; outside of the philosophy room we’ll say
that there is a way to win this chess match or that we have more in common
with our college friends than with our high school friends, whereas in the
metaphysics room we will say these things only if we think that there really
are—in the joint-carving sense—such entities as ways or things-in-common.
So: instead of introducing the fundamental language by means of an
explicit stipulation (“The category of sentence is to carve at the joints”), on
the quietist approach we instead introduce it implicitly, by exhibiting the
raised standard. The hope is that by doing so, the semantics of the resultant
language will disallow all unarticulated parameters. If we do this, then
assuming that reality does indeed contain structure in the vicinity of
saturation (however that might be understood), it would take a perverse
interpreter indeed to interpret our sentences using implicit parameters. It must
be admitted, though, that the quietist approach would be a major departure
from the approach to joint-carving that I have been developing in this book,
since it gives up on the idea that one can characterize joint-carving by means
of a distinctive addition to ideology.
The reader may have a sense that we have gone off the rails. To be honest,
I share that sense. The claim that the category of sentence carves at the joints,
for example—and that it may be regimented as (P)—strains to the
breaking-point my intuitive grip on the notion of joint-carving. But on the
other hand, it’s evident from examples that there just is a metaphysically
significant notion of saturation. I invite the skeptical reader not to simply
dismiss the issue, but rather to join my struggle to make sense of this notion,
and perhaps come up with something better.

11.9 Varieties of passage

However exactly it is understood, the idea of saturation can be used to
distinguish amongst various varieties of the doctrine that time passes. All
varieties agree that the tense operators are metaphysically basic, but they
disagree over what objects exist, and in what sense the “current perspective”
is fundamental.
The first variety is just presentism. We have already discussed presentism,
but now we can clarify what it means to say that “The presentist’s
fundamental sentences describe reality from the current perspective.” Not
only does this mean that ϕ↔Nϕ is true, for any sentence ϕ; it also means—
we can now say—that present-tense sentences like ‘Ted is sitting’ and
‘Someone is sitting’ are saturated. Intuitively: the facts are from the current
perspective.
The second variety I call Williamsonian passage because it is inspired by
(albeit independent of) Williamson’s views on modality.26 This view is like
presentism except that it claims that objects exist permanently. More exactly,
to the presentist’s i) primitivism about tense operators; and ii) claim that the
current perspective is fundamental, it adds iii) the claim that all instances of
the temporal Barcan and converse Barcan schemas are true:
 Williamson himself accepts the modal Barcan schema, and
its converse, These lead to surprising ontological conclusions.
To use Williamson’s example, Wittgenstein might have fathered something,
so, by the Barcan schema, there is something that Wittgenstein might have
fathered. Given plausible essentialist assumptions, this something couldn’t be
any ordinary object. (No person could have been fathered by Wittgenstein
since Wittgenstein in fact fathered no one, and no one could have had a father
different from his or her actual father; and no electron or mountain or …
could have been fathered by Wittgenstein since no electron or mountain or …
could have been fathered by anything.) According to Williamson this
possible Wittgensteinian offspring is a “merely modal” entity: its only
positive properties are modal properties, such as the property of possibly
being fathered by Wittgenstein.27 It doesn’t have a mass, or a charge, or a
smell, or a color. It isn’t even in space, and thus is “nonconcrete” in one
sense.
Analogously, the defender of Williamsonian passage accepts the (current)
existence of extraordinary objects. Since there once existed dinosaurs
(P∃xDx), by the past Barcan schema there (now) exist things that were
dinosaurs (∃xPDx). These “past dinosaurs” are not dinosaurs (the current
perspective is fundamental for the Williamsonian, remember; and nothing is
now a dinosaur); they are rather things that were dinosaurs. Since no ordinary
current object was once a dinosaur, these past dinosaurs are extra entities, in
addition to the ordinary objects that the presentist accepts. Their only positive
properties are temporal properties like previously being a dinosaur.28 Like
Williamson’s possible objects, they are nonconcrete in the sense of
(currently) not being located in space.
In a sense, the Williamsonian accepts the spatializer’s ontology. But since
the Williamsonian takes the current perspective to be fundamental, he
disagrees with the spatializer’s descriptions of those objects; he rejects, for
example, the spatializer’s claim that there are dinosaurs (∃xDx). Thus the
Williamsonian has a ready answer to the question: given that you accept the
spatializer’s ontology, why do you take the tense operators as basic? Since
the Williamsonian does not take the atemporal perspective to be the
fundamental one, he does not take his objects to have the features attributable
from that perspective (such as the feature of being a dinosaur); but it is
precisely these features that are needed to provide metaphysical truth-
conditions for tensed sentences. For example, the spatializer’s reductive
truth-condition for an utterance at t0 of ‘There were dinosaurs’ is ‘There are
dinosaurs that are located before t0’; but the Williamsonian rejects this truth-
condition since he denies that there are dinosaurs. The only existential claims
in the neighborhood that he accepts are claims like ∃xPDx which, of course,
cannot be used in a reduction of tense.29
(Notice that the difference between Williamsonians and spatializers over
which perspective is fundamental emerges even in connection with the
present. I am currently sitting, so the defender of Williamsonian passage will
say that ‘Ted is sitting’ is saturated. But the spatializer will deny this—
continuing to assume no temporal parts—since it is missing a temporal
parameter.)
The third variety is the “moving spotlight” view. It may be introduced by
contrast to Williamsonian passage. The Williamsonian admits the existence
of “past and future objects” but says that their only features are temporal. Past
dinosaurs aren’t dinosaurs (though they were), past noises make no sound
(though they did), and so on. The defender of the spotlight theory also
embraces past and future objects, but she accepts a “fuller” conception of
these objects than the Williamsonian. According to her, these objects have all
the features that the spatializer thinks they have. We can put this by saying
that the spotlight theorist joins the spatializer in accepting the block universe.
But the spotlight theorist accepts something in addition: a joint-carving
monadic property of presentness (or instead a joint-carving one-place
predicate ‘is present’), which is possessed by just one moment of time, and
which “moves”, to be possessed by later and later times, as time passes. This
motion, on my construal, amounts to tensed facts about which moment of
time possesses the property.
For the moving spotlight theorist, it is an external yet temporal perspective
on the block universe that is fundamental. More carefully: sentences such as
‘Ted is sitting at 12:00 AM EST on 1/1/10’ and ‘World War 2 is occurring in
1943’ are saturated. Sentences like ‘Ted is sitting’ and ‘World War 2 is
occurring’ are not saturated.30 In addition, statements of the form ‘Time t has
presentness’ are saturated. Presentness is possessed absolutely; exactly one
time has presentness simpliciter; all other times lack presentness simpliciter.
Finally, the spotlight theorist accepts primitive tense operators, which
generate saturated sentences when prefixed to saturated sentences. Using the
tense operators, the spotlight theorist can describe the change in which
moment has the property of presentness, by means of sentences like: ‘There
is a time, t, such that t does not have presentness but P(t has presentness).’
(Note that the change described here is “genuine” change—change in the
truth of a saturated statement.) Describing this change— the “motion of the
spotlight”—is, in fact, the only purpose of the tense operators in the spotlight
theory: tense operators, according to the spotlight theory, have no effect when
prefixed to statements that do not concern presentness. For any such sentence
ϕ, ϕ is true iff is true, where is any of the tense operators we have
discussed.
So intuitively, the facts are those that one would see if looking at the block
universe from an external perspective. From this perspective, one would see
everything that the spatializer believes in, but one would also see a single
time “lit up”. That is the time that possesses presentness. Further, if one
continued to look from this atemporal perspective—“continued” in the sense
of the tense operators—everything would continue to look exactly the same
except that the lit time would move from past to future.
Nonpresentist defenders of passage face the question of why they accept
primitive tense operators, given that they accept past and future objects. As
we saw, the Williamsonian had a quick answer: since he rejects the atemporal
perspective, he must reject the spatializer’s reduction of the tense operators.
The situation is very different with the spotlight theorist. For example, she
accepts that there exist dinosaurs located before 2011; but this is the
spatializer’s proposed truth-condition for an utterance in 2011 of ‘There were
dinosaurs.’ More generally, the spotlight theorist can accept the spatializer’s
reduction of tense for all tensed statements except those concerning
presentness. For the latter, there can be no reduction. “1776 once had
presentness” cannot be analyzed as saying that 1776 has the property of
presentess relative to itself, since presentness is a monadic property, and is
not had relative to anything. 2011 has presentness simpliciter; nothing else
has presentness; there are no other nontensed facts about presentness; hence
we cannot analyze the tensed claim that “1776 once had presentness” in
nontensed terms.
 So the spotlight theorist can say something to justify her primitivism about
the tense operators. But one might wonder whether it’s convincing. The tense
operators play no role in her theory other than characterizing change in the
possession of the monadic property of presentness. And the only reason for
invoking this property at all is to be able to say that there is genuine change
in which moment is present. But notice that the spotlight theorist does not
admit genuine change for anything else! For her there is no genuine change in
whether I am sitting, or in whether there are dinosaurs, or in whether a war is
occurring, since her account of these matters is identical to the spatializer’s.
All that genuinely changes is which moment has presentness. Is securing this
smidgen of genuine change worth the postulation of primitive tense?
A spotlight theorist could admit more genuine change by adopting a
hybrid theory according to which more statements are treated in the presentist
rather than spatializing way. The spotlight theorist might, for example, treat
pain the way a presentist would. On this view, statements of the form “α is in
pain” are saturated despite lacking a temporal parameter; indeed, no
fundamental statements about pain include a temporal parameter. Moreover,
the tense operators are not inert when prefixed to statements about pain; “P(α
is in pain)” is not equivalent to “α is in pain”. Sentences of the form “α is in
pain” can be saturated and true and temporary in the sense that “P(α is not in
pain)” is also true; hence there is genuine change in the matter of pain, not
just the matter of which moment is present.31 (Depending on how many other
properties are treated in this way, the spotlight theorist might do away with
the property of presentness altogether.) But notice that following this course
to an extreme would turn the position into Williamsonian passage.
There is a powerful epistemic objection to the moving spotlight theory
(and to related theories such as the “growing block” theory, to be discussed
below).32 We believe that we exist in the present; indeed, we take ourselves
to know this. But given the spotlight theory, there are ever so many people,
with similar evidence to our own, who also think they are in the present but
are wrong—they’re wrong because the times at which they are located do not
have the monadic property of presentness. George Washington, for example,
thinks in 1776 that 1776 is present; we think, here in 2011, that 2011 is
present. We cannot both be right, since the property of presentness is
monadic and possessed by only one moment. And our evidence is no better
than Washington’s (we see flowers brightly blooming in 2011; he sees
flowers brightly blooming in 1776, and so on), so it’s hard to believe that
we’re more likely to be right than Washington. Indeed, it seems likely that
we’re both wrong, since 1776 and 2011 are merely two of the infinitely many
times, only one of which has presentness. The spotlight theory leads to
skepticism about whether we’re in the present.
Notice that Williamsonian passage is immune to this objection. The
objection turns on the fact that the spotlight theorist thinks that many people
such as Washington i) think they’re in the present; and ii) are wrong because
the times at which they’re located don’t have presentness. But the
Williamsonian rejects both i) and ii). He rejects i) because Washington, for
example, is a noncrete object, and can’t think anything at all. (What’s true is
that he was a concrete object, and did think that he was in the present.) And
the Williamsonian rejects ii) because he rejects the spotlight theorist’s
property of presentness.33 The epistemic objection doesn’t work against
Williamsonian passage (not that it was intended to) because it’s designed to
refute views that accept the spatializer’s atemporal perspective, but then add
to it some notion of the objective present. Williamsonian passage doesn’t do
this; rather, it rejects the atemporal perspective. In this respect it’s more like
presentism (indeed, it survives the argument for the same reason that
presentism does).
The fourth variety of time’s passage is a little weird. It says that there are
two perspectives on reality that are equally fundamental: the current
perspective and the atemporal perspective. From the current perspective one
says what the presentist says, but one can make a “gestalt switch” to the
atemporal perspective and say what the spatializer says.
The root problem for nonpresentists who accept primitive tense operators
is that of how to combine the atemporal perspective that is apparently
demanded by quantification over past and future entities with the current
perspective that is apparently demanded by the tense operators. The
Williamsonian solves the problem by rejecting the atemporal perspective but
accepting the Barcan and converse Barcan schemas. The spotlight theorist
solves the problem by rejecting the current perspective but applying the tense
operators to certain statements from the external perspective on the block
universe (statements about presentness). The gestalt-switcher solves the
problem by embracing both perspectives.34
I have been cashing out talk of “perspectives” in terms of saturation, and
saturation in terms of truth in a language in which the category of sentence
carves at the joints. I cash out this second spotlight view, therefore, as the
claim that there are two joint-carving categories of sentence; nature has two
joints in the vicinity of a metaphysically complete statement, not one.
(Compare jadeite and nephrite.) There are, therefore, two equally good
languages one could speak in the metaphysics room. And so there are two
different sorts of saturation: call them atemporal saturation and tensed
saturation.35
Thus there are two fundamental descriptions of the world. One is an
atemporal description of reality from the atemporal perspective; the other is a
tensed description of reality from the current perspective. Neither description
is incomplete; rather, the notion of giving the entirety of the fundamental
facts is just bifurcated.
What exactly can be said from these two perspectives? What statements
are atemporally saturated, and what statements are tense-saturated? And how
do the tense operators interact with these statements? The answer depends on
temporal parts.
Suppose, first, that temporal parts are accepted. In that case, from the
atemporal perspective the gestalt-switcher speaks exactly like a spatializer
who accepts temporal parts. Statements like ‘Ted0 is the (instantaneous)
temporal part of Ted at time t0’, ‘Ted0 is sitting’, ‘Ted is sitting at t0’, and so
on, are atemporally saturated. Moreover, in this context, the tense operators
are inert. From the tensed perspective, on the other hand, the spotlight
theorist speaks like a presentist who accepts temporal parts. If t0 is the current
time, then ‘Ted is sitting’ and ‘Ted0 is sitting’ are both tense-saturated and
true. If nothing is currently standing, ‘Nothing is standing’ is tense-saturated
and true. Further, the tense operators are no longer inert, so one can truly
utter: ‘Nothing is standing, but P(something is standing).’
Suppose, on the other hand, that temporal parts are rejected. There are
then various possibilities.

1. Atemporal perspective: indexing—things talk, sit, etc., with respect to

times. Statements missing their temporal indices, such as ‘Ted is sitting’
are atemporally unsaturated. Quantification is over the entire block
universe: there are dinosaurs. Tensed perspective: like presentism. ‘Ted
is sitting’ is tense-saturated and true. There are no dinosaurs (and no
things that were dinosaurs), though there were dinosaurs.
2. Atemporal perspective: as with 1. Tensed perspective: one can ascribe a
 sui generis monadic property of presentness—and one cannot ascribe any
other properties.
3. Atemporal perspective: extreme version: one can say nothing. Less
extreme version: one can quantify over and name objects and ascribe
identity and distinctness to them. Still less extreme version: one can
additionally apply sortal predicates. Tensed perspective: like presentism.

View 1 has an odd feature: each (temporary) predicate has a monadic version
usable from the present perspective and a distinct indexed version usable
from the atemporal perspective. For example, from the present perspective
we say that Ted is sitting; from the atemporal perspective we say that Ted is
sitting at various times. How does sitting relate to sitting-at? Views 2 and 3
avoid this by “thinning” one of the two perspectives. (There are also
intermediate views, according to which one or both perspectives are more or
less thinned.)
The fifth and final variety is the doctrine of the growing block universe.
According to C. D. Broad (1923), the past and present are real, but the future
is not. Reality consists of the present and past portions of the spatializer’s
block universe, which grows by the addition of new “layers of being” to its
future edge. We can clarify the content of these metaphors by using our
framework.
Like the defender of the moving spotlight, the growing blocker believes in
a changing external perspective on the block universe—changing in the sense
of the tense operators. But unlike the moving spotlight view, it’s not just a
smidgen of the facts about the block that change; and the changing facts have
nothing to do with a primitive property of presentness. The changing facts
involve the constant addition to the block.
Like the spatializer, the growing block theorist thinks that the saturated
statements are those that are from an external perspective on the block. Thus
each of the following is saturated:

There is a second World War.

There are human colonies on Mars.
Ted sits at time t.

But unlike the spatializer, the growing blocker thinks that ‘There are human
colonies on Mars’ is false. It’s false because the block ends at the present,
and humans haven’t yet colonized Mars.
Also, the growing block theorist—given my construal, anyway—accepts
primitive tense operators.36 These are not inert when prefixed to saturated
sentences about the contents of the block universe. Using these tense
operators we can describe the growth of the block:

There is a second World War but P (There is no second World War).

There are no human colonies on Mars but F (There are human colonies on
Mars).

Thus the facts about the block universe change, according to the growing
block view. But there is a constraint on this change.37 Once “positive”
statements (examples include ‘Ted is sitting at t’ and ‘There exists a
dinosaur’, but I won’t try to make the notion precise) become true they
remain true forever after (e.g., A(∃xDx→G∃xDx)). For the only change
from the atemporal perspective is that of the addition of new layers of being.
Many defenders of the growing block view are motived primarily by the
thought that the future is “open” whereas the past is not. (They’re less
motivated by a desire for genuine change; thus the view perhaps doesn’t
belong in this section.) One sort of openness is secured simply by denying the
truth of statements that assert the existence of future individuals and facts,
such as ‘There are human colonies on Mars’ and ‘Ted lives in Ithaca in
2012.’ A stronger sort of openness would be secured by denying the truth or
(nonepistemic) determinacy of futuretensed statements such as ‘F (There are
human colonies on Mars)’ and ‘F(Ted lives in Ithaca in 2012).’38
To conclude. I myself am a spatializer. I reject all the A-theories of time
that I have considered, since I deny that tense operators are metaphysically
fundamental. The A-theorists’ fundamental languages are therefore deeply
defective, by my lights. Nevertheless, my disagreement with the A-theorists
is substantive, since it concerns the nature of the correct fundamental
ideology.

1 Arthur Prior’s (1967; 1968a; 1970; 1976; 1996) tense-logical approach to presentism justly
dominates the market.
2 See Quine (1960c, §36); Russell (1915); Sider (2001b, chapter 2); Smart (1963, chapter 7).
3 Prior and Fine (1977, p. 32).
4 The usual tense operators are insufficient; presentists also need a way to talk about temporal
distances. This is often accomplished by “metrical tense operators”, such as Px: “it was the case x
minutes in the past that”. But taking such an operator as metaphysically primitive would involve real
numbers in the ultimate metaphysics of time, and would apparently require a distinguished unit of
temporal measure. The presentist might instead adapt Mundy’s (1987) approach to quantities: i) replace
Px with infinitely many properties of propositions, which are properties of having been true at some
particular fixed distance in the past; ii) introduce primitive relations between these properties (e.g. the
relation of p1 being twice p2, meaning in effect that the temporal distance associated with p1 is twice
that of p2); and iii) lay down assumptions about these primitive relations from which one can prove a
representation theorem guaranteeing a function f (unique up to an arbitrary choice of unit) that assigns
one of these properties to each real number; and finally iv) define Px ϕ as meaning that the proposition
that ϕ has the property f(x). But this involves propositions in the ultimate metaphysics of time. A better
strategy, I think, is less platonistic: replace Px with primitive polyadic tense operators such as
C(ϕ,ψ,α,β) (C for congruence), meaning intuitively that some pair of cases of ϕ and ψ are separated in
time by the same amount, and in the same direction, as some pair of cases of α and β. (Unlike the usual
presentist operators, C is time-reversal symmetric.) Project: figure out how tightly the sentences
containing such operators constrain their Kripke models. (The issues in Hawthorne and Sider (2002) are
relevant.)
5 I write the predicate D without a parameter for a time because I think of an entity’s species as
being permanent. Spatializers who deny this would regiment the atemporal “There are dinosaurs” as
∃x∃tDxt—“Something is at some time a dinosaur”.
6 See Lewis (1986b, 202–4); Hinchliff(1996); Merricks (1994).
7 Just for fun: how might it go? Take Λ, V, ∃, and ∀ primitive; and suppose that there are two
points, r and u (real and unreal), relative to which objects have features. For instance, I am human
relative to r and a dragon relative to u—that is: philosopher(Ted,r) and dragon(Ted,u). Let’s recursively
define the expression “(ϕ) x” (read: “the translation of sentence ϕ with respect to point x”), where x may
be either r or u, as follows:

The theory, then, analyzes any sentence, ϕ, where ϕ may

contain ∼ in addition to Λ,V,∃, and ∀, as (ϕ) r. Notice that the theory requires both Λ and V, and both
∀ and ∃, to be primitive.
8 Is there a more general ontological claim that is the canonical locus of the dispute? One might
propose “Everything is present”, but a dispute over this is apt to be nonsubstantive since neither
presentists nor spatializers should accept a fundamental feature of “presentness” (compare Williamson
(1988)). In fact, it’s unclear whether the dispute over dinosaurs is even diagnostic, since it’s unclear
whether Williamsonian passage (see below) is a kind of presentism.
9 See, for example, Lombard (1999); Savitt (2004).
10 Jackson (2010) defends such a view.
11 Matters would be even worse if relationalism about time is true; compare Sider (2001b, chapter
4, section 8).
12 Attributed in Armstrong (1986).
13 See Jackson (2010).
14 I am not sure how my discussion relates to the view that Tim Maudlin (2007c) calls time’s
passage. Part of Maudlin’s view is that time has an intrinsic direction, and directionality is not the issue
here (directionality is accepted by many philosophers who self-identify as spatializers; it can be thought
of as the view that there is a fundamental asymmetric temporal relation). Moreover, he disavows the
rhetoric of “flow”. But he says that there’s more to his position than directionality, and he seems to tie
time’s passage to the traditional conception of genuine change.
15 See Smart (1963, chapter 7); Williams (1951).
16 See Markosian (1993).
17 See Zimmerman (2005).
18 A counterpart theorist (Lewis, 1968) might object to premise 4 by extending counterpart theory
to claims of the form “possibly, ψ”, where ψ contains a name of a possible world. Thus “Possibly, not:
there are no talking donkeys in @” would be true iff some counterpart of @—some other possible
world, presumably—contains talking donkeys. In reply to such a move, Williamson might shift his
focus from saturation to restriction: counterpart theory cannot provide for the intuitively correct claim
that it is contingent whether there unrestrictedly are talking donkeys (and not just whether @contains
talking donkeys). Thanks to Williamson for discussion. See Dorr (2010) for a discussion of a number of
issues in this vicinity.
19 Tim Williamson pointed out that similar questions confront other attempts to formulate logical
principles and schemas—such as the indiscernibility of identicals—in natural language.
20 I have in mind only beliefs about objective matters, not beliefs about, e.g., left and right.
21 This is still vague; “specifying a value” needs to include doing so by quantification, as in
“∃x∃y∃z x is between y and z”.
22 Not that this is incoherent; see Fine’s (2000) discussion of “positionalism” (a view he ultimately
rejects).
23 Must the marks of punctuation, or indeed, the typeface and color also carve at the joints? No,
only representational facets must carve. Saying this does not require any sort of metaphysically
fundamental distinction between representational and nonrepresentational facets of a language, since
the notion of a fundamental language itself—a partly linguistic notion—need not be fundamental.
Recall section 10.2.
24 The distinction between saturated and unsaturated sentences could be extended to cases where
sentences are not so syntactically similar to their metaphysical truth-conditions. We would need to
identify some natural correspondence between parts of sentences and parts of their metaphysical truth-
conditions, under which S is saturated if its parts correspond to all the parts of µ(S), whereas it is
unsaturated if µ(S) has some extra parts under the correspondence.
25 Throughout I am taking the natural language sentence ‘Ted is sitting’ at face value, as containing
no syntactic constituent that represents a time. But perhaps this is false at some deep level of syntactic
analysis. See Stanley (2000) for a discussion of the general issues here.
26 See Williamson (1998; 2002). Although these works are primarily concerned with modality, p.
245 of the latter article contains remarks that are consistent with Williamsonian passage.
27 And perhaps also mathematical/logical properties like being self-identical and being a member of
its unit set.
28 And perhaps modal and logical and mathematical properties.
29 The Williamsonian could accept claims of the form “Something is a dinosaur at time t” if
analyzed in terms of tensed claims; but then they could not be used in a reduction of tense.
30 If the spotlight theorist believed in temporal parts, she could say instead: sentences like
‘Temporal part t is sitting’, ‘Temporal part t is part of Ted’, ‘Event e occurred in 1943’, and ‘Event e is
part of World War 2’ are saturated. Sentences like ‘Ted is sitting’ are saturated but false since ‘Ted’
denotes a space-time worm that is not sitting simpliciter (though see Wasserman (2005, chapter 3)).
31 See Zimmerman (1998) on related matters. Forrest (2004) defends a related view (although
Forrest (2006) strikes me as backtracking).
32 See Bourne (2002); Braddon-Mitchell (2004); Merricks (2006); and compare Lewis (1970a,
section 9) on actuality.
33 There are two ways a Williamsonian could construe the thought “I am in the present”; in neither
way could anyone mistakenly think this thought. He could construe it as meaning “This very thought, t,
is now occurring”; thus t is true iff N(t is occurring). But then no one thinking t could be mistaken: if
they’re thinking t then t is occurring; so t is now occurring (the Williamsonian accepts ϕ↔Nϕ, recall);
so t is true. Alternatively, he could construe it as meaning “I am concrete.” But only concrete things
think; so again, no one thinking the thought could be mistaken.
34 Views that embrace the atemporal perspective are in the best position to solve the problem of
relations (Sider, 2001b, chapter 2, section 2). The Williamsonian, for example, has as much trouble
with cross-time relations as does the presentist.
35 Compare Fine’s (2005) idea of reality relative to a time. Fine speaks of what is real at t for
arbitrary times t. Perhaps in my terms the idea is that there is a different joint-carving category of
sentence for each time.
36 There’s an interesting question of whether the defender of the block universe who thinks that the
future is “open” could do away with the primitive tense operators.
37 Compare van Inwagen (2010, pp. 11–12).
38 Such denials are often restricted to statements that are not settled by the past plus laws of nature.
But if it were me, I wouldn’t restrict the denial—provided we’re talking about what is fundamental.
The intuitive idea here is that there simply aren’t any facts about what is going to happen, and the fact
that the laws and past sometimes yield determinate predictions doesn’t put any pressure on this intuitive
idea. These are only predictions, after all; why regard them as constituting facts about what will
happen?
The denials are also open to presentists and Williamsonians. A related but distinct option for the
Williamsonian is to deny the future Barcan formula and its converse. And, in the spirit of multiplying
views beyond necessity, note this Williamsonian variant on the growing block view: there exist
nonconcrete, spatiotemporally unlocated entities with future-tense properties; these individuals will
eventually join the block universe by gaining concreteness and a spatiotemporal location, which they
will then retain forever after.
12 Modality
At bottom, the world is an amodal place. Necessity and possibility do not
carve at the joints; ultimate reality is not “full of threats and promises”.1 The
book of the world says how things are, not how they must or might be.
This is not to say that there is no modality. The book of the world does not
mention cities, smiles, or candy either; yet there are cities, smiles, and candy.
Previous chapters have discussed logical and nomic necessity—logical
truth and laws of nature—and have argued that neither is fundamental. The
kind of necessity here in question is so-called “metaphysical necessity”, a
kind supposedly intermediate in strictness between logical and nomic
necessity. This chapter argues that metaphysical necessity is not fundamental,
and shows how to fit it into a fundamentally amodal world.

12.1 No fundamental modality

Why hold that metaphysical necessity is nonfundamental? Let us first
disavow a bad reason. You may have noticed that the philosopher’s
conception of metaphysical necessity is frustratingly thin. We are told that
logic and mathematics is metaphysically necessary. We are usually told that
laws of nature are not. We are told that it is metaphysically necessary that
“nothing can be in two places at once”, and so on. This conception falls far
short of a full criterion. But a thin conception is not in itself problematic. For
when a notion is taken to be fundamental, one often assumes that the facts
involving the notion will outrun one’s conception. (Why take to be
fundamental what you can define?) We cannot define the primitive predicates
of physics, but no one supposes our conceptions of those predicates to settle
all cases. We rather introduce such predicates, more or less, by a theoretical
analog of pointing. “What do I mean by ‘charge’? Here are some examples;
here is a partial theory of charge; charge is what is common to (enough of)
the examples and makes (enough of) the theory come out true.” Defenders of
metaphysical necessity will say similar things. “What do I mean by
‘metaphysical necessity’? Here are some examples (logic and mathematics
are necessary, laws of nature aren’t necessary, and so on); here is a partial
theory of necessity (this might include principles of modal logic).
Metaphysical necessity is whatever is common to (enough of) the examples
and makes (enough of) the theory come out true.” Now, such speeches are
often phrased as introductions to a concept or meaning. Thus Alvin
Plantinga:
But what exactly do these words—‘necessary’ and ‘contingent’—mean? What distinction do they
mark? Just what is supposed to be the difference between necessary and contingent truths? We can
hardly explain that p is necessary if and only if its denial is impossible; this is true but insufficiently
enlightening. It would be a peculiar philosopher who had the relevant concept of impossibility well in
hand but lacked that of necessity. Instead, we must give examples and hope for the best. (1974, p. 1, my
emphasis)
Taken as introductions to a concept, such speeches will invariably disappoint.
I don’t know how to begin with the necessity of logic, mathematics, not being
in two places at once, and answer new questions about metaphysical
necessity (such as the question of whether the dimensionality of spacetime is
metaphysically necessary or contingent). But the speeches are, or should be,
nothing of the kind. What they do is supply enough of a semantic constraint
so that, if reality contains the requisite structure, the term ‘metaphysically
necessary’ will be semantically determinate.
The good reason for opposing modal primitivism is simply: ideological
economy.2 Modal talk is certainly common in ordinary and special-science
discourse. But we do not generally take notions from these high-level
domains as good candidates for being metaphysically basic, as “proper
annex[es] to austere scientific language” (Quine, 1976b, p. 863), since they
are unneeded for the most fundamental inquiries of mathematics and physics.
We tend to think of psychological, economic, political, and other special-
science notions as being nonfundamental. Many of us think the same about
various philosophical notions: semantic, moral, epistemic, causal. Since
modality is unneeded for the most fundamental inquiries, it too is
metaphysically nonfundamental, however conceptually fundamental it may
be.
(There is similar reason to oppose primitivism about essence, construed
non-modally Kit Fine convincingly argues against the standard modal
definition of essence: a is essentially F iff it is necessary that a is F (provided
a exists). Even though it is necessary that Socrates is a member of his unit set
if he exists, Fine points out, we do not think it part of the essence of Socrates
—part of “what it is to be Socrates”— that he be a member of his unit set
(though we do think it part of the essence of his unit set that it contain
Socrates). Essence, for Fine, is not to be understood modally at all.3 This is
an attractive view; but we should not regard nonmodal essence as being
metaphysically basic (not that Fine does): fundamental theories need essence
no more than they need modality. Perhaps a reductive account of essence
along the lines of the reduction of modality to be given below could be
developed.)

12.2 A Humean strategy for reduction

If modality is not fundamental, then it must be nonfundamental. (The
alternative would be to say that there is no modality at all, not even in a
nonfundamental sense, which is about as reasonable as saying the same about
cities, smiles, and candy.) We need a reduction of modality—in my terms, a
metaphysical semantics for modal language.
A familiar strategy for reduction is to define necessity as truth in all
possible worlds. But this constitutes a reduction of modality only if possible
worlds both exist and can be nonmodally defined. David Lewis’s (1986b)
concrete possible worlds are nonmodally defined,4 but it is very difficult to
believe that they exist. The usual nonLewisian alternative is to construct
possible worlds out of preexisting “abstract” materials: sentences,
propositions, and so on.5 But the usual constructions are defined modally
(viz., “a possible world is a maximal consistent set of propositions”).6
A very different strategy locates modality in linguistic convention.
According to the old “linguistic” or “conventionalist” theory of necessity,
analyticity in the sense of truth by convention was held to explain necessity.
If analytic truths “do not make any assertion about the empirical world, but
simply record our determination to use symbols in a certain fashion”, as Ayer
put it (1936, p. 31), then, it was thought, we can understand how they are
guaranteed to be true no matter how the world turns out to be. In its crudest
form, conventionalism claimed that a statement of necessity, such as ‘It is
necessary that all bachelors are male’, says that the component sentence (‘All
bachelors are male’) is true by convention.
Today this theory has few adherents, for the most part with good reason. If
 ϕ says that ϕ is true by convention, then ϕ would apparently turn out
contingent, since statements about what conventions we adopt are not
themselves true by convention. Thus the characteristic axioms of modal
systems S4 ( ϕ→ ϕ) and S5 (∼ ϕ→ ∼ ϕ) would be in all cases false.
Further, conventionalism is apparently inapplicable to Kripke and Putnam’s
examples of the necessary a posteriori (and, relatedly, to de re modality). But
most importantly, as we saw in section 6.5, the notion of truth by convention
is thoroughly confused. Linguistic conventions can put semantic relations in
place (since facts about semantic relations are partially constituted by
linguistic conventions). Thus by adopting the convention that ‘bachelor’
means the same as ‘unmarried man’, we can bring it about that ‘An object is
a bachelor iff it is an unmarried man’ means the same as ‘An object is an
unmarried man iff it is an unmarried man’. But there convention’s power runs
out; it plays no role in explaining why the latter sentence is true, or necessary.
Nor does convention play any role in explaining the truth or necessity of
mathematics, or of any other domain. As Quine says, “… definitions are
available only for transforming truths, not for founding them” (Quine, 1936,
p. 81).
I prefer a third strategy for reducing modality, which I’ll call Humean, for
lack of a better word.7 To say that a proposition is necessary, according to the
Humean, is to say that the proposition is i) true; and ii) of a certain sort. A
crude Humean view, for example, would say that a proposition is necessary
iff it is either a logical or mathematical truth. What determines the “certain
sort” of propositions? Nothing “metaphysically deep”. For the Humean,
necessity does not carve at the joints. There are many candidate meanings for
‘necessary’, corresponding to different “certain sorts” our linguistic
community might choose. Since none of these candidates carves at the joints,
our linguistic community is free to choose whichever of these it likes.
Perhaps the choice is arbitrary, in which case the facts about necessity are
“conventional” in the sense of section 4.3. Perhaps the choice reflects
something important about the role ‘necessary’ plays in our conceptual lives,
in which case the facts are “subjective” (or “projective”). More likely, the
truth is somewhere in between. But at any rate, the conceptual choice is not
forced on us by the facts.
Contrast Humeanism with conventionalism. Conventionalists claim that
statements that are necessary (such as logical and mathematical truths) are
made true by linguistic convention, and they deny that such statements are
fundamentally true (if they speak of fundamentality at all). Further, at least
the cruder ones claim that a statement of necessity, ϕ, says that ϕ is true by
convention. The Humean denies each of these claims. Neither logical nor
mathematical truths owe their truth to convention (except in the banal sense
in which every true sentence partly owes its truth to the conventions that
secure its meaning). If mathematical and logical notions carve at the joints,
then mathematical and logical truths are fundamental truths.8 Statements of
necessity say nothing about conventions. What ‘It is necessary that 2 + 2 = 4’
says is that 2 + 2 = 4 is either a logical or mathematical truth (given the crude
Humean view, anyway).
For the Humean, convention plays no role in truthmaking. Insofar as the
Humean wants to speak of truthmaking at all, she will say the following. The
proposition that 2+2 = 4 is made true by whatever makes mathematical truths
true generally (facts about mathematical entities, perhaps). Its status as a
mathematical truth is secured by whatever generally makes mathematical
truths mathematical (perhaps the fact that its subject matter is solely
mathematical). And the further proposition that the proposition that 2 + 2 = 4
is either a logical truth or a mathematical truth—that is, the proposition that it
is necessary that 2 + 2 = 4—is made true by all of the above, plus a general
(logical) fact about disjunction: a disjunction is true whenever one of its
disjuncts is.
Nevertheless, convention, or something like it, does play a role for the
Humean. The Humean shares the conventionalist’s goal of accounting for
modality in a fundamentally amodal world. To that end, it’s important that
the “certain sorts” of propositions invoked by the Humean are not objectively
distinguished, that no joint in reality encircles the class of logical and
mathematical truths. So what does select this class? Something about us, says
the Humean. Perhaps the choice of the “certain sorts” is conventional.
Convention can do this without purporting to make true the statements of
logic or mathematics (or, for that matter, statements to the effect that these
truths are necessary), for the choice of the certain sorts is just a choice about
what to mean by ‘necessary’. Or perhaps the choice is partly
subjective/projective rather than purely conventional. Either way, necessity is
accounted for without metaphysically privileging the class of necessary
truths.
Truth by convention was an essential part of conventionalist
epistemology: it was to explain how we know logical and mathematical
truths. The Humean account of necessity, in contrast, is metaphysically,
rather than epistemologically driven. It tries to fit modality into a demanding
metaphysics, not into a demanding epistemology. Thus the Humean is free to
simply ditch the idea of truth by convention. The price is that no headway is
made on epistemology: nothing in the Humean account of necessity sheds
any light on the epistemology of logic or mathematics.
Truth by convention was also perceived as an essential part of
conventionalist metaphysics, insofar as there was such a thing. As Paul
Boghossian (1997, p. 336) puts it:
Guided by the fear that objective, language-independent, necessary connections would be
metaphysically odd, [conventionalists] attempted to show that all necessities could be understood to
consist in linguistic necessities … Linguistic meaning, by itself, was supposed to generate necessary
truth; a fortiori, linguistic meaning, by itself, was supposed to generate truth.

But the Humean account shows that we do not need truth by convention to
avoid the “metaphysically odd” in our account of modality.9
Distinguish between governance and classification conceptions of
necessity. According to the governance conception, necessity is a source of
truth. When ϕ is true, ϕ is true because ϕ is true.10 (Compare the
governance conception of laws of nature, on which laws “guide” the
evolution of the world.11) On the classification conception, on the other hand,
the truth of ϕ comes first. Necessity plays no role in truthmaking. To say that
a true proposition is necessary is to classify that proposition as being of a
certain sort, but the proposition is true on its own merits.12
The Humean theory is on the classification side of this divide. (As is
combinatorialism, and Lewis’s modal realism. What is most fundamental for
Lewis is truth from the perspective of the pluriverse; necessities are just
certain kinds of truths about the pluriverse.13 For the combinatorialist, a
necessary truth is just a truth that remains true in all combinations.) Everyday
modal thought may fit the governance conception better than the
classification conception (much as everyday thinking about laws of nature
arguably fits the governance conception). But according to the Humean,
nothing answers to the governance conception; modality isn’t what we
ordinarily take it to be.
The core idea of the Humean account, then, is that necessary truths are
truths of certain more or less arbitrarily selected kinds. More carefully: begin
with a set of modal axioms and a set of modal rules. Modal axioms are
simply certain chosen true sentences; modal rules are certain chosen truth-
preserving relations between sets of sentences and sentences. To any chosen
modal axioms and rules there corresponds a set of modal theorems: the
closure of the set of modal axioms under the rules.14 Any choice of modal
axioms and modal rules, and thus of modal theorems, results in a version of
Humeanism: to be necessary is to be a modal theorem thus understood.15
(“Modal” axioms, rules, and theorems are so-called because of their role in
the Humean theory of modality, but the goal is to characterize them
nonmodally; otherwise the theory would fail to be reductive. Also, the set of
modal “axioms” need not be recursive.)
A simple version of Humeanism to begin with: the sole modal rule is first-
order logical consequence, and the modal axioms are the mathematical truths.
(Logical truths are logical consequences of any propositions whatsoever, and
so do not need to be included as modal axioms.) The following sections
develop the Humean account by considering a series of worries, and
responding with a combination of refinement and argument. The resulting
account will be partial16 in various ways; but it will be enough, I think, to
justify the claim that Humean metaphysical truth-conditions for modal
statements can in principle be given.

12.3 Logical consequence and mathematical truth

“How can the notions of logical consequence and mathematical truth be
characterized?”
Humeanism is incompatible with some theories of logical consequence.
Any modal account would obviously render the theory circular. The simplest
modal account is that logical consequence is just necessary consequence;
another modal account says that logical consequences are modal
consequences that involve only logical words essentially, in Quine’s sense
(see below).17 The Humean must reject each.
Humeanism could consistently be combined with primitivism about
logical consequence. But the reasons for rejecting primitive modality lead
also to rejecting primitive logical consequence: there seems to be no need for
the metalogical notion of logical consequence in our most fundamental
inquiries. Humeanism plus primitivism about logical consequence would, I
think, be an advance over primitivism about metaphysical necessity. But the
true lover of desert landscapes will want to reject each form of primitivism.
 Quine once defined logical truths as those truths which involve only
logical words essentially:18
First we suppose indicated, by enumeration if not otherwise, what words are to be called logical words;
typical ones are ‘or’, ‘not’, ‘if’, ‘then’, ‘and’, ‘all’, ‘every’, ‘only’, ‘some’. The logical truths, then, are
those true sentences which involve only logical words essentially. What this means is that any other
words, though they may also occur in a logical truth (as witness ‘Brutus’, ‘kill’, and ‘Caesar’ in ‘Brutus
killed or did not kill Caesar’) can be varied at will without engendering falsity.

Extended to a theory of logical consequence, this account could be combined

noncircularly with Humeanism. But it would generate implausible results. If
‘=’ is counted as a logical word, certain true sentences specifying information
about the number of things, for example ‘∼∃x∀y x=y’ (“it is not the case
that there exists exactly one thing”) count as logical truths. Since the Humean
account implies that all logical truths are necessary, this would rule out the
possibility of there existing only one thing. And even if ‘=’ is not counted as
a logical word, if there are in fact a finite number, n, of entities, then any
sentence of the following form turns out to be necessarily true:19

Any sentence of this form is true whenever there are n or fewer entities (the
only way to make the antecedent true is to assign n distinct entities to x1…xn,
each of which satisfies F; but if there are no more than n entities, then y must
be assigned one of these things as well, in which case the consequent comes
out true.) So no matter what substitutions we make for the predicates, the
sentence remains true; thus it counts as logically true given Quine’s criterion,
and so counts as necessarily true given Humeanism. But some sentences of
this form seem only contingently true: they could have been falsified if there
had existed n things satisfying the antecedent, and some additional entity that
failed to satisfy the consequent. It might be responded that these
consequences are unproblematic: since abstract entities are infinite in number
and exist necessarily, it is impossible for there to exist just one thing, or just
finitely many things.20 But a Humean might deny the existence of abstracta.
Moreover, the contextualist (see below) might want to allow contexts in
which ‘abstract entities exist necessarily’ is false. Whatever its other merits,
Quine’s theory of logical truth does not mesh with the Humean account of
modality.
 The “best-system” account of logical truth mentioned in section 10.3, on
the other hand, if extended to a theory of logical consequence, would mesh
just fine with Humeanism. So would a model-theoretic conception of logical
consequence. The model-theoretic conception (and Quine’s as well) faces the
question of how to characterize the notion of a logical constant. Only logical
constants receive constant interpretations in all models, so the
characterization affects the resultant notion of logical consequence.21 But an
element of conventionality or subjectivity or projection or semantic
indeterminacy in the notion of a logical constant, resulting in such an element
in the notion of necessity, will be acceptable to the Humean.
The Humean theory is mostly neutral on the characterization of
mathematical truth, provided the characterization is non-modal and plausibly
yields necessary truths. One vague conception: a mathematical truth is a
proposition that concerns just mathematics and is true. There is a question
parallel to that of how to characterize the logical constants: what is to count
as a part of mathematics? But again, an element of conventionality,
subjectivity, or the like will be tolerable to the Humean. If there are sharp
lines to be drawn around logic and mathematics, then necessity here is sharp;
if not, not. The spirit of Humeanism is that necessity is not a realm to be
discovered. We draw the lines around what is necessary. It should be no
surprise that we sometimes do this incompletely, arbitrarily, or projectively
Note that the Humean need not assume that mathematical statements are true.
If they are not, then there are no mathematical truths, and hence no
mathematical necessities, which would seem to be the right consequence
given this philosophy of mathematics.

12.4 Analyticity
“Analytic truths should turn out necessary.”
Granted. We should include each analytic truth—in the sense of section
9.8— as a modal axiom. Analytic sentences were there characterized, recall,
as sentences that are both true and definitional (where a definitional sentence
is a sentence intended to constrain the meanings of its terms). This does not
require truth by convention, since part of the characterization of an analytic
sentence is that that sentence be (“already”) true.22
12.5 Laws of metaphysics
“Statements of fundamental metaphysics should turn out noncontingent.”
It seems to be part of the concept of metaphysical necessity that certain
statements of fundamental metaphysics are noncontingent. Consider for
example:23
Parthood is transitive.
There are no (merely) past or future objects.
Objects have temporal parts.
Any objects have a mereological sum.
In particular, consider these sentences as they occur in discussions of
fundamental metaphysics, where the terms in question are intended to carve
at the joints, and where the truth of the sentences is a matter of controversy.
In such contexts, the sentences seem to be noncontingent: necessarily true if
true; necessarily false if false.24 But in such contexts the sentences do not
seem analytic. They are the subject of controversy, and so are not construed
as definitional of the terms they contain. It might be argued that certain
ordinary sentences containing the disputed vocabulary (‘part’, quantification,
temporal vocabulary) are definitional; but philosophers who think that this
vocabulary carves at the joints and engage in dispute over the nature of these
joints suspend any such definitional constraints. If the sentences (in this
context) are not analytic, how shall the Humean account for their
noncontingency?
By adding a new group of modal axioms: the true propositions about such
fundamental and abstract matters. What justifies their status as modal
axioms? This is just how the concept of necessity works. Such propositions
have no further feature that explains their inclusion as modal axioms.
Exactly what sorts of truths about “fundamental and abstract matters” get
included as modal axioms? I’m not sure how much sharpening of this vague
formulation is possible (or necessary), but we might wheel in the generalized
Lewisian account of lawhood from section 3.1. If complexity is expensive,
only the laws of logic count as laws (recall section 10.3). If it is cheap, then
physical and special-science generalizations also count as laws. If its cost is
middling, the special-science laws drop away but the laws of physics remain.
Now, consider a cost of complexity intermediate between the first
(“expensive”) and the last (“middling”). Here we drop the laws of physics but
still retain more than the laws of logic: we also retain what one might call
laws of metaphysics. (Despite the grand name, no heavy-duty metaphysics is
intended; the underlying account of lawhood is reductive.) The sacrifice in
simplicity needed for their inclusion is greater than that required for the laws
of logic, since the relevant terms (such as mereological and temporal
vocabulary) are not quite so topic-neutral and pervasive as those of logic. But
the laws of metaphysics are simpler because more abstract, pervasive, and
topic-neutral than the laws of fundamental science, and because less
“specific”.
The Humean view about necessity requires no commitment to any
particular laws of metaphysics. It says merely that the laws of metaphysics,
whatever they happen to be, are to be included as modal axioms. The
Humean could, for example, be neutral about whether objects have temporal
parts; she would say simply that whichever of ‘Things have temporal parts’
and ‘Things do not have temporal parts’ is true is a modal axiom. The roster
of modal axioms is a function of whatever the nonmodal facts concerning
fundamental and abstract matters turn out to be.
The laws of metaphysics provide examples of synthetic necessary truths.
Supposing it to be true that past and future objects do not exist, it is
impossible that they exist, despite the fact that there is no conceptual
incoherence in supposing that they do. Some metaphysicians find synthetic
necessities deeply disturbing, and go to great lengths to avoid them. Ross
Cameron (2007), for example, argues that i) mereological sums exist; but ii)
existential claims are never analytic; and so, desiring to avoid synthetic
necessities, concludes that iii) it is contingent that mereological sums exist.
Thus, for example, although the subatomic particles in my body compose
something—namely, me—there is a possible world exactly like ours at the
subatomic level in which those particles compose nothing. A second
example: Cian Dorr’s (2004) hostility to synthetic necessities plays an
important role in his argument against non-symmetric relations. And
Cameron and Dorr are just two recent examples of a centuries-old tradition
that finds synthetic necessities metaphysically mysterious. As Dorr (2004,
section 3) puts it, “Metaphysical necessity is never ‘brute’: when a logically
contingent sentence is metaphysically necessary there is always some
explanation for this fact.”
I wonder whether the aversion to synthetic necessities survives the full
repudiation of truth by convention. Dorr requires necessities to be
“explained”. He explains the necessity of the logically contingent “All water
is H2O”, for example, by the fact that metaphysical analysis reveals it to
express the same proposition as the logical truth “All H2O is H2O.” But why
doesn’t the necessity of this logical truth then need to be explained?
Assuming that truth by convention is off the table, the question must be
faced.
A certain sort of picture-thinking is initially compelling. It couldn’t be
necessary that mereological sums exist, for what would force the existence of
a further whole, in addition to the existence of a given set of parts? The fact
of the parts’ existence is one fact, the existence of the whole another fact;
how could the first necessitate the second? But the picture does not withstand
scrutiny. Why not reason also as follows: the fact that it is raining is one fact;
the fact that it is either raining or snowing is a distinct fact; how could the
first necessitate the second? One is tempted to respond that the cases are
different: the fact that it’s either raining or snowing isn’t a separate fact from
the fact that it’s raining. But the temptation dissipates once truth by
convention has been repudiated. Once we recognize that logical content is as
“wordly” as nonlogical content, we see that there is no interesting construal
of fact-identity on which the first pattern of reasoning is sustained but the
second is not. The picture-thinking loses its grip. For the picture was that we
perceive in allegedly unproblematic cases of logical necessitation a certain
kind of connection, a connection which explains the neces-sitation; and we
see that this connection is missing in the case of mereological sums. But now
we see that the perceived connection was never there in the first place.
From a Humean perspective, synthetic necessities are no more puzzling
than analytic or logical necessities. As it happens, we use the word
‘necessary’ to encompass certain synthetic truths in addition to analytic and
logical truths. But there’s no deep fact here: we could have used ‘necessary’
for just the analytic and logical truths—or for a more inclusive set of truths.
The fact that certain synthetic truths are necessary is no more surprising than
the fact (if it is a fact) that cups are glasses.
The aversion to synthetic necessities makes a bit more sense from a non-
Humean perspective, but even then its status is hardly that of metaphysical
bedrock. Suppose, for example, that necessity is a primitive feature of the
world, an irreducible “glow” in some true propositions. What then would be
wrong with saying that certain synthetic propositions glow, in addition to
logical and analytic propositions? There can be no “constitutive” explanation
of why the logical truths glow, for example, in the sense of an explanation
which says: to glow is to be such-and-such; logical truths are examples of
such-and-suches; so logical truths glow. For the glow is primitive. And if the
modal primitivist lacks a constitutive explanation of why the logical truths
glow, what’s so bad about lacking a constitutive explanation of why certain
further propositions glow? The modal primitivist might respond that a leaner,
more attractive view about primitive modality would restrict its application as
much as possible. The more restriction of the glow the better; best-case
scenario: restriction solely to the logical truths. Fair enough, but the resulting
doctrine is then one speculative metaphysical hypothesis among many, not a
bedrock truth whose denial would be “unintelligible” (as Lewis (1983b, p.
366) describes a related denial).
The Humean treatment of the necessity of laws of metaphysics
undermines “arguments from possibility” for conclusions in fundamental
metaphysics. Such arguments begin by claiming that a certain proposition is
possible—perhaps on the grounds that “modal intuition” informs us that it is
possible, perhaps because of some general presumption in favor of
possibility. Next, the possibility is argued to be incompatible with a certain
fundamental proposition of metaphysics. So the fundamental proposition is
possibly false. Finally, it is concluded that the fundamental proposition is
actually false, since fundamental propositions of metaphysics are
noncontingent. For example, I once argued that mereological nihilism (the
view that only mereologically simple entities exist) is actually false on the
grounds that gunk is possible (Sider, 1993b). And even arguments about
weightier matters, such as David Chalmers’s (1996) argument against
materialism, fall in this category.
Such arguments are undermined by Humeanism. This is not because
Humeanism undermines the premise that fundamental propositions of
metaphysics are noncontingent. On the contrary, Humeanism vindicates this
premise, since fundamental propositions of metaphysics count as modal
axioms when true.25 It is rather because Humeanism undermines our reason
for accepting the possibility in question. Intuitively, this is because for
propositions of fundamental metaphysics, possibility boils down to the actual
falsity of rivals.
A proposition p of fundamental metaphysics typically has rivals: other
propositions of fundamental metaphysics that are incompatible with p. (I
have in mind competing accounts of the same subject matter. For example,
rivals to materialism include dualism and idealism.) Now, the set of modal
axioms is defined as containing, among other things, the propositions of
fundamental metaphysics that are true, whatever those happen to be. And to
say that a proposition is possible, on the Humean view, is to say that that
proposition’s negation is not a logical consequence of the modal axioms.
Thus, to say that the fundamental metaphysical proposition p is possible is to
say that its negation is not a logical consequence of a set that is defined as
containing its true rivals. Given this, there is next to no epistemic difference
between asserting that p is possible and asserting that its rivals are false.26
Assuming the Humean theory, then, neither modal intuition nor a putative
presumption in favor of possibility can be regarded as probative in matters of
fundamental metaphysics—unless intuition or the presumption are somehow
probative concerning the actual falsity of rivals; but then there would be no
need to bring in possibility; one could argue directly against the rivals.
Arguments from possibility in effect assume that modality is separate, in a
certain sense, from actuality. They assume that the status of possibility is one
we can assess independently of questions about actuality, and so, in the case
of noncontingent propositions, make genuine epistemic advances on
questions of actuality by reasoning through this separate status. But for the
Humean, modal notions are not separate from actuality in this way. To be
necessary just is to be true and of a certain sort.
I have been disparaging arguments from possibility; but it may be
objected that surely some such arguments are good. Imagine a physical
theory that predicts bizarre results if there are exactly seventeen particles, but
makes sensible predictions otherwise. Can we not object that the theory
makes the wrong predictions with respect to those physically possible
scenarios involving exactly seventeen particles? We could; but we could also
object without bringing in modality. Since the theory makes an exception for
the case of seventeen particles, it is surely needlessly complex, and is
therefore less explanatory than an otherwise similar theory without the
exception. I suspect that something similar is true generally: when there is
good in an argument from possibility, it can be recast in other terms—
explanation, for instance.

12.6 Determinates and determinables

“Why are distinct determinates of a given determinable incompatible? Why is
it impossible for something to be both 5g mass and 6g mass, for example?”
It’s tempting to respond that it’s analytic that nothing has two distinct
masses. ‘Nothing has two masses’ is both true and a definitional constraint on
the interpretation of the term ‘mass’. But suppose we come across a new
property which we name ‘schmass’.27 Since schmass has only just been
discovered, it is not definitional of ‘schmass’ that different values of schmass
are incompatible. Indeed, let us stipulate that we are not yet sure whether
schmass comes in degrees. But in fact, suppose, schmass does come in
degrees—indeed, the determinate schmass properties have the structure of the
positive real numbers, just like the masses. Further, nothing has two of these
determinate properties. In such a case, it’s surely necessary, but not analytic,
that nothing has two distinct schmasses.
The explanation of necessity here lies not in analyticity, but rather in the
fact that it’s a law of metaphysics that nothing has two schmasses. (Likewise,
it’s a law of metaphysics that nothing has two masses.) In general, facts about
determinate/determinable structure are “fundamental and abstract” features of
reality of the sort discussed in the previous section.
For the sake of concreteness, let’s go into the metaphysics of quantities a
bit. Consider first Brent Mundy’s (1987) view of quantities.28 According to
Mundy, numerical representation of quantities is justified by the first-order
and higher-order structure of the quantities in question. The infinitely many
determinate mass properties, for example, are structured by higher-order
relations, for instance a binary relation, R, signifying “is a larger mass”, or a
ternary relation S of summation (m1m2Sm3 iff masses m1 and m2 “sum to”
mass m3). Assuming that certain constraints hold, one can prove
representation theorems to the effect that there exists a unique (up to an
appropriate transformation) function from the first-order mass properties into
the real numbers that “meshes” with the higher-order relations. Mesh
requires, for example, that the number assigned to one mass property be
greater than the number assigned to another iff the first bears the relation R to
the second, and that whenever three masses stand in S, the sum of the
numbers assigned to the first two masses must equal the number assigned to
the third. The constraints assumed by the representation theorem include
constraints on the higher-order structuring relations (for example, R must be
transitive) as well as constraints on the distribution of the first-order
determinate mass properties (for example, no particular can have two of
them). Notice that similar constraints are obeyed by other quantities as well.
Some quantities are structured somewhat differently; for example,
temperature has a zero point. Still, there is a pattern of constraints that are
mostly obeyed by multiple quantities. The pattern is widespread and
fundamental enough to count as a law of metaphysics. The first-order portion
of the pattern—nothing instantiates two of the determinates—is what
accounts for the incompatibility between distinct determinates.
A rival account of quantities is the first-order account of Krantz et al.
(1971), according to which the numerical representation of quantities is
justified by the holding of first-order relations over particulars. In the case of
mass, in place of Mundy’s relations R and S over the determinate mass
properties we have corresponding structuring relations more-massive-than
and summation over massive objects. As with Mundy, representation
theorems can be proved on the assumption that the structuring relations hold
in certain patterns; and as with Mundy, similar (though not identical) patterns
recur with other determinables. We can again include these patterns as modal
axioms under the rubric of laws of metaphysics.
Some confirmation that this account is on the right track can be gained by
reflecting on some examples. First example: suppose some quantity is
discrete, and has a small number of values—say, eight. The structuring
relations (whether higher-order or first order) hold in a rather different pattern
from the pattern for other, continuous, quantities. Does the pattern for the
discrete quantity count as a law of metaphysics? There seems to be no fact of
the matter. The concept of a law of metaphysics does not decide the case one
way or the other, and that concept is certainly not a fundamental one. The
present account thus predicts that there is no fact of the matter whether it’s
possible for a thing to have two “determinates” of the discrete
“determinable”. This prediction seems intuitively correct.
Second example: imagine discovering that one particular determinable
property—mass, say—lacks most of the structure it is generally presumed to
have. In Mundy’s terms, suppose there are no second-order relations over the
determinate masses (no joint-carving second-order relations, I would say); in
the first-orderist’s terms, suppose there are no (joint-carving) relations of the
sort described above over massive objects. What we call mass is in fact
heterogeneous. (Perhaps the class of masses has various structured
subgroups, perhaps not.) Now, even if no object in fact has two masses in this
example, having two masses seems less impossible, intuitively, than having
two determinates of one of the other determinables where the structure is not
missing. And this is what the present account predicts: ‘No object has two
masses’ less clearly fits into the fundamental and abstract pattern exhibited
by the other determinables, and hence is less clearly a law of metaphysics.

12.7 Contextualism
It’s a familiar point that natural language modal words are contextually
variable. “Johnny can’t watch TV next door”, I might say, indicating that
going next door is forbidden to Johnny. But I might later say, with no change
in the facts being described: “Johnny can watch TV next door (since our
neighbors leave their doors unlocked), but he can’t watch across the street
(since the doors there are locked).” Still later I might say: “Johnny can go
across the street (since that’s close by), but he cannot go across town (since
he doesn’t yet know how to drive).” More extremely, I might say: “Johnny
can go to the moon (since the technology exists), but not to Mars”, whereas
later I might say “Johnny can travel to Mars but not to star systems 10,000
light years away (since supraluminal travel violates the laws of nature and
humans don’t live to be 10,000 years old).”
This contextual variation might be regarded as happening only with
restricted modalities. There is some one “outer” modality, according to this
view, of which each contextual modality is a restriction.29 The outer modality
might be claimed to be “metaphysical necessity”, for example. On this view,
a restricted necessity “necessarily, ϕ” is analyzed as (R→ϕ), and a restricted
possibility “possibly, ϕ” is analyzed as ♢(RΛϕ), where and ♢ express the
outer notions of metaphysical necessity and possibility and R is a restricting
sentence. In the case of ‘Johnny can’t watch TV next door’, for example, the
restrictor R might be the sentence ‘Johnny satisfies his obligations’; in the
case of ‘Johnny can go to the moon’, R might specify the current state of
space travel (but not the fact that Johnny has no money for a ticket). On this
picture, though R can vary contextually, talk of outer possibility and necessity
is contextually constant.
A more attractive picture, I think, is that there is no such contextually
constant outer modality. Even the outer and ♢ vary contextually. In
Humean terms, there can be contextual variation both in the modal axioms
and in the modal rules, resulting in variation in the modal theorems and hence
in what is necessary and possible.
 This contextual variation may be motivated both from within the
perspective of Humeanism and directly from ordinary use of modal words.
From within a Humean perspective, contextual variation is natural to admit,
since nothing deep holds the modal axioms and rules together in the first
place. If there were a primitive glow associated with all necessary truths, then
it would be natural to have a contextually invariant word for that glow; but in
the absence of such a glow, why not allow contextual variation? And such
contextual variation seems called for by ordinary usage. Just as the following
speeches seem admissible:

“Johnny can’t go to the moon. Well, he could if he had enough money to buy
a ticket from NASA.” [shift from a kind of personal possibility to
technological possibility]
“Things can’t just disappear. Well, they could if the law of conservation of
matter were false.” [shift from nomic possibility to metaphysical possibility]

so do the following:

“Nothing can be at two places at once. Well, something could if the

occupation relation held in a one–many pattern.” [metaphysical possibility to
logical possibility?]
“Contradictions can’t be true. Well, they could if dialetheism were true.”
[logical possibility to ??]

It might be objected that we can follow up each of the second two speeches
with disavowals: “but the occupation relation just couldn’t hold in a one–
many pattern”, “but dialetheism just couldn’t have been true”. But these seem
to be no better conversational moves than the analogous moves in the first
speeches: “but he just couldn’t have obtained enough money for a ticket”,
“but the law of conservation of matter just couldn’t have been false”. English
modals are context-dependent through and through; there is no stable “outer
modality”.
This instability can also be seen in talk of the necessity of laws of
metaphysics. First, although philosophers—especially metaphysicians—tend
to speak of the laws of metaphysics as being noncontingent, in ordinary
contexts it isn’t unnatural to speak of the laws of metaphysics as being
contingent. And second, there seems to be some contextual variation, or at
least indeterminacy, in what counts as a law of metaphysics. Think, for
example, of the boundary between metaphysics and fundamental physics.
Supposing spacetime to in fact have four dimensions: is this a “mere physical
necessity”, or is it a “metaphysical necessity”? There’s no contextually
constant answer. Likewise for “Points of spacetime exist.”
What I have been characterizing in this chapter has been an outer
modality, a modality that is usually restricted in context. But what has
emerged in this section is that even this outer modality varies contextually. In
context, various modal axioms or rules may be contextually dropped,
resulting in a more restrictive notion of necessity and hence a broader sort of
possibility. Some modal axioms or rules may be harder to suspend than
others, as may be some patterns of suspension; but any pattern is in-principle
possible.30 Thus, there is no sharp line between restricted and unrestricted
metaphysical necessity.

12.8 The necessary a posteriori

“What about the necessary a posteriori?”
Hilary Putnam (1975b) argues by thought experiment that a substance on
another planet with all the superficial features of water would nevertheless
not count as water if it were not made up of H2O, but were instead made up
of some alien molecule XYZ. He concludes that it is necessary that all water
is made up of H2O. Since it was not knowable a priori what the chemical
composition of water was, we have an example of a necessary truth that is
knowable only a posteriori.
As Alan Sidelle (1989) argues, the necessary a posteriori provides a prima
facie challenge for “conventionalist” theories of modality. Avery simple
conventionalism that identified necessity with analyticity, construed as truth
by convention, would face the challenge most directly, for conventionalists
regarded analytic truths as being a priori. More vaguely, conventionalists
reject the picture that there is an independent modal realm “out there waiting
to be discovered”, holding instead that necessity somehow, has its source in
us; it is then tempting to think that all modal truths would be knowable a
priori. Since Humeans also reject the picture of modality being “out there”,
they too face the vague challenge.
Sidelle defends conventionalism from this objection. His strategy is to
“factor” a necessary a posteriori truth into an analytic component and a
nonmodal component. All the “modal force” (p. 37) of the necessary truth is
to come from the analytic component, whereas the a posteriority is to come
from the nonmodal component.31 In the case of ‘All water is made up of
H2O’, the analytic truth is:
(W) Whatever water’s deep explanatory feature, F, happens to be, it is
necessary that all water has F.
And the nonmodal component is that the deep explanatory feature of water is
in fact: being made of H2O.
There is a question of how the analyticity of (W) is supposed to help the
conventionalist. As Stephen Yablo (1992) points out, it wouldn’t do to think
of (W) as a definitional constraint on ‘water’, in the sense of a stipulation that
nothing is to count as water unless it possesses its deepest explanatory
feature necessarily. For that would seem to presuppose the existence of
modal features (and select the reference of ‘water’ on their basis), rather than
show modal features to be conventional. Sidelle has a different view of the
role of (W); but rather than discussing it, I will put forward my own view—in
the service of the Humean theory rather than conventionalism.32
The crucial thing is to regard (W) as a definitional constraint on
‘necessary’, not on ‘water’. Thus we stipulate that no property is to count as
the property of being a necessary proposition unless it is had by the
proposition that water has F, where F is water’s deep explanatory feature.
Unlike the stipulation considered by Yablo, this stipulation is nonmodal, and
thus does not presuppose modal features in its selection of reference.
Concretely: Humean necessity satisfies this stipulation provided we
introduce a new class of modal axioms, the natural kind axioms:
Natural kind axiom: a proposition expressed by a sentence of the form All
Fs are Gs , where F is a natural-kind term (such as ‘water’) and G
expresses the deep explanatory feature of the property expressed by F
Once these new axioms are included, the Humean theory implies that
‘Necessarily, water is made of H2O’, as well as Putnam’s other examples of
the necessary a posteriori, are indeed true.

12.9 Micro-reduction
“What about Lewis’s example of the talking donkey?”
 One of the chief advantages Lewis claimed for his modal realism against
rival theories of possible worlds was that only modal realism can reduce
modality. Consider for example “linguistic ersatzism”, according to which
possible worlds are maximal consistent sets of sentences in some suitable
world-making language. This rival apparently cannot reduce modality, since
the relevant notion of a consistent set of sentences must apparently be modal.
Mere logical consistency will not rule out possible worlds with married
bachelors, for instance. Lewis considered the move of defining consistency as
logical consistency with a set of “meaning postulates”, but rejected it with an
argument along the following lines (1986b, pp. 150–7).33
Let C0 completely describe New York City in every microphysical detail
(relationally as well as intrinsically). It is surely necessary that any C0 is a
city. The ersatzer must therefore include a sentence of the form:
(C) Any C0 is a city
among the meaning postulates. But no one knows what different
arrangements at the micro-level would suffice for the existence of a city, and
so no one knows what meaning postulates of the form (C) to add. (Lewis’s
example making this point involved ‘talking donkey’ rather than ‘city’.)
Even if the ersatzer does not know which particular sentences to add, she
might still produce some defining condition that would delineate the class of
meaning postulates so as to include the appropriate instances of (C). But this
must be done nonmodally; the ersatzer cannot, for instance, stipulate that all
necessary truths of form (C) are to count as meaning postulates.
Mark Heller has argued that the ersatzer need not solve this problem. In
Heller’s (1998) version of linguistic ersatzism, possible worlds are sets of
sentences in a language that can describe only the distribution of fundamental
properties over spacetime. Since the language contains no macro-predicates,
it lacks the predicate ‘is a city’. Consistency can then be just logical
consistency.34 The problem of micro-reduction then becomes one of how to
interpret these worlds, how to say when it is true in one of them that there
exists a city. But, Heller claims (1996), interpretation need not be regarded as
part of the theory of possible worlds. It is part of a project of the analysis of
ordinary concepts, and is optional for a philosopher of modality.
If all we wanted were a theory of possible worlds, then we could indeed
postpone solving the problem of micro-reduction in this way. But we want a
reduction of modality as well as a theory of worlds. The two are generally
connected thus:
It is necessary that ϕ iff for every possible world, w, it is true that ϕ in w.
This biconditional contains the locution it is true that ϕ in [possible world] w
. If the locution remains undefined then no analysis of modality has been
given. But to define it requires solving the problem of micro-reduction, for
the locution must be defined in the case where w is a world with subatomic
particles arranged in way C0 and ϕ is ‘There exists a city’.
This problem threatens the Humean as well, since sentences like (C) do
not seem to be mathematical truths, analytic truths, laws of metaphysics,
natural kind axioms, or logical consequences of such sentences.
The Humean should introduce another group of modal axioms, derived
from a theory of metaphysical truth-conditions (section 7.4) for the language
in question. There are, I assume, metaphysical truth-conditions for statements
about cities, smiles, and candy. The metaphysical semantics generating these
metaphysical truth-conditions, let’s assume, contains axioms of the form
“Predicate F applies to object x iff ϕ(x)”, where F may be ‘city’, ‘smile’,
‘candy’, and so on, and where ϕ(x) is phrased in purely fundamental terms.35
For each such axiom of the metaphysical semantics, we should add a
corresponding modal axiom:

∀x(Fx↔ϕ(x))

It is surely indeterminate exactly what the axioms of the true metaphysical

semantics are. Correspondingly, it is indeterminate which sentences of the
form (C) count as modal axioms. Note, though, that if C0 exactly describes
the actual microstate of New York City (as stipulated above), then surely (C)
will count as necessary on any reasonable way of resolving this
indeterminacy. For surely, on any reasonable way of resolving the
indeterminacy of what the theory of metaphysical truth-conditions looks like,
the actual microstate of New York City will be subsumed under ϕ in the
relevant axiom:36

∀x(city(x)↔ϕ(x))

I will not, because I cannot, give metaphysical truth-conditions for ‘city’,

‘smile’, or ‘candy’. I have not, therefore, fully specified metaphysical truth-
conditions for ‘necessarily’, since my specification of those truth-conditions
makes reference to the general theory of metaphysical truth-conditions for the
rest of the language. Nevertheless, we have all we should want in a reduction
of modality. For we never succeed in giving anything more than “toy”
metaphysical truth-conditions anyway, for notions of interest; recall section
7.6. The purpose of sketching toy truth-conditions is to show that words like
‘cause’, ‘necessary’, and so on, are as capable of resting upon the
fundamental as more mundane words like ‘city’, ‘smile’, and ‘candy’. We do
this by showing what form the unfathomably complex theory of metaphysical
truth-conditions might take to deal with them. The present theory
accomplishes this: if the theory of metaphysical truth-conditions can deal
with ‘city’, ‘smile’, and ‘candy’, then it can deal with ‘necessary’ as well—
the parts dealing with ‘city’, ‘smile’, and ‘candy’ get incorporated into the
part dealing with ‘necessary’.
It must be admitted that Lewis’s account of the necessity of statements
like (C) has an advantage. Lewis and I are both modal reductionists; thus we
both want to show that the necessitations of sentences like (C) can be given
nonmodal metaphysical truth-conditions (if I may transpose Lewis’s talk of
“analysis” into this key), thus showing that they fit into amodal fundamental
reality. My Humean approach assumes that the theory of metaphysical truth-
conditions contains bi-conditionals of the sort described above. Lewis, on the
other hand, requires no such assumption. For Lewis, the metaphysical truth-
condition of “Necessarily, every C0 is a city” is basically37 just “Absolutely
every C0 is a city”—a mundane quantified statement containing ‘city’. So to
show that the necessity of (C) fits into amodal fundamental reality, all Lewis
needs to assume is that mundane quantified sentences containing ‘city’ have
some nonmodal metaphysical truth-conditions or other.
Notice, by the way, that even a modalist, who takes necessity to carve at
the joints, must concede an advantage here to Lewis. Unlike Lewis and me,
the modalist is not trying to fit the necessitation of (C) into amodal reality.
But given purity (section 7.2), even a modalist cannot say that (C) is a
fundamental fact, for (C) contains the nonfundamental term ‘city’. So even
the modalist must supply metaphysical truth-conditions for (C) and other
micro-reduction sentences. Unlike me, the modalist can appeal to primitive
modal notions in doing so, but given purity, the sentences containing those
primitive modal notions cannot involve any nonfundamental notions, and so
don’t seem to be of any particular help here. Modalists, it would seem, can do
no better than I in giving metaphysical truth-conditions for microreduction
sentences; they must apparently appeal to the same biconditional in the
theory of metaphysical truth-conditions to which I appeal.

12.10 De re modality
“What about de re modality?”
De dicto modal claims, it is often said, concern the modal features of
propositions, whereas de re modal claims concern the modal features of
objects. “Necessarily: the number of the planets is odd” is said to be de dicto
because it attributes necessity to a proposition, the proposition that the
number of the planets is odd; whereas “The number of the planets is such
that: it is necessarily odd” is de re since it attributes a modal feature, being
necessarily odd, to an object, namely the object that in fact numbers the
planets—8, as we’re now told. This characterization is imperfect, but all I
need here is a rough-and-ready sense of the distinction.
The modal axioms and rules laid out so far need to be augmented to
deliver the full range of de re modal truths. For example, ‘Necessarily, Ted is
human’ is commonly supposed to be true: any entity of a nonhuman species,
no matter how like me in other respects, simply would not be me. But ‘Ted is
human’ is not a modal theorem, given the current definition.
One approach would parallel section 12.8’s treatment of the necessary a
posteriori. ‘Ted is human’ names a biological organism and specifies its
species; and it’s plausible that any such sentence is necessarily true if true at
all. So we could simply add to our list of modal axioms, all such sentences
that are true. More generally, Saul Kripke (1972) taught us of a wider class of
sentences containing proper names that are necessarily true if true at all.38
These include sentences of the following forms: i) “α is F”, where α is a
proper name and F is a “sortal” predicate (species predicates are a special
case); ii) “α is F”, where F specifies material origins; and iii) “α = β”, where
α and β are proper names. All true sentences of these forms could be added as
modal axioms. (A familiar theme: nothing deep need underlie the selection of
these new modal axioms; the selection reflects how our concept of necessity
happens to work.)
A second kind of de re modal claim involves “quantifying into modal
contexts”: a variable inside the scope of a modal operator is bound to a
quantifier outside, as in ‘∃x human(x)’ (“Something is such that it is
necessarily human”). The Humean theory so far supplies truth-conditions for
 ϕ only when ϕ lacks free variables; but to account for quantifying-in, truth-
conditions must also be given in the case where ϕ may contain free variables
(‘ human(x)’). The usual Tarskian strategy for assigning truth-conditions to
formulas with free variables is to relativize those truth-conditions to variable
assignments—functions from free variables to entities. Following this
strategy, the Humean could allow modal axioms and theorems to contain free
variables, and could relativize modal axiom-hood, rulehood, and
theoremhood to variable assignments. In categories i)–iii) of modal axioms
mentioned in the previous paragraph, α and β can now be variables as well as
proper names. So, for example, ‘human(x)’ would count as a modal axiom
relative to a variable assignment g that assigned me to x (category i) above),
and thus would count as a modal theorem relative to g; thus ‘ human(x)’
would be true relative to g; thus ‘∃x human(x)’ would be true simpliciter.
That, then, is one approach the Humean could take to de re modality. An
alternate approach is counterpart-theoretic. David Lewis (1968; 1971; 1986b)
defined modality in terms of possible worlds, but famously denied that
individuals are located in more than one possible world. As a result, he could
not agree with the usual possible-worlds definition of de re modality,
according to which an object is necessarily F iff that very object is F in every
possible world. Rather, he said, an object is necessarily F iff in every possible
world, that object’s counterpart is F, where a counterpart of an object in a
world is something that is similar enough to that object, and is more similar
to that object than anything else in its world. A Humean could base her
theory of de re modality on these ideas, as follows. First, take the Humean
account offered in the preceding sections as an account of de dicto necessity.
Second, use de dicto necessity to construct abstract possible worlds and
individuals, in any of the standard ways. For example, possible individuals
could be defined as certain sets of formulas, or certain sorts of properties.39
Third, introduce a counterpart relation over the abstract possible individuals.
Fourth, give counterpart-theoretic truth-conditions for de re modal sentences.
Thus, ‘Necessarily, Ted is human’ will turn out true iff all of my abstract
counterparts represent me as being human.40
Even Humeans who reject the counterpart-theoretic approach to de re
modality may pursue the general strategy of the previous paragraph to enrich
their account of modality in other directions. The general approach is to take
the Humean account as an account of some initial notion of necessity; use
that initial notion to introduce abstract possible worlds and individuals; and
then use those abstract possible worlds and individuals to define a further,
enriched, sort of necessity. For example, to introduce an actuality operator,
@, the Humean might build up abstract worlds using a @-free language, and
then give truth-conditions for @ in terms of those worlds. A similar strategy
could be used to give truth-conditions for statements about counterfactuals,
dispositions, supervenience, and other modal notions.

12.11 Family resemblances

Why are logical (or mathematical, or analytic, or …) truths necessary? The
Humean’s answer is that this is just how our concept of necessity works. One
can give no deeper answer to this question than to the question of why a
water glass counts as a cup (assuming that it does). There are many possible
meanings we could have chosen for ‘cup’; some include glasses, some don’t;
none carve nature at the joints better than any others; the meaning we have in
fact chosen includes glasses; and that’s all there is to it. Likewise for
‘necessary’.
This is the most flatfooted Humean answer, at any rate. But a Humean
need not be quite so flat-footed. Consider the various sorts of modal axioms.
She resists the idea that there is a single necessary and sufficient condition for
being a modal axiom. Nevertheless, she is free to exhibit similarities between
various modal axioms, just as one might exhibit similarities between things
that fall under our concept of a game, to use Wittgenstein’s example. Doing
this would help to show that the Humean concept of necessity is not utterly
arbitrary or heterogeneous.
In fact, many of the modal axioms I have listed do seem to have
something in common: a prioricity. The sort of evidence relevant to logical
and mathematical truths, to laws of metaphysics, and to most analytic truths,
is what is traditionally called a priori. This is not to say that they are known a
priori (for they may not be known at all), nor that they are even knowable a
priori (perhaps we have no way of knowing certain mathematical or
mereological truths), nor even to insist on a deep or sharp or absolute
distinction between the a priori and the a posteriori. Still, such access as we
do have to their truth does seem different from the kinds of access we have to
empirical truths of everyday life or science.41
Many of the axioms also seem to share the feature of being difficult to
conceive as being false; this, too, partially unifies the set of modal axioms.
 There are notorious difficulties with attempts to define necessity in terms
of a priority or conceivability. These would be explained if necessity is given
by a list that is not unified by any single feature, but which is partly unified
by a priority and conceivability. Compare the notorious difficulties with
attempts to define ‘game’ in terms of its unifying concepts (fun, competition,
and so on).
So some unification of the class of modal axioms is possible. But a
Humean will live with some disunification. She will be willing to admit that
there is no one feature that all the modal axioms have in common. The spirit
of Humeanism, after all, is that the line between the necessary and contingent
is not discovered, but rather drawn by us—perhaps somewhat arbitrarily.

12.12 Spreading arbitrariness

Humeanism’s implications might seem disturbing given how central modality
is to our cognitive lives. Historians speculate about what could have or would
have happened; bankers make decisions based on what must happen;
Humphrey regretted what might have been. If modal concepts are so
disunified, up to us, metaphysically second rate, then isn’t all this reasoning
radically defective? Have we centered our cognitive lives on a notion so
arbitrary as the notion of a bachelor, or cup, or martini?
Suppose someone viciously attacks a homeless person and steals his
possessions. We want to blame the attacker. Now, blame is often thought to
require the possibility of doing otherwise.42 But, one might worry,
Humeanism allows multiple, equally non-joint-carving, candidate meanings
of ‘could’, some counting the following sentence true, others counting it
false:

(AP) The attacker could have done otherwise.

Therefore, the worry continues, the sentence ‘The attacker is blameworthy’

also comes out true on some candidates and false on others. So this and other
attributions of blame are nonsubstantive in the sense of section 4.2.
In fact, Humeanism has no such consequence. First, although Humeanism
allows multiple candidates for modal words, there are limits to these
candidates. In particular, there surely are no candidates under which ‘it is
metaphysically impossible for the attacker to have done otherwise’ comes out
true. Any population using this sentence to express a truth would be
semantically alien to us. (Not that they would be failing to carve at the joints.
They would be like people who use ‘hut’ to refer to skyscrapers.)
This first observation is insufficient to rescue the substantivity of (AP),
since the ‘could’ in (AP) does not express metaphysical possibility. The
context of utterance of (AP) under consideration is an ordinary one, in which
moral responsibility is at issue; and in such a context, ‘could’ does not
express the broad notion of metaphysical possibility (the notion analyzed by
the Humean), but rather expresses some much more restricted sort, which we
might call “personal possibility”. Thus (AP) means something like this:
(AP′) It’s metaphysically possible that: R and the attacker do otherwise
where R is some “restrictor” sentence that specifies certain features of the
agent. (The nature of R is part of what gets disputed in debates over free will.
Compatibilists will include only a small set of features about the agent’s
nature and environment in R, whereas incompatibilists will include all facts
about the past and laws of nature.) So despite the previous paragraph, if the
restrictor sentence R has an appropriate range of candidate meanings, then
(AP′) may itself have a range of (equally joint-carving) candidate meanings,
some of which count it true, others of which count it false.
Thus if restricted notions of possibility have certain candidates, there
would indeed be a disturbing sort of nonsubstantivity of attributions of
blame.43 But this is not forced on us by Humeanism. For it was candidates for
R, not candidates for ‘metaphysically possible’, that generated the
nonsubstantivity. It is open to a Humean to claim that the relevant restrictor R
does not have candidates that count (AP′) as being false.
Furthermore, even if statements of blame do come out nonsubstantive, it is
important to recall the distinction between conceptual and metaphysical
substantivity from section 4.2. The existence of a range of equally joint-
carving candidates generating a mix of truth-values establishes metaphysical
nonsubstantivity, but is consistent with conceptual substantivity. It may be
that our choice of a candidate for ‘personally possible’ is deeply embedded in
our conceptual lives, and reflects our conception of moral responsibility.
Alternate ways of speaking about possibility and blame may be morally alien,
even if they aren’t semantically alien. And so, it may be argued, the
metaphysical nonsubstantivity of statements of blame isn’t particularly
disturbing, since such statements are conceptually substantive.
 Another example: counterfactual reasoning is obviously integral to our
cognitive lives in countless ways. Suppose, as is common, that
counterfactuals are understood in the Lewis/Stalnaker way, as being variably
strict conditionals defined over a space of possible worlds;44 and suppose we
reduce possible worlds to necessity in one of the usual ways. Given
Humeanism, there will be multiple candidates for the space of worlds
corresponding to the candidates of ‘necessary’. But the differences between
the candidates will affect only which “distant” possible worlds—distant in
the sense of the nearness metric introduced by Lewis and Stalnaker—are
included in the space, and thus will not affect the truth-values of ordinary
counterfactuals, which depend only on nearby worlds. For example, perhaps
some candidates for ‘necessary’ include the laws of metaphysics whereas
others do not, so that some candidate spaces of worlds include
metaphysically odd worlds (containing multi-located objects, singing
numbers, and the like) whereas others exclude such worlds. This multiplicity
of candidates would not cause an ordinary counterfactual like “If I had struck
the match, it would have lit” to be nonsubstantive, since this counterfactual is
true iff the match lights in the nearest world in which I strike it, and that
nearest world is not going to be a metaphysically odd one, even if
metaphysically odd worlds are present. It is the nearness metric, not the space
of possible worlds, that is responsible for whatever objectivity or
nonarbitrariness counterfactuals have.

1 Goodman (1955, p. 40)

2 Contrast the less direct approach of Sider (2001b, chapter 2, section 3), which disparaged
primitive modality as a gateway drug to stronger stuff: primitive counterfactuals, dispositions, and
tense.
3 Indeed, Fine defines modality in terms of essence: to be metaphysically necessary is to be true in
virtue of the essences of all objects, including abstract objects (1994a, p. 9; see also 1994b). One
reservation: assume Platonism about set theory, and that there is a fact of the matter about the
continuum hypothesis. We would then expect the continuum hypothesis to be either necessarily true or
necessarily false. But it’s hard to see how its truth or falsity could flow from the essences of the
relevant entities: existence, sethood, and membership.
4 Although see Divers andMelia (2002); Shalkowski (1994); I reply in my (2003).
5 See Adams (1974); Plantinga (1976); Stalnaker (1976).
6 Other nonLewisian stories about possible-worlds talk are combinatorialist, fictionalist, or both
(Armstrong (1989; 1997); Rosen (1990)), but I have argued elsewhere (2002; 2005) that these projects
fail to reduce modality
 7 I choose the term because of my denial of modal “further facts”, but the term is imperfect since
universality plays no role in my account. I’m not sure whom else to count as Humeans, but to varying
degrees the following philosophers share the spirit: Heller (1996; 1998); Mortensen (1989); Nolan
(2011); Peacocke (1997; 1999); Sidelle (2009).
8 Statements of necessity are not fundamental truths, since they contain the non-joint-carving term
‘necessary’.
9 Truth by convention would not have demystified modality on its own. Even if linguistic meaning
somehow generated truth, it would not follow immediately that it also generated necessary truth.
10 Michael Jubien (2007) and Marc Lange (2005) seem to defend governance theories.
11 See Beebee (2000); Loewer (1996).
12 There is an intermediate conception, which like the governance conception holds that necessity is
a “further status” that is intimately connected with truth, but unlike the governance conception does not
hold that the further status “produces” truth. But what exactly is the nature of the intimate connection, if
not production? Could it be understood in terms of necessity itself: it’s necessary that any proposition
with the further status is true? Thanks to Ryan Robinson and Kit Fine.
13 This is what drives Jubien’s (2007, section 1) objection to Lewis:
Suppose it’s necessary that all A’s are B’s. This is supposed [by the possible worlds theorist] to mean
that in every possible world, all A’s are B’s. So the necessity arises from what goes on in all the worlds
taken together. There’s nothing about any world individually, even in all of its maximal glory, that
forces all of its A’s to be B’s. It’s as if it just happens in each world that all of its A’s are B’s, that from
the strictly internal point of view of any world, it’s contingent, a mere coincidence. But then shouldn’t
we expect that this internal contingency will not be repeated in every world, that there will be worlds
where some As “happen” not to be B’s? After all, nothing within any given world prevents it, and these
are supposed to be all the possible worlds.
14 That is, the intersection of all sets A such that: i) each modal axiom is in A; and ii) for any B ⊆
A, if B bears some modal rule to some sentence ϕ then ϕ ∈A.
15 It’s best to proceed in terms of rules and axioms because, e.g., logical consequences of
propositions of different types may not fall under any type, but are nevertheless necessary.
16 For example, the account defines a property of propositions that do not themselves concern
modality, and thus is insufficient to interpret iterable modal operators.
17 Compare Hanson (1997).
18 Quine (1960a); p. 103 in Quine (1966).
19 See Etchemendy (1990).
20 A similar response could be based on the views of Williamson (1998; 2002).
21 The set of logical words should not include modal words. That would threaten circularity; also,
some modal languages arguably contain logical truths, such as If ϕ then Actually ϕ , that are not
necessary.
22 Like the Humean account of necessity, this is a classification rather than governance account.
23 For simplicity, I write as if mereological vocabulary is fundamental, but see chapter 13.
24 Although see Cameron (2007).
 25 Qualification: Humeanism doesn’t quite imply that fundamental propositions of metaphysics are
necessarily false when false; what it implies is that this holds whenever the fundamental proposition has
a true “rival” in the sense to be explained.
26 See Sider (2011, section 9) for a fuller presentation of this point.
27 Thanks to Ernie Sosa here.
28 Similar remarks apply to Funkhouser (2006).
29 Perhaps epistemic and deontic modalities are included in this picture, perhaps not.
30 We may want to allow addition as well as suspension. For example, in a context where we are
considering the possibility of dialetheism, but in which we don’t want to count everything as possible,
we must suspend classical rules of inference, but we must then put some weaker rules in their place.
31 See also Chalmers (1996); Coppock (1984); Hirsch (1986); Jackson (1998).
32 Sidelle himself regards (W) as a definitional constraint on how to use ‘water’ in counterfactual
situations, where those counterfactual situations are given in “identity-neutral” terms (in terms of “pre-
objectual stuff”). I prefer the account in the text for two reasons. First, my account does not require a
metaphysics of stuff (about which I am skeptical—recall section 9.6.2). Second, Sidelle’s account
presupposes talk of counterfactual situations and thus is not fully reductive (though I grant that since he
does not presuppose talk of identities of natural kinds and ordinary objects between distinct
counterfactual situations, his account is in an important sense conventionalist about de re modality).
33 A second argument was that no one knows whether it is metaphysically impossible or merely
nomically impossible for a particle to be both positively charged and negatively charged, and so the
ersatzer does not know whether to include a meaning postulate prohibiting this combination (1986b, pp.
154–5). The Humean can respond that if ‘nothing is both negatively and positively charged’ is a law of
metaphysics (like analogous sentences concerning distinct determinates of a given determinable), then
it is a modal axiom and hence is necessary (Perhaps it is indeterminate whether it is a law of
metaphysics, in which case it is indeterminate whether it is necessary)
34 A question I will not pursue: why should fundamental properties necessarily apply to points
rather than regions?
35 Given a restrictive fundamental ontology, existential quantification over Fs in natural language
might have nonexistential truth-conditions, in which case the relevant axioms would have a different
form (see section 7.4).
36 ‘Subsumed’ here is deliberately vague. Two ways of making it precise. One: the language of the
metaphysical truth-conditions is infinitary; the axiom has the form ∀x(city(x)↔(ϕ1(x)V…)); C0 is one
of the ϕ is (on each resolution of the indeterminacy of ‘metaphysical truth-conditions’); the modal rule
of inference in the Humean theory is infinitary logical consequence; thus (C) is a modal theorem. Two:
the language of the metaphysical truth-conditions can represent quantification over properties; the
axiom has the form ∀x(city(x) ↔ ∃p(x instantiates pΛϕ(p)); ‘the property of being C0 is ϕ’ is a modal
axiom (including this sort of axiom would be natural given this higher-order setting; such sentences are
analogous to mathematical truths) as are suitable sentences of the form ∀x(ψ(x) ↔ x instantiates the
property of being ψ); therefore (C) is again a modal theorem.
37 Not quite: it’s only the C s that are parts of maximal spatiotemporally interrelated entities—the
0
entities that Lewis calls possible worlds—that must be cities.
38 Accommodating the analogous sentences containing indexicals and demonstratives (Kaplan,
1989) would require relativizing modal axiomhood to context, and would also require other
adjustments if, for example, the contingent sentence ‘I am here now’ is regarded as a logical truth.
39 See Lewis (1986b, section 3.2).
40 See Sider (2002; 2006) for more discussion.
41 Those sympathetic with the two-dimensional approach to semantics might try to extend this
similarity further, even to the necessary a posteriori.
42 Let’s not enter into the subtleties to which this formula has been subjected.
43 See Hawthorne (2001) for a discussion of this issue.
44 See Lewis (1973b); Stalnaker (1968). Similar remarks would apply to more recent contextualist
accounts of counterfactuals as strict conditionals; see Lycan (2001); von Fintel (2001).
13 A Worldview
Let us end on a concrete note. What might a comprehensive “worldview”
look like, given realism about structure?
Think of a worldview as consisting of i) an ideology; ii) a fundamental
theory phrased in terms of the ideology, specifying laws of metaphysics and
perhaps other principles; and iii) a sketch of a metaphysical semantics for
nonfundamental discourse in terms of the ideology. I will put forward a
worldview according to which fundamental reality contains nothing but
physics, logic, and set theory. While I believe that this worldview may well
be true, I won’t say much in its defense; the point is to illustrate.
First, ideology. My primitive notions are those of first-order quantification
theory (with identity), plus a predicate ∈ for set-membership, plus predicates
adequate for fundamental physics, plus the notion of structure.
Next, the fundamental theory. Since my ideology includes the first-order
quantifiers, one part of giving the laws of metaphysics will be the statement
of an ontology—a statement, in general terms, of what there is. My
worldview’s ontology contains only points of spacetime1 and sets, both pure
and impure. Thus it contains no composite objects. This is not to say,
however, that ‘There are no composite objects’ or even ‘Everything is a set or
a point of spacetime’ is a law of metaphysics. For ‘composite’, ‘set’, and
‘point’ are not in my fundamental ideology. What might an official statement
of a law of metaphysics giving my ontology look like? Suppose one of the
physical predicates is the topological predicate ‘x is open’ (meaning: open
set). In that case the law could be:

There is exactly one thing that has no members but is not a member of any
open thing [this is the null set]; everything else either has a member [and so is
a set], or is a member of some open thing [and so is a point of spacetime].

Since there are no composite entities in its ontology, my worldview is a

version of mereological nihilism.2 Since its ontology contains spacetime but
no further entities occupying spacetime, it is a version of
supersubstantivalism. Since there are no tense operators in its ideology, it is a
version of the spatializing, “B-” theory of time. (Temporal discourse rests on
the attributions of geometric predicates— which I assume to be required for
physics—to spacetime points.) Since there are no aesthetic, moral, or
supernatural notions in its ideology, it is a version of naturalism. Since there
are no causal, nomic, or modal notions in its ideology, it is a version of
Humeanism.
Next we have laws of metaphysics concerning the behavior of ∈. These
can be taken to be the axioms of impure first-order ZFC set theory.3 (Perhaps
these should not be called laws of metaphysics, since they are also laws of
mathematics.)
Next we have laws of metaphysics concerning the behavior of the physical
predicates. These are given to us by physics. (Perhaps these should not be
called laws of metaphysics, since they are also laws of physics.)
Next there are pseudo-laws of metaphysics concerning the behavior of the
logical constants. These are the logical truths of first-order quantification
theory (such as ‘∃x x∈x →∃x x∈x’), which I take to be classical. They are
perhaps not genuine laws because they are not generalizations (not all of
them, anyway), but I call them pseudo-laws because they fall under
generalizations such as “∀P (P→P)” or “For every sentence, ϕ, ϕ→ϕ is
true.” (Since my fundamental ideology contains neither sentential
quantification nor the notion of a sentence nor the corner-quotes nor a truth
predicate, the generalizations themselves are not laws, and can only be stated
in a nonfundamental language.)
Finally, there are principles asserting that my primitive notions carve at
the joints: and so on.
What remains is to sketch a metaphysical semantics for ordinary and
scientific language. This is a formidable task, since the worldview is so
austere.
Let’s begin with talk about ordinary physical objects. Although my
ontology contains no physical objects per se, it does contain entities with
which they may naturally be identified: the sets of spacetime points that they
occupy.4 I, for example, can be identified with a set whose earliest points are
around 1967, whose temporal cross-sections are person-shaped, and which
continues on into the future for an unknown duration.5
A metaphysical semantics based on this identification would construe talk
of ordinary physical objects as being about sets of spacetime points. It would
interpret names of physical objects as referring to the sets with which the
objects are identified, predicates as applying to ’tuples of sets of spacetime
points, and quantifiers as ranging over sets of spacetime points.
Despite the fact that my ontology is in a sense mereological nihilist, the
proposed metaphysical semantics is not like the one at the end of section 7.7.
It assigns to existential sentences about physical objects:

There is an x such that: ψ(x)

metaphysical truth-conditions that are themselves existential in form:

There is an x such that: ψ*(x).

Even though my fundamental ideology lacks physical-object-theoretic

notions, my fundamental ontology is rich enough to include entities—sets—
that may be identified with physical objects. There is no need for an
ontological reduction of physical objects, so to speak, only a predicate
reduction of ordinary physical predicates ψ to fundamental predicates ψ*.
What will this predicate reduction look like? Some predicates are
straightforward. For example, the metaphysical truth-condition for ‘Physical
object x is located at spacetime point p’ could be that x (i.e., the set identified
with x) contains p as a member. The truth-condition for ‘Physical object x is
atemporally part of physical object y’ could be ‘x ⊆ y.’ The truth-condition
for ‘Physical object x is part of physical object y relative to time
(hypersurface) t’ could be ‘x ⋂ t ⊆ y ⋂ t.’ Thus our metaphysical semantics
encompasses talk of the location and mereology of ordinary physical
objects.6
But what of predicates like ‘x is a table’ or ‘y is a person’? How can we
say, in the language of physics, logic, and set theory, which sets of points of
spacetime are tables and persons?
Of course I have no clue. I don’t even have much of a clue about how to
construct a “toy” metaphysical semantics (section 7.4) for such predicates.
Nevertheless, I am confident that a metaphysical semantics could in principle
be given.
The reason is that the array of definable relations is extremely rich, given
the iterative resources of set theory.7 Let’s identify relations with their
extensions. (This is a harmless simplifying assumption, since we could
instead substitute “structured relations” constructed set-theoretically from the
extensions of fundamental predicates.8) Suppose we have defined some initial
predicates of relations, ϕ(r) and ψ(r, s), say. We can then define “higher-
order” predicates of individuals using quantification over relations that satisfy
the initial predicates, for example:

(R1(x,y) holds iff x bears each ϕ relation to everything that is more massive
than y; R2(x,y) holds iff some pair of relations standing in ψ hold between x
and y in opposite directions.) Moreover, we can define higher-order
predicates of relations, again by quantifying over relations that satisfy the
initial predicates. This procedure can then be iterated. After defining higher-
order predicates of relations using quantification over relations that satisfy ϕ
and ψ, we can quantify over relations that satisfy those predicates to define
still higher-order predicates of individuals and relations; and so on.
These iterative constructions can at each stage make use of any notions
that can be defined in fundamental terms. For example, suppose we define a
notion of metaphysical necessity as described in chapter 12; then modal
notions may be employed in the iterations. Given a notion of necessity, we
might define corresponding possible worlds—maximal consistent sets of
sentences, perhaps. We might then use these worlds in a reduction of
causation, counterfactuals, chance, and other notions, as Lewis envisaged.9
All such reductionist programs may be brought into the iterative construction
of further relations. Given these possibilities for definition, I see no in-
principle barrier to the possibility of giving a metaphysical semantics for
predicates of ordinary physical objects, as well as for the distinctive
vocabularies of chemistry, biology, psychology, economics, semantics, and
other disciplines.
We needn’t always identify objects with sets of spacetime points;
sometimes more complex set-theoretic constructions are called for. Consider,
for example, metalinguistic talk—talk about language. My worldview
includes no linguistic ideology, no notions of predicate, sentence, conjunct,
satisfies, true, means, pragmatically implicates. So if we want to speak of
languages (whether formal or natural), we need a metaphysical semantics.
Now, it is perhaps natural to identify a “linguistic atom” (symbol, for a
formal language; word, morpheme, phoneme, or whatever, for a spoken
language) with a set of spacetime points: the set of points located inside some
production (inscription or utterance) of that symbol. But we cannot construe
all linguistic entities this way, since many complex linguistic entities will
never be produced (the conjunction of the longest two sentences ever
produced, for example). So it’s natural instead to take complex linguistic
entities to be set-theoretic constructions (perhaps sequences, perhaps tree-
structures) of linguistic atoms, and thus as set-theoretic constructions of sets
of points, rather than as mere sets of points.
This metaphysical semantics associates existential truth-conditions with
sentences that quantify existentially over linguistic entities—set-theoretic
constructions out of sets of spacetime points really exist. The hard work, as
with physical-object talk, comes with the predicates. Syntactic predicates are
comparatively easy. For example, if c is the set of spacetime points identified
with the sign for conjunction, then the metaphysical truth-condition for the
syntactic predicate ‘s is the conjunction of s1 and s2’ might be something like
this: ‘s is a tree with top node c containing two child nodes, one of which is
s1, the other of which is s2.’ Semantic, pragmatic, and other linguistic notions
will be more difficult, but the project here is continuous with the general
project of giving a metaphysical semantics for talk about macro-entities, and
the iterative techniques described above may again be used.
The identifications I have proposed—of physical objects with sets of
occupied points, linguistic atoms with sets of production points, linguistic
complexes with set-theoretic constructions from linguistic atoms—are
somewhat arbitrary. They’re not wholly arbitrary; it’s not as if I’ve simply
observed that the pure settheoretic hierarchy has enough entities with which
to identify everything, and left it at that. But there’s no denying that there are
multiple alternates to my sketch of a metaphysical semantics. Fortunately,
this multiplicity is harmless. What we want out of a metaphysical semantics
for L is a good explanation of the linguistic behavior of speakers of L, and
there is often an element of arbitrariness when explaining higher-level
phenomena. What we’re after in linguistics (and psychology, and economics,
and …) is a good model, not a unique model.10
I have imagined one way the book of the world might be. It is not a tale of
common sense. But we can, I think, recognize it as our own.
 1 Or perhaps points of some higher-order space, if the foundations of quantum mechanics lead in
this direction.
2 Ideological (rather than ontological) parsimony leads me to this view: nihilism lets us eliminate
parthood from fundamental ideology; set-membership can take up the slack. See Sider (2011).
3 Define an urelement to be any member of anything that is open.
4 Compare Quine (1976b).
5 An alternate approach would be that of Sider (1996a).
6 My initial description of this worldview as a version of mereological nihilism must therefore be
qualified. Although the worldview denies that there are fundamental ascriptions of parthood, it allows
that sentences of nonfundamental languages can truly ascribe parthood. Moreover, since both
composites and their parts are identified with sets, it even allows that there exist, in the fundamental
sense of ‘there exist’, entities that are asserted by nonfundamental sentences to have parts.
7 See also the discussion of infinite definitions in section 7.11.1.
 8 Compare Lewis (1986b, section 1.5).
9 See Lewis (1973a; b; 1994).
10 The arbitrariness would not be harmless if we applied notions from our fundamental ideology to
the arbitrarily constructed entities—qua arbitrarily constructed entities, so to speak. It would not be
harmless, for example, to construe sentences (qua sentences) as being fundamentally located in
spacetime, in the sense of being subsets of open sets. On the one hand, we are surely free to choose any
reasonable method for identifying sentences with sets; but on the other hand, what if the chosen method
identifies sentences with sets that are not subsets of open sets? (See Forrest and Armstrong (1984);
Sider (1996c).)
Recall the claim in section 6.3 that since semantics isn’t fundamental, we can’t take the facts about
structure to fundamentally involve semantic entities. I wrote as if the problem was that semantic entities
don’t fundamentally exist. But that isn’t quite right, since our metaphysical semantics might identify
semantic entities with certain set-theoretic constructions, which do fundamentally exist. The problem is
rather this: since which sets are identified with semantic entities is arbitrary, and since the notion of
structure, , is part of our fundamental ideology, we can’t apply it to semantic entities (qua semantic
entities).
References
Adams, Robert Merrihew (1974). Theories of actuality. Noûs, 8, 211–31.
Albert, David Z. (1996). Elementary quantum metaphysics. In Bohmian Mechanics and Quantum
Theory: An Appraisal (eds. J. T. Cushing, A. Fine, and S. Goldstein), pp. 277–84. Kluwer
Academic Publishers, Dordrecht.
Armstrong, David M. (1978a). Nominalism and Realism, volume 1 of Universals and Scientific
Realism. Cambridge University Press, Cambridge.
— (1978b). A Theory of Universals, volume 2 of Universals and Scientific Realism. Cambridge
University Press, Cambridge.
— (1983). What Is A Law Of Nature? Cambridge University Press, Cambridge.
— (1986). In defence of structural universals. Australasian Journal of Philosophy, 64, 85–8.
— (1989). A Combinatorial Theory of Possibility. Cambridge University Press, New York.
— (1997). A World of States of Affairs. Cambridge University Press, Cambridge.
— (2004). Truths and Truthmakers. Cambridge University Press, Cambridge.
Arntzenius, Frank (2008). Gunk, topology, and measure. In Oxford Studies in Metaphysics (ed. Dean
W. Zimmerman), volume 4, pp. 225–47. Oxford University Press, Oxford.
Arntzenius, Frank and Hawthorne, John (2005). Gunk and continuous variation. The Monist, 88,
441–65.
Ayer, Alfred Jules (1936). Language, Truth and Logic. Victor Gollancz Ltd, London. 2nd edition.
Bach, Kent (1994). Conversational implicature. Mind and Language, 9, 124–62.
Bealer, George (1982). Quality and Concept. Clarendon Press, Oxford.
Beall, J. C. and Restall, Greg (2006). Logical Pluralism. Clarendon Press, Oxford.
Beebee, Helen (2000). The non-governing conception of laws of nature. Philosophy and
Phenomenological Research, 61, 571–94.
Bennett, Karen (2004). Spatio-temporal coincidence and the grounding problem. Philosophical
Studies, 118, 339–71.
— (2006). Zombies everywhere! MS. Available at http://courses.cit.cornell.edu/kb383/zombies.pdf.
— (2009). Composition, colocation, and metaontology. In Chalmers et al. (2009), pp. 38–76.
Bigelow, John (1988). The Reality of Numbers: A Physicalist’s Philosophy of Mathematics.
Clarendon Press, Oxford.
Blackburn, Simon (1993). Essays in Quasi-Realism. Oxford University Press, Oxford.
Boghossian, Paul (1997). Analyticity. In A Companion to the Philosophy of Language (eds. Bob Hale
and Crispin Wright), pp. 331–68. Blackwell, Oxford.
— (2003). Blind reasoning. Aristotelian Society, Supplementary Volume, 77, 225–48.
Boolos, George (1984). To be is to be the value of a variable (or to be some values of some
variables). Journal of Philosophy, 81, 430–49.
Bourne, Craig (2002). When am I? A tense time for some tense theorists? Australasian Journal of
Philosophy, 80, 359–71.
Boyd, Richard (1988). How to be a moral realist. In Essays on Moral Realism (ed. Geoffrey Sayre-
McCord), pp. 181–228. Cornell University Press, Ithaca, NY.
Braddon-Mitchell, David (2004). How do we know it is now now? Analysis, 64, 199–203.
Båve, Arvid (2010). How to precisify quantifiers. Journal of Philosophical Logic, 40, 103–11.
Bricker, Phillip (1992). Realism without parochialism. Read to the Pacific Division of the APA.
Available at http://www.umass.edu/philosophy/PDF/Bricker/realism_without_parochialism.pdf.
— (1993). The fabric of space: Intrinsic vs. extrinsic distance relations. In Midwest Studies in
Philosophy XVIII: Philosophy of Science (eds. Peter French, Theodore E. Uehling, Jr, and
Howard K. Wettstein), pp. 271–94. University of Notre Dame Press, South Bend, IN.
Britton, Karl (1947). Are necessary truths true by convention? Aristotelian Society, Supplementary
Volume, 21, 78–103.
Broad, C. D. (1923). Scientific Thought. Harcourt, Brace and Company, New York.
— (1936). Are there synthetic a priori truths? Aristotelian Society, Supplementary Volume, 15, 102–
17.
Burgess, John (2005). Being explained away. Harvard Review of Philosophy, 13, 41–56.
Cameron, Ross P. (2007). The contingency of composition. Philosophical Studies, 136, 99–121.
— (2008). Turtles all the way down: Regress, priority and fundamentality in metaphysics.
Philosophical Quarterly, 58, 1–14.
— (2010). From humean truthmaker theory to priority monism. Noûs, 44, 178–98.
— (2011). Truthmakers. In Oxford Handbook of Truth (ed. Michael Glanzberg). Oxford University
Press, Oxford. Forthcoming.
Carnap, Rudolf (1937). The Logical Syntax of Language. Routledge & Kegan Paul, London.
— (1950). Empiricism, semantics and ontology. Revue International de Philosophie, 4, 20–40.
Reprinted in Meaning and Necessity: A Study in Semantics and Modal Logic, 2nd edn. University
of Chicago Press, Chicago.
Chalmers, David (1996). The Conscious Mind. Oxford University Press, Oxford.
— (2005). The Matrix as metaphysics. In Philosophers Explore the Matrix (ed. Christopher Grau),
pp. 132–76. Oxford University Press, New York.
— (2009). Ontological indeterminacy. In Chalmers et al. (2009), pp. 77–129.
— (2011). Verbal disputes. Philosophical Review. Forthcoming.
Chalmers, David, Manley, David, and Wasserman, Ryan (eds.) (2009). Metametaphysics. Oxford
University Press, Oxford. Chisholm, Roderick (1976). Person and Object: A Metaphysical Study.
Open Court Publishing Co., La Salle, Illinois.
Chomsky, Noam (2000). New Horizons in the Study of Language and Mind. Cambridge University
Press, New York.
Churchland, Paul M. (1981). Eliminative materialism and the propositional attitudes. Journal of
Philosophy, 78, 67–90.
Coppock, Paul (1984). Review of Nathan U. Salmon, Reference and Essence. Journal of Philosophy,
81, 261–70.
Correia, Fabrice and Schnieder, Benjamin (eds.) (2012). Grounding. Cambridge University Press,
Cambridge. Forthcoming.
Dasgupta, Shamik (2009a). Individuals: An essay in revisionary metaphysics. Philosophical Studies,
145, 35–67.
— (2009b). Symmetry in reality. Ph.D. thesis, New York University, New York.
— (2010). On the plurality of grounds. MS. Available at
http://www.shamik.net/Research_files/dasgupta%20on%20the%20plurality%20of%20grounds.pdf
Davidson, Donald (1967a). The logical form of action sentences. In The Logic of Decision and
 Action (ed. Nicholas Rescher), pp. 81–95. University of Pittsburgh Press, Pittsburgh, PA.
— (1967). Truth and meaning. Synthese, 17, 304–23.
— (1970). Mental events. In Experience and Theory (eds. L. Foster and J. Swanson), pp. 79–101.
Duckworth, London.
— (1977). The method of truth in metaphysics. In Midwest Studies in Philosophy II: Studies in the
Philosophy of Language (eds. Peter French, Theodore E. Uehling, Jr, and Howard K. Wettstein),
pp. 244–54. University of Minnesota Press, Minneapolis, MN.
deRosset, Louis (2010). Getting priority straight. Philosophical Studies, 149, 73–97.
Devitt, Michael and Sterelny, Kim (1999). Language and Reality: An Introduction to the Philosophy
of Language. MIT Press, Cambridge, MA, 2nd edition.
Divers, John (2002). Possible Worlds. Routledge, London.
Divers, John and Melia, Joseph (2002). The analytic limit of genuine modal realism. Mind, 111, 15–
36.
Dorr, Cian (2002). The simplicity of everything. Ph.D. thesis, Princeton University.
Dorr, Cian (2004). Non-symmetric relations. In Oxford Studies in Metaphysics (ed. Dean W.
Zimmerman), volume 1, pp. 155–92. Oxford University Press, Oxford.
— (2005). What we disagree about when we disagree about ontology. In Fictionalism in Metaphysics
(ed. Mark Kalderon), pp. 234–86. Oxford University Press, Oxford.
— (2007). There are no abstract objects. In Sider et al. (2007), pp. 32–63.
— (2010). How to be a modal realist. http://users.ox.ac.uk/~sfop0257/papers/ModalRealism.pdf.
Dowty, David R., Wall, Robert E., and Peters, Stanley (1981). Introduction to Montague Semantics.
Kluwer, Dordrecht.
Dreier, James (2004). Meta-ethics and the problem of creeping minimalism. Philosophical
Perspectives, 18, 23–44.
Dretske, Fred (1977). Laws of nature. Philosophy of Science, 44, 248–68.
Dummett, Michael (1973). The philosophical basis of intuitionist logic. In Proceedings of the Logic
Colloquium, Bristol, July 1973 (eds. H. E. Rose and J. C. Shepherdson), pp. 5–49. North-
Holland, Amsterdam.
Eddon, Maya (2012). Fundamental properties of fundamental properties. In Oxford Studies in
Metaphysics (eds. Karen Bennett and Dean W. Zimmerman), volume 8. Oxford University Press,
Oxford.
Eklund, Matti (2002). Inconsistent languages. Philosophy and Phenomenological Research, 64, 251–
75.
— (2006). Neo-fregean ontology. Philosophical Perspectives, 20, 95–121.
— (2007). The picture of reality as an amorphous lump. In Sider et al. (2007), pp. 382–96.
— (2008). Putnam on ontology. In Following Putnam’s Trail: On Realism and Other Issues (eds.
Maria Uxia Rivas Monroy, Concepcion Martinez Vidal, and Celeste Cancela), pp. 203–22.
Rodopi, Amsterdam.
— (2009). Carnap and ontological pluralism. In Chalmers et al. (2009), pp. 130–56.
Etchemendy, John (1990). The Concept of Logical Consequence. Harvard University Press,
Cambridge, MA.
Everett, Anthony and Hofweber, Thomas (eds.) (2000). Empty Names, Fiction and the Puzzles of
Non-Existence. CSLI Press, Stanford.
Field, Hartry (1980). Science Without Numbers. Blackwell, Oxford.
— (1989). Realism, Mathematics and Modality. Blackwell, Oxford.
— (2003). The semantic paradoxes and the paradoxes of vagueness. In Liars and Heaps (ed. J. C.
Beall), pp. 262–311. Oxford University Press, Oxford.
— (2008). Saving Truth from Paradox. Oxford University Press, Oxford.
Fine, Kit (1994a). Essence and modality. In Philosophical Perspectives 8: Logic and Language (ed.
James Tomberlin), pp. 1–16. Ridgeview, Atascadero, CA.
— (1994b). Senses of essence. In Modality, Morality, and Belief (ed. Walter Sinnott-Armstrong), pp.
53–73. Cambridge University Press, New York.
— (2000). Neutral relations. Philosophical Review, 109, 1–33.
— (2001). The question of realism. Philosopher’s Imprint, 1, 1–30.
— (2003). The problem of possibilia. In Loux and Zimmerman (2003), pp. 161–79.
— (2005). Tense and reality. In Modality and Tense, pp. 261–320. Oxford University Press, New
York.
— (2009). The question of ontology. In Chalmers et al. (2009), pp. 157–77.
— (2010). Some puzzles of ground. Notre Dame Journal of Formal Logic, 51, 97–118.
— (2011). The pure logic of ground. MS.
— (2012). Guide to ground. In Correia and Schnieder (2012). Forthcoming.
Fitelson, Branden (2006). The paradox of confirmation. Philosophy Compass, 1, 95–113.
Fodor, Jerry A. (1974). Special sciences (or: The disunity of science as a working hypothesis).
Synthese, 28, 97–115.
— (1987). Psychosemantics. MIT Press, Cambridge, MA.
Forrest, Peter (2004). The real but dead past: A reply to Braddon-Mitchell. Analysis, 64, 358–62.
— (2006). Uniform grounding of truth and the growing block theory: A reply to Heathwood.
Analysis, 66, 161–3.
Forrest, Peter and Armstrong, David M. (1984). An argument against David Lewis’ theory of
possible worlds. Australasian Journal of Philosophy, 62, 164–8.
Frege, Gottlob (1879). Begriffsschrift, ein der arithmetischen nachgebildete Formelsprache des
reinen Denkens. Verlag L. Nebert, Halle/Saale. Translated as Concept Script, A Formal
Language of Pure Thought Modelled Upon That of Arithmetic, by S. Bauer-Mengelberg in From
Frege to Gödel: A Source Book in Mathematical Logic, 1879–1931 (ed. J. van Heijenoort, 1967),
pp. 1–82. Harvard University Press, Cambridge, MA.
— (1884). The Foundations of Arithmetic. Blackwell, Oxford, 2nd edition.
— (1952/1892). On sense and reference. In Translations of the Philosophical Writings of Gottlob
Frege (eds. Peter Geach and Max Black). Blackwell, Oxford.
— (1997/1918). Thought. In The Frege Reader (ed. Michael Beany), pp. 323–45. Blackwell, Oxford.
Frost, Robert (1936). A Further Range. H. Holt, New York.
Funkhouser, Eric (2006). The determinable-determinate relation. Noûs, 40, 548–69.
Gamut, L. T. F. (1991). Logic, Language, and Meaning, Volume 2: Intensional Logic and Logical
Grammar. University of Chicago Press, Chicago, IL.
Gettier, Edmund L. (1963). Is justified true belief knowledge? Analysis, 23, 121–3.
Goodman, Nelson (1955). Fact, Fiction, and Forecast. Harvard University Press, Cambridge, MA.
Goodman, Nelson (1978). Ways of Worldmaking. Hackett, Indianapolis.
Grice, Paul (1975). Logic and conversation. In Syntax and Semantics, Volume 3: Speech Acts (eds.
Peter Cole and Jerry Morgan), pp. 41–58. Academic Press, New York.
Grice, Paul and Strawson, Peter F. (1956). In defense of a dogma. Philosophical Review, 65, 141–58.
Grünbaum, Adolf (1973). Philosophical Problems of Space and Time. D. Reidel, Dordrecht, 2nd
edition.
Gunderson, Keith (ed.) (1975). Language, Mind and Knowledge, volume 7 of Minnesota Studies in
the Philosophy of Science. University of Minnesota Press, Minneapolis.
Hale, Bob (1987). Abstract Objects. Blackwell, Oxford.
Hale, Bob and Wright, Crispin (2001). The Reason’s Proper Study. Oxford University Press, Oxford.
Hall, Ned (2006). Philosophy of causation: Blind alleys exposed; promising directions highlighted.
Philosophy Compass, 1, 86–94.
Hamilton, Andrew (2007). Laws of biology, laws of nature: Problems and (dis)solutions. Philosophy
Compass, 2/3, 592–610.
Hanson, William H. (1997). The concept of logical consequence. Philosophical Review, 106, 365–
409.
Harman, Gilbert (1999). Reasoning, Meaning and Mind. Oxford University Press, Oxford.
Haslanger, Sally (1995). Ontology and social construction. Philosophical Topics, 23, 95–125.
Hawthorne, John (2001). Freedom in context. Philosophical Studies, 104, 63–79.
— (2006a). Epistemicism and semantic plasticity. In Zimmerman (2006), pp. 289–322.
— (2006b). Metaphysical Essays. Oxford University Press, Oxford.
— (2006c). Plenitude, convention, and ontology. In Hawthorne (2006b), pp. 53–70.
— (2007). Craziness and metasemantics. Philosophical Review, 116, 427–40.
— (2009). Superficialism in ontology. In Chalmers et al. (2009), pp. 213–30.
Hawthorne, John and Sider, Theodore (2002). Locations. Philosophical Topics, 30, 53–76.
Hawthorne, John (O’Leary-) and Cortens, Andrew (1995). Towards ontological nihilism.
Philosophical Studies, 79, 143–65.
Heller, Mark (1996). Ersatz worlds and ontological disagreement. Acta Analytica, 15, 35–44.
— (1998). Property counterparts in ersatz worlds. Journal of Philosophy, 95, 293–316.
Hempel, Carl (1945). Studies in the logic of confirmation I & II. Mind, 54, 1–26 & 97–I2I.
Hinchliff, Mark (1996). The puzzle of change. In Philosophical Perspectives (ed. James Tomberlin),
volume 10, pp. 119–36. Blackwell, Cambridge, MA.
Hirsch, Eli (1986). Metaphysical necessity and conceptual truth. In Midwest Studies in Philosophy
XI: Studies in Essentialism (eds. Peter French, Theodore E. Uehling, Jr, and Howard K.
Wettstein), pp. 243–56. University of Minnesota Press, Minneapolis, MN.
— (1993). Dividing Reality. Oxford University Press, New York.
— (2002a). Against revisionary ontology. Philosophical Topics, 30, 103–27.
— (2002b). Quantifier variance and realism. Philosophical Issues, 12, 51–73.
— (2005). Physical-object ontology, verbal disputes, and common sense. Philosophy and
Phenomenological Research, 70, 67–97.
— (2007). Ontological arguments: Interpretive charity and quantifier variance. In Sider et al. (2007),
pp. 367–81.
— (2008). Language, ontology, and structure. Noûs, 42, 509–28.
— (2009). Ontology and alternative languages. In Chalmers et al. (2009), pp.231–59.
Hodes, Harold (1984). Logicism and the ontological commitments of arithmetic. Journal of
Philosophy, 81, 123–49.
Hofweber, Thomas (2000). Quantification and non-existent objects. In Everett and Hofweber (2000),
pp. 249–73.
— (2005). A puzzle about ontology. Noûs, 39, 256–83.
— (2007). Innocent statements and their metaphysically loaded counterparts. Philosopher’s Imprint,
7, 1–33.
— (2009). Ambitious, yet modest, metaphysics. In Chalmers et al. (2009), pp. 260–89.
Horgan, Terence and Potrč, Matjaž (2000). Blobjectivism and indirect correspondence. Facta
Philosophica, 2, 249–70.
— (2002). Addressing questions for blobjectivism. Facta Philosophica, 4, 311–22.
Horn, Laurence R. (1989). A Natural History of Negation. University of Chicago Press, Chicago.
Horwich, Paul (1990). Truth. MIT Press, Cambridge, MA.
Hudson, Hud (2006). The Metaphysics of Hyperspace. Oxford University Press, Oxford.
Jackson, Alex (2010). Time and logic. MS.
Jackson, Frank (1998). From Metaphysics to Ethics: A Defence of Conceptual Analysis. Oxford
University Press, Oxford.
Jubien, Michael (2007). Analyzing modality. In Zimmerman (2007), pp. 99–139.
Kaplan, David (1989). Demonstratives. In Themes from Kaplan (eds. Joseph Almog, John Perry, and
Howard Wettstein). Oxford University Press, New York.
Kim, Jaegwon (1982). Psychophysical supervenience. Philosophical Studies, 41, 51–70.
Kim, Jaegwon (1989). The myth of nonreductive materialism. Proceedings of the American
Philosophical Association, 63, 31–47.
— (1992). Multiple realization and the metaphysics of reduction. Philosophy and
PhenomenologicalResearch, 52, 1–26.
King, Jeffrey C. (2005). Anaphora. Available at http://plato.stanford.edu/entries/anaphora/.
Krantz, David H., Luce, R. Duncan, Suppes, Patrick, and Tversky, Amos (1971). Foundations of
Measurement. Academic Press, New York.
Kripke, Saul (1972). Naming and necessity. In Semantics of Natural Language (eds. Donald
Davidson and Gilbert Harman), pp. 253–355, 763–9. D. Reidel, Dordrecht. Revised edition
published in 1980 as Naming and Necessity (Harvard University Press, Cambridge, MA).
— (1975). Outline of a theory of truth. Journal of Philosophy, 72, 690–716.
— (1982). Wittgenstein on Rules and Private Language. Harvard University Press, Cambridge, MA.
Lange, Marc (2005). A counterfactual analysis of the concepts of logical truth and necessity.
Philosophical Studies, 125, 277–303.
Lewis, David (1968). Counterpart theory and quantified modal logic. Journal of Philosophy, 65, 113–
26.
— (1969). Convention: A Philosophical Study. Blackwell, Oxford.
— (1970a). Anselm and actuality. Noûs, 4, 178–88.
— (1970b). How to define theoretical terms. Journal of Philosophy, 67, 427–46.
— (1971). Counterparts of persons and their bodies. Journal of Philosophy, 68, 203–11.
— (1973a). Causation. Journal of Philosophy, 70, 556–67.
— (1973b). Counterfactuals. Blackwell, Oxford.
— (1975). Languages and language. In Gunderson (1975), pp. 3–35.
— (1979). Counterfactual dependence and time’s arrow. Noûs, 13, 455–76.
— (1983a). Extrinsic properties. Philosophical Studies, 44, 197–200.
— (1983b). New work for a theory of universals. Australasian Journal of Philosophy, 61, 343–77.
— (1984). Putnam’s paradox. Australasian Journal of Philosophy, 62, 221–36.
— (1986a). Events. In Lewis (1986c), pp. 241–69.
— (1986b). On the Plurality of Worlds. Blackwell, Oxford.
— (1986c). Philosophical Papers, Volume 2. Oxford University Press, Oxford.
— (1991). Parts of Classes. Blackwell, Oxford.
— (1992). Armstrong on combinatorial possibility. Australasian Journal of Philosophy, 70, 211–24.
— (1994). Humean supervenience debugged. Mind, 103, 473–90.
— (2004). Void and object. In Causation and Counterfactuals (eds. John Collins, Ned Hall, and L.
A. Paul), pp. 277–90. MIT Press, Cambridge, MA.
Lewy, Casimir (1976). Meaning and Modality. Cambridge University Press, Cambridge.
Loewer, Barry (1996). Humean supervenience. Philosophical Topics, 24, 101–27.
Lombard, Lawrence Brian (1999). On the alleged incompatibility of presentism and temporal parts.
Philosophia, 27, 253–60.
Loux, Michael J. and Zimmerman, Dean W. (eds.) (2003). Oxford Handbook of Metaphysics. Oxford
University Press, Oxford.
Lycan, William (2001). Real Conditionals. Oxford University Press, Oxford.
MacFarlane, John (2005). Logical constants. Available at http://plato.stanford.edu/entries/logical-
constants/.
Mackie, J. L. (1977). Ethics: Inventing Right and Wrong. Penguin, London.
Malcolm, Norman (1940). Are necessary propositions really verbal? Mind, 49, 189–203.
Markosian, Ned (1993). How fast does time pass? Philosophy and Phenomenological Research, 53,
829–44.
— (1998). Simples. Australasian Journal of Philosophy, 76, 213–26.
— (2004a). Simples, stuff, and simple people. The Monist, 87, 405–28.
— (2004b). Soc it to me? Reply to McDaniel on maxcon simples. Australasian Journal of
Philosophy, 82, 332–40.
Marshall, Dan (2009). Can ‘intrinsic’ be defined using only broadly logical notions? Philosophy and
Phenomenological Research, 78, 646–72.
Maudlin, Tim (1993). Buckets of water and waves of space: Why spacetime is probably a substance.
Philosophy of Science, 60, 183–203.
— (1996). On the unification of physics. Journal of Philosophy, 93, 129–44.
— (2007a). The Metaphysics Within Physics. Oxford University Press, New York.
— (2007b). A modest proposal concerning laws, counterfactuals, and explanations. In Maudlin
(2007a), pp. 5–49.
— (2007c). On the passing of time. In Maudlin (2007a), pp. 104–42.
McDaniel, Kris (2003). Against maxcon simples. Australasian Journal of Philosophy, 81, 265–75.
— (2007a). Brutal simples. In Zimmerman (2007), pp. 233–66.
— (2007b). Extended simples. Philosophical Studies, 133, 131–41.
— (2009). Ways of being. In Chalmers et al. (2009), pp. 290–319.
— (2011a). Being and almost nothingness. Noûs, Forthcoming.
— (2011b). A return to the analogy of being. Philosophy and Phenomenological Research,
Forthcoming.
McGee, Vann (2005). Two conceptions of truth? Philosophical Studies, 124, 71–104.
McGinn, Colin (1981). Modal reality. In Reduction, Time and Reality (ed. Richard Healey), pp. 143–
88. Cambridge University Press, Cambridge.
— (2000). Logical Properties. Oxford University Press, Oxford.
McLaughlin, Brian and Bennett, Karen (2005). Supervenience. Available at
http://plato.stanford.edu/entries/supervenience/.
McTaggart, J. M. E. (1908). The unreality of time. Mind, 17, 457–74.
Melia, Joseph (1995). On what there’s not. Analysis, 55, 223–9.
— (2000). Weaseling away the indispensability argument. Mind, 109, 455–79.
Merricks, Trenton (1994). Endurance and indiscernibility. Journal of Philosophy, 91, 165–84.
— (2006). Good-bye growing block. In Zimmerman (2006), pp. 103–10.
— (2007). Truth and Ontology. Oxford University Press, Oxford.
Millikan, Ruth Garrett (1989). Biosemantics. Journal of Philosophy, 86, 281–97.
Montague, Richard (1973). The proper treatment of quantification in ordinary English. In
 Approaches to Natural Language (eds. Jaakko Hintikka, J. M. E. Moravcsik, and Patrick
Suppes), pp. 221–42. Reidel, Dordrecht, Boston.
Mortensen, Chris (1989). Anything is possible. Erkenntnis, 30, 319–37.
Mundy, Brent (1987). The metaphysics of quantity. Philosophical Studies, 51, 29–54.
Nerlich, Graham (1976). The Shape of Space. Cambridge University Press, Cambridge.
— (1982). The present state of realism. Philosophical Quarterly, 32, 272–9. Review of Richard
Healey (ed.), Reduction, Time and Reality: Studies in the Philosophy of the Natural Sciences.
Nolan, Daniel (2011). The extent of metaphysical necessity. MS.
North, Jill (2009). The ‘structure’ of physics: A case study. Journal of Philosophy, 106, 57–88.
— (2010). Structure in classical mechanics. Available at
http://courses.cit.cornell.edu/north/Structure_CM_Rev_newer.pdf.
Oliver, Alex (1996). The metaphysics of properties. Mind, 105, 1–80.
Pap, Arthur (1958). Semantics and Necessary Truth. Yale University Press, New Haven.
Parsons, Josh (2003). Entension, or how it could happen that an object is wholly located in each of
many places. MS. Available at
http://www.otago.ac.nz/philosophy/Staff/JoshParsons/papers/entension2.pdf.
Parsons, Terence (1980). Nonexistent Objects. Yale University Press, New Haven.
Paul, L. A. (2009). Counterfactual theories of causation. In Oxford Handbook of Causation (eds.
Helen Beebee, Christopher Hitchock, and Peter Menzies), pp. 158–84. Oxford University Press,
Oxford.
— (2010). A new role for experimental work in metaphysics. Review of Philosophy and Psychology,
1, 461–76.
Peacocke, Christopher (1997). Metaphysical necessity: Understanding, truth and epistemology. Mind,
pp. 521–74.
— (1999). Being Known. Clarendon Press, Oxford.
Penrose, Roger (2005). The Road to Reality. Alfred A. Knopf, New York.
Pietroski, Paul (2003). The character of natural language semantics. In Epistemology of Language
(ed. Alex Barber), pp. 217–56. Oxford University Press, Oxford.
Plantinga, Alvin (1974). The Nature of Necessity. Oxford University Press, Oxford.
— (1976). Actualism and possible worlds. Theoria, 42, 139–60.
Poincaré, Henri (1952). Science and Hypothesis. Dover, New York. Republication of the first English
translation, published by Walter Scott Publishing, London, 1905.
Priest, Graham (1987). In Contradiction: A Study of the Transconsistent. Martinus Nijhoff,
Dordrecht.
Prior, A. N. (1960). The runabout inference ticket. Analysis, 21, 38–9.
— (1967). Past, Present, and Future. Oxford University Press, Oxford.
— (1968a). Papers on Time and Tense. Oxford University Press, Oxford.
— (1968b). Quasi-propositions and quasi-individuals. In Prior (1968a), pp. 135–44.
— (1970). The notion of the present. Studium Generale, 23, 245–8.
— (1971). Objects of Thought. Oxford University Press, Oxford.
— (1976). Thank goodness that’s over. In Papers in Logic and Ethics, pp. 78–84. Duckworth,
London.
— (1996). Some free thinking about time. In Logic and Reality: Essays on the Legacy of Arthur
Prior (ed. Jack Copeland), pp. 47–51. Clarendon Press, Oxford.
Prior, A. N. and Fine, Kit (1977). Worlds, Times, and Selves. Duckworth, London.
Putnam, Hilary (1975a). Mathematics, Matter and Method. Cambridge University Press, Cambridge.
— (1975b). The meaning of meaning. In Gunderson (1975), pp. 131–93.
— (1975c). On properties. In Putnam (1975a), pp. 305–22.
— (1975d). What is mathematical truth? In Putnam (1975a), pp. 60–78.
— (1978). Meaning and the Moral Sciences. Routledge and Kegan Paul, Boston.
— (1980). Models and reality. Journal of Symbolic Logic, 45, 464–82.
— (1981). Reason, Truth and History. Cambridge University Press, Cambridge.
— (1987). Truth and convention: On Davidson’s refutation of conceptual relativism. Dialectica, 41,
41–67.
Quine, W. V. O. (1936). Truth by convention. In Philosophical Essays for A. N. Whitehead (ed. O. H.
Lee), pp. 90–124. Longmans, New York.
— (1948). On what there is. Review of Metaphysics, 2, 21–38.
— (1951a). Ontology and ideology. Philosophical Studies, 2, 11–15.
— (1951b). Two dogmas of empiricism. Philosophical Review, 60, 20–43.
— (1953). Notes on the theory of reference. In From a Logical Point of View, pp. 130–8. Harvard
University Press, Cambridge, MA.
— (1960a). Carnap and logical truth. Synthese, 12, 350–74.
— (1960b). Variables explained away. Proceedings of the American Philosophy Society, 104, 343–7.
— (1960c). Word and Object. MIT Press, Cambridge, MA.
Quine, W. V. O. (1966). The Ways of Paradox. Random House, New York.
— (1969). Natural kinds. In Ontological Relativity and other Essays, pp. 114–38. Columbia
University Press, New York.
— (1976a). On multiplying entities. In The Ways of Paradox, pp. 259–64. Random House, New
York. Revised and enlarged edition.
— (1976b). Worlds away. Journal of Philosophy, 73, 859–63.
Quinton, Anthony (1958). Properties and classes. Proceedings of the Aristotelian Society, 58, 33–58.
Railton, Peter (1986). Moral realism. Philosophical Review, 95, 163–207.
Rayo, Agustín and Uzquiano, Gabriel (1999). Toward a theory of second-order consequence. Notre
Dame Journal of Formal Logic, 40, 315–25.
Rayo, Agustín and Yablo, Stephen (2001). Nominalism through de-nominalization. Noûs, 35, 74–92.
Reichenbach, Hans (1958). The Philosophy of Space and Time. Dover, New York.
Restall, Greg (1996). Truthmakers, entailment and necessity. Australasian Journal of Philosophy, 74,
331–40.
Rosen, Gideon (1990). Modal fictionalism. Mind, 99, 327–54.
— (1994). Objectivity and modern idealism: What is the question? In Philosophy In Mind: The Place
of Philosophy in the Study of Mind (eds. Michaelis Michael and John O’Leary-Hawthorne), pp.
277–319. Kluwer, Dordrecht.
— (2010). Metaphysical dependence: Grounding and reduction. In Modality: Metaphysics, Logic,
and Epistemology (eds. Bob Hale and Aviv Hoffmann), pp. 109–36. Oxford University Press,
Oxford.
Routley, Richard (1980). Exploring Meinong’s Jungle and Beyond. An Investigation ofNoneism and
the Theory of Items. Philosophy Department Monographs, Research School of Social Sciences,
Australian National University, Canberra.
Russell, Bertrand (1903). The Principles of Mathematics. Routledge, London.
— (1915). On the experience of time. The Monist, 25, 212–33.
— (1919). Introduction to Mathematical Philosophy. Routledge, London.
Russell, Jeffrey Sanford (2010). The structure of gunk: Adventures in the ontology of space. In
Zimmerman (2010), pp. 248–74.
Sagan, Carl (1980). Cosmos. Random House, New York.
Savitt, Steven (2004). Presentism and eternalism in perspective. Available at http://philsci-
archive.pitt.edu/archive/00001788/01/PEP.pdf.
Schaffer, Jonathan (2007). From nihilism to monism. Australasian Journal of Philosophy, 85, 175–
91.
— (2009a). On what grounds what. In Chalmers et al. (2009), pp. 347–83.
— (2009b). Spacetime the one substance. Philosophical Studies, 145, 131–48.
— (2010a). The internal relatedness of all things. Mind, 119, 341–76.
— (2010b). The least discerning and most promiscuous truthmaker. Philosophical Quarterly, 60,
307–24.
— (2010c). Monism: The priority of the whole. Philosophical Review, 119, 31–76.
Schiffer, Stephen (2003). The Things We Mean. Clarendon Press, Oxford.
Schnieder, Benjamin (2011). A logic for ‘because’. Review of Symbolic Logic. Forthcoming.
Shalkowski, Scott (1994). The ontological ground of the alethic modality. Philosophical Review, 103,
669–88.
Sidelle, Alan (1989). Necessity, Essence and Individuation. Cornell University Press, Ithaca, NY.
— (2002). Is there a true metaphysics of material objects? Philosophical Issues, 12, 118–45.
— (2007). The method of verbal dispute. Philosophical Topics, 35, 83–114.
— (2009). Conventionalism and the contingency of conventions. Noûs, 43, 224–41.
— (2010). Modality and objects. Philosophical Quarterly, 60, 109–25.
Sider, Theodore (1993a). Naturalness, intrinsicality and duplication. Ph.D. thesis, University of
Massachusetts.
— (1993b). Van Inwagen and the possibility of gunk. Analysis, 53, 285–9.
— (1995). Sparseness, immanence and naturalness. Noûs, 29, 360–77.
— (1996a). All the world’s a stage. Australasian Journal of Philosophy, 74, 433–53.
— (1996b). Intrinsic properties. Philosophical Studies, 83, 1–27.
— (1996c). Naturalness and arbitrariness. Philosophical Studies, 81, 283–301.
— (2001a). Criteria of personal identity and the limits of conceptual analysis. Philosophical
Perspectives, 15, 189–209.
— (2001b). Four-Dimensionalism. Clarendon Press, Oxford.
— (2002). The ersatz pluriverse. Journal of Philosophy, 99, 279–315.
— (2003). Reductive theories of modality. In Loux and Zimmerman (2003), pp. 180–208.
— (2004). Précis of Four-Dimensionalism, and replies to critics. Philosophy and
PhenomenologicalResearch, 68, 642–7, 674–87.
— (2005). Another look at Armstrong’s combinatorialism. Noûs, 39, 679–95.
— (2006). Beyond the Humphrey objection. Available at
http://tedsider.org/papers/counterpart_theory.pdf.
— (2007a). Neo-Fregeanism and quantifier variance. Aristotelian Society, Supplementary Volume,
81, 201–32.
— (2007b). Parthood. Philosophical Review, 116, 51–91.
— (2008a). Monism and statespace structure. In Being: Developments in Contemporary Metaphysics
(ed. Robin Le Poidevin), pp. 129–50. Cambridge University Press, Cambridge.
— (2008b). Yet another paper on the supervenience argument against coincident entities. Philosophy
and Phenomenological Research, 77, 613–24.
— (2009). Ontological realism. In Chalmers et al. (2009), pp. 384–423.
— (2011). Against parthood. MS.
Sider, Theodore, Hawthorne, John, and Zimmerman, Dean W. (eds.) (2007). Contemporary Debates
 in Metaphysics. Blackwell, Oxford.
Siegel, Susanna (2006). Which properties are represented in perception? In Perceptual Experience
(eds. Tamar Szabó Gendler and John Hawthorne), pp. 481–503. Oxford University Press,
Oxford.
Skow, Bradford (2010). Extrinsic temporal metrics. Oxford Studies in Metaphysics, 5, 179–202.
Smart, J. J. C. (1963). Philosophy and Scientific Realism. Routledge & Kegan Paul, London.
Soames, Scott (1992). Truth, meaning, and understanding. Philosophical Studies, 65, 17–35.
Sober, Elliot (1994). A Bayesian primer on the grue problem. In Grue! The New Riddle of Induction
(ed. Douglas Stalker), pp. 225–40. Open Court, Chicago.
Sorensen, Roy (2001). Vagueness and Contradiction. Clarendon Press, Oxford.
Sosa, Ernest (1999). Existential relativity. In Midwest Studies in Philosophy XXIII: New Directions in
Philosophy (eds. Peter French and Howard K. Wettstein), pp. 132–43. Blackwell, Oxford.
Sperber, Dan and Wilson, Deirdre (1986). Loose talk. Proceedings of the Aristotelian Society, 86,
153–71.
Stalnaker, Robert (1968). A theory of conditionals. In Studies in Logical Theory: American
Philosophical Quarterly Monograph Series, No. 2. Blackwell, Oxford.
— (1976). Possible worlds. Noûs, 10, 65–75.
— (1977). Complex predicates. The Monist, 60, 327–39.
Stanley, Jason (2000). Context and logical form. Linguistics and Philosophy, 23, 391–434.
Strawson, Peter F. (1952). Introduction to Logical Theory. Methuen, London.
— (1959). Individuals: An Essay in Descriptive Metaphysics. Routledge, London.
Street, Sharon (2006). A Darwinian dilemma for realist theories of value. Philosophical Studies, 127,
109–66.
Strevens, Michael (2006). Scientific explanation. In Encyclopedia of Philosophy. Macmillan
Reference USA.
Tappenden, Jamie (1993). Analytic truth—it’s worse (or perhaps better) than you thought.
Philosophical Topics, 21, 233–61.
Tarski, Alfred (1959). What is elementary geometry? In The Axiomatic Method, with Special
Reference to Geometry and Physics (eds. Leon Henkin, Patrick Suppes, and Alfred Tarski), pp.
16–29. North-Holland, Amsterdam.
Tarski, Alfred and Givant, Steven (1999). Tarski’s system of geometry. Bulletin of Symbolic Logic, 5,
175–214.
Thomasson, Amie L. (2003). Fictional characters and literary practices. British Journal of Aesthetics,
43, 138–57.
— (2005). The ontology of art and knowledge in aesthetics. The Journal of Aesthetics and Art
Criticism, 63, 221–9.
— (2006). Debates about the ontology of art: What are we doing here? Philosophy Compass, 1, 245–
55.
— (2007). Ordinary Objects. Oxford University Press, New York.
— (2009). Answerable and unanswerable questions. In Chalmers et al. (2009), pp. 444–71.
Tooley, Michael (1987). Causation: A Realist Approach. Clarendon Press, Oxford.
Turner, Jason (2008). Ontology, quantification, and fundamentality. Ph.D. thesis, Rutgers University.
— (2010a). Ontological nihilism. In Oxford Studies in Metaphysics (eds. Karen Bennett and Dean W.
Zimmerman), volume 6, pp. 3–54. Oxford University Press, Oxford.
— (2010b). Ontological pluralism. Journal of Philosophy, 107, 5–34.
Unger, Peter (1979). I do not exist. In Perception and Identity: Essays Presented to A. J. Ayer with
His Replies to Them (ed. G. F. Macdonald), pp. 235–51. Macmillan, New York.
van Fraassen, Bas (1989). Laws and Symmetry. Clarendon Press, Oxford.
van Inwagen, Peter (1990). Material Beings. Cornell University Press, Ithaca, NY.
— (1998). Modal epistemology. Philosophical Studies, 92, 67–84.
— (2010). Changing the past. In Zimmerman (2010), pp. 3–28.
von Fintel, Kai (2001). Counterfactuals in a dynamic context. In Ken Hale: A Life in Language, pp.
123–52. MIT Press, Cambridge, MA.
Wasserman, Ryan (2005). The problem of change. Ph.D. thesis, Rutgers University.
Weatherson, Brian (2003). What good are counterexamples? Philosophical Studies, 115, 1–31.
Williams, Donald C. (1951). The myth of passage. Journal of Philosophy, 48, 457–72.
Williams, J. Robert G. (2005). The inscrutability of reference. Ph.D. thesis, University of St
Andrews.
— (2007). Eligibility and inscrutability. Philosophical Review, 116, 361–99.
Williamson, Timothy (1985). Converse relations. Philosophical Review, 94, 249–62.
— (1988). Equivocation and existence. Proceedings of the Aristotelian Society, 88, 109–27.
— (1994). Vagueness. Routledge, London.
— (1998). Bare possibilia. Erkenntnis, 48, 257–73.
— (2002). Necessary existents. In Logic, Thought and Language (ed. A. O’Hear), pp. 233–51.
Cambridge University Press, Cambridge.
— (2003). Everything. Philosophical Perspectives, 17, 415–65.
— (2007). The Philosophy of Philosophy. Blackwell, Malden, MA.
Wilson, Deirdre and Sperber, Dan (2004). Relevance theory. In Handbook of Pragmatics (eds.
Laurence R. Horn and Gregory Ward), pp. 607–32. Blackwell, Malden, MA.
Wittgenstein, Ludwig (1958). Philosophical Investigations. Blackwell & Mott, Oxford, 3rd edition.
Translated by G. E. M. Anscombe.
Wright, Crispin (1983). Frege’s Conception of Numbers as Objects. Aberdeen University Press,
Aberdeen.
— (1992). Truth and Objectivity. Harvard University Press, Cambridge, MA.
Yablo, Stephen (1992). Review of Alan Sidelle, Necessity, Essence and Individuation. Philosophical
Review, 101, 878–81.
— (1998). Does ontology rest on a mistake? Aristotelian Society, Supplementary Volume, 72, 229–
61.
— (2000). A paradox of existence. In Everett and Hofweber (2000), pp. 275–312.
— (2001). Go figure: A path through fictionalism. In Midwest Studies in Philosophy XXV: Figurative
Language (eds. Peter French and Howard K. Wettstein), pp. 72–102. Blackwell, Oxford.
Zimmerman, Dean W. (1998). Temporary intrinsics and presentism. In Metaphysics: The Big
Questions (eds. Dean W. Zimmerman and Peter van Inwagen), pp. 206–19. Blackwell,
Cambridge, MA.
— (2005). The A-theory of time, the B-theory of time, and ‘taking tense seriously’. Dialectica, 59,
401–57.
Zimmerman, Dean W. (ed.) (2006). Oxford Studies in Metaphysics. Oxford University Press, Oxford.
— (2007). Oxford Studies in Metaphysics, volume 3. Oxford University Press, Oxford.
— (2010). Oxford Studies in Metaphysics, volume 5. Oxford University Press, Oxford.
Index
A-theory, see passage (time)
abstract entities, 85–6, 90–1
aesthetics, 57–60, 62
alien meanings, 50, 70, 71
analyticity, 191–5
and substantivity, 193–4
and truth, 192–3
Ayer’s conception of, 98
defined, 193
epistemology of, 194–5
Quine’s critique, 191–2

B-theory, see spatializing (time)

Barcan schema, temporal, 257
Bayesianism, 35–8, 64
best-system theory of laws, see laws of nature, Lewis’s theory
block universe, see spatializing (time)

candidates, 46, 49, 71

right sort, 48–9
versus precisifications, 49
canonical names, see “the proposition (fact, state of affairs) that”
causal metasemantics, 33–5
causal nihilism, 151–3, 199
causation, 15, 16, 34, 50, 60, 73, 75–6, 151–3
change (genuine), 246, 247, 249–50, 260–1
charity to use, 31, 32
cheaters, 154–6
combinatorial principle for determinacy, 150
combinatorial principle for fundamental truth, 148
completeness, 105–6, 115–16
composite material objects, 149, 169, 173–6
conceptual analysis, see metaphysics, and ordinary truth
conceptual role, see understanding connection (between fundamental and
nonfundamental), 109–15, 144
contingency (genuine), 247–9
continuum hypothesis, 150–1
conventionalist theory of necessity, see modality, conventionalism
conventionality, 54–7
candidate-selection, 54–7
content, 54
symbol, 54
counterpart theory, 287–8
cross-time relations, 262

definitional constraints, 191–3

definitional sentence, see definitional constraints
deflationism about metaphysics, 67–72, 180
deflationism about modality, 204
deflationism about truth, see disquotational truth
dependence, see ground
descriptivist metasemantics, 23, 30, 31, see also simple charity-based
descriptivism
determinacy, 137, 150–1
determinates and determinables, 278–80
dialetheism, 228
disquotational truth, 28, 45, 113, 126, 232, 251, see also facts, disquotational

eligibility, 31
entity-grounding, see ground, Schaffer’s approach
entrenching, 159–61, 163
epistemic value, 61–6
objectivity of, 38, 53
essence, 267
excluded middle, 227–8, 234
explanation, 23, 28, 30, 64, 131
ultimate, 159–60, 163
expressivism, 45, 57, 113–14, 125–6
extended simples, 79–82

fact-level metaphysics, 185

facts
disquotational, 126, 250–1
robustly conceived, see saturation
flow, see passage (time)
free will and determinism, 75
fundamental entities, 164–5, 170–1
fundamental language, 8, 152, 254
fundamental notion, 106, 144
fundamental truth (fundamental fact), 105, 147–51
fundamentality, 105–65

geometric conventionalism, 39–43

goal of inquiry, see epistemic value
governance versus classification (modality), 270–1
grasp, see understanding ground, 106, 134–5, 205
Fine, 124–7, 142–7, 162
nonfundamental, 124–7
Schaffer, 161–4, 199–200
growing block, see passage (time), growing block
grue, 34
gunk, 133–5

holes, 68, 71, 72, 166, 167

ideological commitment, see indispensability

ideology, vii–viii, 10, 12–15, 83, 86, 95, 97, 141, 154–7, 170, 201, 230,
145–6
nonpsychologically conceived, 12
uniqueness of, 13, 217–22
in-virtue-of, 105–7, 109–11, 134–5
indifference principle, 37
indispensability, 13, 166–7, 169–70
of quantifiers, 182–8
infinite descent, 115, 133–6, 145–7
innocent higher-order quantification, see quantification, higher-order
intuitionism, 227

K3, 233–8
knee-jerk realism, 18–20

laws of metaphysics, 22, 145, 274–8

laws of nature, 15, 21–3, 60, 64, 156
Lewis’s theory, 17, 21, 60, 64
Lewis’s theory augmented, 22, 223–4
levels, 129, 132, 133
liar paradox, 126, 231–8
linguistic theory of necessity, see modality, conventionalism
location, 79
logic
classical, 137, 231, 234–5
in fundamental and nonfundamental languages, 231–8
laws of, 22, 223–5
metaphysics of, 97–8, 216–22, 276
nonclassical, 225–8
logical consequence, 222–30, 272–3
logical conventionalism, 97–104
logical pluralism, 225–8
logical truth, 222–30, 272–3

martini (Bennett example), 44

maximalism, 200–1
Meinong, 206–7
Melianism, 138–9, 145
mereological nihilism, 120, 122, 150, 292–4
metalogic
nonfundamental, 214–15, 222–8
versus logic, 216, 227
metametaphysics, 6–8, 67–84, 86–7
not methodological, 82–4
metaontology, 7, 167–215
metaphysical causation, 144–5, 205
metaphysical deflationism, see deflationism about metaphysics
metaphysical semantics, 112–24
expressivist form, 113–14, 125–6
truth-conditional form, 113
versus linguistic semantics, 122–3
metaphysical truth-conditions, 114–24
needn’t preserve logical form, 120–1, 235–7
“toy”, 117–18, 132, 285–6, 294–6
metaphysics
ambitions of, 140
and apriority, 136
and ordinary truth, 116–18, 154
characterized, 1
epistemology of, 11–12, 82, 83, 96–7, 166–7, 169, see also
indispensability, ontological commitment
relation to special sciences, 123–4
relation to study of language, 122–3
metaphysics room, 74–7, 171–3
metasemantics, 23
minimality, see nonredundancy
modal axioms and rules, 271
modalism, 107–8, 266–7, 286
modality, 4, 16, 80–1, 95, 105, 107–8, 199, 247–9, 266–91
arguments from possibility, 136, 277–8
conventionalism, 268–70
de re, 286–8
Humean theory, 80–1, 136, 204, 269–91
modal realism, 201, 247–9, 268
model theory, 228–30
monism, 79, 159–60, 169
morality, 56, 59, 62
mountain (Hawthorne example), 44
moving spotlight, see passage (time), moving spotlight

natural kind axioms (modality), 283

natural properties, 4, 85–8, 128–31, 138
necessary a posteriori, 282–3
negative facts, 158
neoFregeanism, 190
nominalism and metalogic, 228–30
nonfactualism, see expressivism
nonredundancy, 218–19

objecthood, 205–6
objectivity, 60–1, 65–6, 72
Ontologese, 171–3, 197
ontological commitment, 166–7, 169–70, 201–3, 209, 211–12
ontological deflationism, 7, 81, 167–8, 173–201
ontological free lunch (Armstrong), 198–9
ontological nihilism, 199
ontological pluralism, 207
ontological realism, 88–90, 95–6, 168–73, 188, 199–201
and presentism, 242–4
ontological semi-realism, 149–50
ontologism, see structure, ontic approach to
ontology, 79, 166–215
easy, 189–91, 195–9

parsimony, see simplicity

passage (time), 239, 246–65
growing block, 263–4
moving spotlight, 259–62
Williamsonian, 257–9
path-dependent distance, 39
perception, 78
permuted interpretations, 19, 25–6, 65
personal identity, 72–3
perspective, 240–1, 249–57, 262
physical geometry, 38–43, 65
plural quantification, see quantification, higher-order
predicate functors, 184–5, 207
predication, 157–8
presentism, 154, 239–46, 257
pretense (abstracta), 229–30
projectibility, 35
projectivism, 57–60
purity, 106–11, 115–16, 124, 129, 137, 143–4, 161, 164, 186

quantification
carves at the joints, see ontological realism
higher-order, 208–15, 229
metaphorical, 197
nonfundamental, 120–1, 127, 171–2, 196–8, 293–4
temporally relative, 243–4
quantifier variance, 89, 151, 175–88, 201
quantity, 218, 240, 279–80

reality (Fine), 142–3, 147–53, 199

reduction, see metaphysics, and ordinary truth
reference magnetism, 23–33, 64, 74
as metametasemantics, 30–3
origin of the term, 27
Williams’s conception of, 27–33
rejection, 236

saturation, 249–57
schmexists, vii, 89, 148
semantic candidates, see candidates
semantic goals, 50–7
semantic pressure, 75, 76
semantics is nonfundamental, 90–1, 235
semantics versus pragmatics, 77–8
similarity, 1, 63, 87–90
of facts, 88–90
simple charity-based descriptivism, 31–3, 74
simplicity, 10–14, 86, 141, 154–6, 169, 201, 209–10, 214–15
and laws of nature, 22
skepticism about the present moment, 261–2
smuggling, 157–9, 161, see also “the proposition (fact, state of affairs) that”
space, 79
spacetime structure, see physical geometry
sparse entities, see structure, ontic approach to
spatializing (time), 239–40, 265, 292
special sciences, 14, 28, 48, 129
laws, 22
states of affairs, 153, 163–4
structural truth, 147, see also fundamental truth (fundamental fact)
structure
absolute versus comparative, see structure, degrees of
and fundamentality, 5–6
argument for positing, 10–11, 132
degrees of, 5, 28, 78, 128–33, 136
determinate, 137, 235
epistemology of, 11–15
generality of, 8, 85–7, 216
introduced, 1–2
is explanatory, 132, 139–41
is structural, 132, 137–41
not a feature of abstract or linguistic entities, 90–2
not reducible, 15–18
objectivity of, 18–20, 38, 65–6
of grammatical categories, 203, 253–7
ontic approach to, 94–6
querying complex expressions, 92–4
regimentation, 91–4, 141, 256
subpropositional, 128, 147–53, 203, 220
understanding of, 9–10
stuff, 183–4
subjectivity, 57–60, 65
substantivity, 6–7, 65, 70, 77–9, 86–7
characterized, 44–54
dimensions of, 38, 52
is metalinguistic, 47, 51
joint-carving sufficient but not necessary for, 46, 48, 71
metaphysical versus conceptual, 50, 73–4
subtleties, 47–54
supersubstantivalism, 292
synthetic necessities, 275–7
tense operators, 154, 240–2
metrical, 240
“the proposition (fact, state of affairs) that”, 86, 111, 157–9
theoretical terms, 32–3, 48, 74
tonk, 102–3, 195
totality facts, 158
Tractarian ontological realism, 203–6
trumping (reference magnetism), 32
truth by convention, 97–104, 191, 194, 268–70, 276
truth supervenes on being, 155–6
truthmaker free-for-all, 160–1
truthmaking, 105, 153–61

understanding, 9–10
universals, 4, 85–6, 94, 96, 157
Ural mountains (Hawthorne example), 32, 48
use, see charity to use

vagueness, 234–8
worldly, 137
verbal questions, see substantivity

well-foundedness (of ground, etc.), see infinite descent

worldview, 292