Terence Parsons Nonexistent Objects

Nonexistent
Objects
Terence Parsons
Yale Unioersity Press

New Haven-and London
Copyright © r!}8o by Tale Universi~.
All rights reserved. This book may not be
reproduad, in whole or in part, in 01!1 form
(beyond that copying permitted by Sections 107
and roB of the U.S. Copyright Law and except by
reviewers for thepublicpress)~ without written
permission from tlu publishers.
Designed by James J4 ]ohnson

andset in Monotype Baskeroille gpe.
Printedin the United States cif America by
Mu"ay Printing Company;, Westford, Mass.
Library of Congress Cataloging in Publication Data
Parsons, Terence.
Nonexistent objects.
Bibliography: p .
Includes index.
L Nonexistent objects (Philosophy) I. Title.
BD2Ig . PS7 III 79-21682
ISBN 0-300-02404-5
10 9 8 7 6 5 4 3 2 I
TO MY PAR.ENTS
without whom I might have been one

Contents
Preface Xl
Introduction I
I. HISTORICAL DEVELOPMENT 2
2. THE CURRENT SCENE 5
3. METHODOLOGICAL PRELIMINARIES 9
I: Initial Exposition
I. Initial Sketch of the Theory 17

I. OBJECTS 17
2. ~XTRANUCLEAR PREDICATES 22
3. IDENTITY 27
2. Meinong and Motivation 30
1. MEINONO'S MOTIVATION 30
2. EVIDENCE FOR THE THEORY 32
3. RUSSELL'S ARGUMENTS AGAINST MEINONG 38
4· INTENTION ALITY 44
3. A Sketch of a Theory of Fictional Objects 49
I. LITERARY THEORY 49
2. IMMIGRANT AND NATIVE OBJECTS 51
3. THE THEORY 52
4· SURROGATE OBJECTS 57
5· RELATIONAL PROPERTIES 59
vii
viii CONTENTS
I I: Formal Development
4. The Language {f} 63

I. THE SYMBOLISM OF (!) 64
2. RULES AND AXIOMS 70
3. THE SEMANTICS OF (f) 78
4. THEOREMHOOD EQ.UALS VALIDITY 84
5. AN ORTHODOX THEORY fL' 8S
6. ~IODELING (J) WITHIN !fJ 88
\ 7. USES OF THE PICTURING 92
5.. Further Developments of the Language (f) 98
I. MODALITIES 98
2. ABSTRACTS AND COMPLEX PROPERTIES 103
3. PROPOSITIONS AND PROPOSITIONAL ATTITUDES 108
4- SINGULAR TERMS I II
5. FORMALIZATION 12 3
APPENDIX: A SEMANTICS FOR A FRAGMENT OF

ENGLISH 134
I I I: Applications and Discussion
6. Summary of the Formal Development and More

on Nuclear and Extranuclear 155
I. SUMMARY OF CHAPTER 4 155
2 .. COMPARISON WITH ORTHODOX SYSTEMS 160
3. SUMMARY OF CHAPTER 5 161
4. MORE ON NUCLEAR AND EXTRANUCLEAR 166
5. MODIFYING THE DISTINCTION 17°
7· Fictional Objects, Dream Objects, and Others 175
I. "IN THE STORY. w .. » 175
2. WHAT ARE THEY LIKE? 182
3· DIGRESSION CONCERNING THE AUTHOR 187

4· UNCLARITIES IN THE THEORY 188
5- INDISCERNIBLES: CROWDS AND SIBLINGS 19°
6. RELATIONS TO UNREAL OBJECTS 194
76 EXTRANUCLEAR PREDICATIONS; STORIES
WITHIN STORIES 197
CONTENTS ix
8. ALTERNATIVE THEORIES 202

9- FANTASY, MYTH, AND LEGEND 206
10. OBJECTS IN DREAMS 207
I I. VISUAL OBJECTS 210
8. Traditional Issues from the Present Perspective 2 12

I. ONTOLOGICAL ARGUMENTS; WHY SHOULD
ANYONE CARE WHETHER GOD EXISTS? 212
2. EPISTEMOLOGY: MIGHT I BE A NONEXISTENT
OBJECT? 217
3. LEIBNIZ'S MONADS AND POSSIBLE WORLDS 219
4. ESSENTIALISM 224
5. PLATONIC FORMS AND COMPLEX PROPERTIES 227
6. THEORETICAL ENTITIES IN SCIENCE 228
9· Global Issues 233

I. ARE THERE TOO MANY OBJECTS? 233
2. ARE THERE TOO FEW OBJECTS? 240
3. SOME CONCLUDING REMARKS 245
Selected Bibliography 247
Index 255
Preface
Alexius Meinong was an Austrian philosopher and psychologist

who did most of his work around the turn of the century. He
believed that he had discovered a whole realm of objects which
had not been studied previously by philosophers or by scientists.
These are the objects that don't exist. As examples he cited the
golden mountain-it doesn't exist, and it has certainly not been
extensively scrutinized by serious scholars-and the round
square, which not only does not exist but also has the interesting
property of being impossible.
I first became acquainted with Meinong's views when I
was slated to teach a course on metaphysics at the University of
Illinois at Chicago Circle during the fall of 1966.. My intent was
to impart to- the class the wisdom I had recently gained as a
graduate student, and a nice vehicle for this was to be W. v.
Quine's essay, "On What There Is." I feared, however, that the
students would find the discussion there too abstract unless they
were already familiar with the views that Quine was opposing in
the first few pages of his essay: they would appreciate Quine's
applications ofOccam's Razor more if they were already familiar
with the engaging but obscure theory against which it was being
wielded. I had been told that Quine's character "Wyman" was
based on the historical Meinong, and so I selected Meinong's
"The Theory of Objects" as an initial reading.
Meinong's views had a profound impact on me. At first, I
was convinced by the criticisms leveled against him by Russell
and Quine; in fact, I first thought of these "refutations" as
xi
xii PREFACE
constituting one of the clearest examples of philosophical pro-

gress that we have. Clear progress is rare in philosophy, and I
was pleased to have an example to cite. But as I thought about
it more, I became increasingly dissatisfied. I found the criticisms
intellectually compelling, but I kept thinking that there was
something very true in Meinong's views that was being missed.
And eventually I began working out a similar kind of theory, a
theory that I thought preserved most of Meinong's radical and
exciting ontological views-such as the belief in nonexistent
objects) even in impossible objects-but a theory that is immune
to the standard criticisms. That theory was developed gradually
in a series of papers and talks, and the present book represents
its most recent form.
This is not an exposition of Meinong. Although I have
gained much in inspiration and ideas from reading his work,
I am not well enough acquainted with his voluminous writings
to qualify as an expert on them, From time to time I will discuss
gross similarities and differences between Meinong's views and
those presented here, but for the most part I will simply treat
Meinong as an ally who had many of the ideas first.
I was most influenced in the early stages of my work by a
manuscript by Richard Routley entitled "Exploring Meinong's
Jungle," a work that is, I suspect, closer to Meinong's views than
is my own. I have not tried to draw detailed comparisons be-
tween Routley's work and my own, and the same goes for many
other contemporary writers on related topics.. Such comparisons
would have required too much work and would probably have
been out of date by the time this is published.. Likewise, I have
not tried to catalogue the ways in which the present work
coheres with or departs from my own earlier work. I intend this
book to be an independent work, readable on its own.
My goal has been to put forth a certain kind of theory of
nonexistent objects. I realize that both the theory and its ex-
position are somewhat crude.. If the theory is of any value, it
will doubtless be developed by others in ways that I do not at
present envisage, and it seems to me time to place it in their
hands. As for the exposition, I ask the reader to try to see past
any poor choices of terminology and minor blunders. This
PREFACE xiii
burden will fall heaviest on those who are not in sympathy with
the idioms and presumptions of current professional "analytic"
philosophy. I've tried to make the main ideas of the book ac-
cessible to a broader audience, especially in the introduction
and part I.
I have benefited enormously from numerous discussions
with students and colleagues and with people outside academia
altogether. In addition to those persons mentioned in the text,
I have received special support and help from Kit Fine, Edmund
Gettier, Gael janofsky, Kathryn Pyne Parsons, Barbara Partee,
Robert Sleigh, David Woodruff Smith, John Vickers, and, most
of all, from Karel Lambert. They helped in various ways, mostly
by keeping me honest by one means or another.
I am grateful for a grant from the National Endowment for
the Humanities which, together with a sabbatical leave from the
University of Massachusetts (Amherst), gave me a year's free
time in which to write this book, and I wish to thank the Phi-
losophy Department at the University of California at Irvine for
providing office facilities and hospitality during that period.
University of Massachusetts
Amherst, Massachusetts
December 1978
Introduction
Are there objects that don't exist? The orthodox, mainstream

answer (in Anglo-American philosophy, anyway) is a resounding
"No I-there's no such thing as a thing that doesn't exist. Though
there may be kinds of things that are nowhere exemplified (e.g.,
being a winged horse) there is no particular. thing that fails to
exist." Or, put in positive terms: "everything exists."
This is a central tenet of contemporary philosophy. I'm
inclined to call it the "Russellian rut": "Russellian" because it
stems principally from Russell, and a "rut" because it's a view
in which most of us are so entrenched that it's hard to see over
the edges. The view is defended (though obliquely) in Russell's
classic paper, "On Denoting'i j! published in 1905, this paper
also contains RUSsell's most terse and unsympathetic treatment
of Meinong. (It is unfortunate that most people are acquainted
with Meinong only as the bad guy in this paper of Russell's, for
Russell published several reviews of Meinong's work which were
much more sympathetic, some even containing lavish praise).2
I. All references are to the bibliography at the end of the book. Most
references will be given in the text by means of the author's last name followed
by an abbreviation of the title of the work in square brackets; for example, a
reference to Russell's article "On Denoting" would be given as "Russell
[OD]," possibly followed by page references. The author's name will be
omitted if it is clear from the con text.
2. For example, "Before entering upon details, I wish to emphasize the
admirable method of Meinong's researches" ([1904 Review] P: 205) and
"Meinong's present position appears to me clear and consistent and fruitful
of valuable results for philosophy" ([I 905 Review] p. 538).
2 INTRODUCTION
'Rut' has an unfavorable connotation; the same point can

be put more favorably, I believe that, until recently, at least,
philosophy has been in a state that Kuhn ([SSR]) calls "normal
science.. " We have a set of paradigm beliefs and techniques which
we work with, and work with very fruitfully, but that we nor-
mally do not seriously question, One of these key beliefs is that
everything exists, and one of the paradigm techniques, in meta-
physics, if not in the philosophy of language so much anymore,
is Russell's famous theory of descriptions (which was first pre-
sented in the paper mentioned above). In this introduction I
want to discuss how such a situation arose, and then I want to
examine the current scene in more detail. This should set the
stage for the departure from orthodoxy envisaged in the body
of the book.
I. Historical Development
In 1900 Bertrand Russell believed in nonexistent objects, a view

he says he got from G . E. Moore (Russell [POM] p. xviii). And
indeed, superficially, it's quite a plausible view. If we forget or
inhibit our philosophical training for the moment, we are all
prepared to cite examples of nonexistent objects: Pegasus, Sher-
lock Holmes, unicorns, centaurs, . . . . Those are all possible
objects, but we can find examples of impossible ones, too;
Quine's example of the round square cupola on Berkeley Col-
lege will do . It is an impossible object, and it certainly doesn't
exist, so it seems to be an example of an impossible nonexistent
object. With so many examples at hand, what is more natural
than to conclude that there are nonexistent objects-lots of
them!
Well, by I9~9 at least, Russell had changed his mind. Non-
existent objects offend against our "robust sense of reality," and
the main task of the metaphysician seems to be to explain away
the apparent examples without committing himself to objects
that don't exist ([IMP] p. 170). And the theory of descriptions,
as all philosophers learn in graduate school, if not sooner, pro-
vides a means for doing that. 3
3.. The theory of descriptions is essentially a method for paraphrasing

INTRODUCTION 3
Most of Western philosophy has agreed with Russell ever

since. Why? The question has a certain poignancy. For in adopt-
ing the theory of descriptions (at least as Russell originally pre-
sented it in "On Denoting"), we paid a rather high price for
avoiding nonexistent entities. Formerly, we could think of a
statement such as 'Pegasus is winged' as a simple predication,
true if the object named is winged and false otherwise.. To use the
theory of descriptions to eliminate the "apparent" reference to
Pegasus, we first assume that 'Pegasus' is, logically speaking, not
a name at all, but rather a kind· of code for 'the winged horse of
Greek mythology' (or perhaps just 'the Pegasizer' as in Quine
[OWTI]); then we say that this description, which logically
underlies the apparent name, is not itself a constituent of the
proposition at all, but that the entire sentence really means
something like 'Some existing thing is a winged horse, and is the
only existing winged horse, and is winged). This is hardly an
intuitive result, and it doesn't even accord well with the apparent
data. For one thing, it makes all simple sentences containing
certain English locutions into the terminology of symbolic logic (or into
literal English renderings of that terminology). Its most famous application
involves the word 'the'. A sentence involving this word gets paraphrased,
roughly, by means of the format:
", . . the A . '. .' => 'Something is such that it is an A, and nothing else is
an A, and . . . it . . . '. For example, 'The king of France is bald' would be
paraphrased as 'Something is such that it is a king of France, and nothing else
is a king of France, and it is bald' (where ~is bald' occupies the position of the
ellipses in the format). On its orthodox construal, you are to read the 'some-
thing' as 'some existing thing' and the "nothing else' as 'no other existing
thing', though this construal is really independent of the paraphrase method.
The main advantage of the theory is supposed to be that it allows us to replace
a sentence that contains a term, 'the A,' which apparently refers to something,
by a sentence which does not contain such a term, The latter sentence can then
be denied without the denier having to (apparently, anyway) refer to 'the A'.
Thus someone who does not believe in the king of France can comfortably
avoid the apparent commitment to such a king that is suggested by 'The king
of France is not bald' by saying instead: 'It is not the case that something is such
that it is a king of France, and nothing else is, and it is bald'. Russell also held
that English proper names are disguised definite descriptions, to which the
paraphrase should be applied (see below in the text for an example); he and
others hoped to extend this treatment to all sorts of other linguistic construc-
tions (see Unnson [PA] chaps. 3,4, 10).
4 INTRODUCTION
names of nonexistents false, a matter that has been controversial

ever since it was proposed. And, worse, it seems not to work at
all for one of Russell's own paradigm tests: the sentence 'George
IV wished to know whether Scott was the author of Waverley' is
said to mean 'George IV wished to know whether one and only
one man wrote Waverleyand Scott was that man', a result that
hardly anyone finds plausible.s
So why did Anglo-American philosophy follow Russell
instead of Meinong? I do not believe that it was because of our
"robust sense of reality." For one- thing, the issue doesn't concern-
reality, but rather unreality; it is not what exists that is inques-
tion, but rather whether there is something more, something
outside the realm of existence.. And, for another, I think that our
intuitions are genuinely in conflict on this matter.. We do tend to
focus on what exists, if this is what a robust sense of reality comes
to. But we also have a contrary tendency to believe in particular
examples of nonexistent objects, such as Pegasus and Sherlock
Holmes.
No) I think we had much better reasons for agreeing with
Russell-at least two. The first is that the contrary view-a
Meinongian bloated ontology-seemed inevitably plagued with
difficulties, absurdities, and outright inconsistencies. I've men-
tioned Russell's attacks on Meinong's theories. Well, Russell
argued very effectively against Meinong. And Meinong did not
provide a persuasive reply. (I believe that Meinong's reply was
correct in part, but his reply was not persuasivej.f Thus we
inherited the belief that, whatever its initial plausibility, an en-
dorsement of nonexistent entities is untenable.
The second reason for following RusselI was this: Russell
took the view that everything exists, plus his theory of descrip-
tions, and on this foundation he erected one of the most impres-
sive philosophical systems ever known. He made great strides in
4. See Linsky [R] chap. V for a discussion, and chapter 5, section 4, of

this book. The reading in question is the de dicta one (see chapter 2, section 4).
5. The initial objections were raised in RusseIl [1905 Review], Meinong's
replies were in [USGSW] sec. 3, and Russell's final comments are in Russell
(1907 Review]. A brief synopsis of the exchange is in Chisholm [REP] pp.
10-I I. The arguments will be discussed in chapter 2, section 3.
INTRODUCTION 5
the development of modern logic, he provided a kind of founda-

tion for mathematics, and he articulated very powerful and inter-
esting metaphysical and epistemological views. More than that,
the techniques he employed-principally modem logic supple-
mented by the theory of descriptions-turned out to have wide-
spread application far beyond Russell's own theories. No wonder,
then, that we inherited many of his ontological views along with
the rest.
2. The Current Scene
Let me repeat my view of the current scene. Metaphysically, we

are just beginning to emerge from a state of·"normal science."
Normal science is characterized by the existence of certain para-
digm views which are simply taken for granted-and taken for
granted in such a way that it is hard to see what it would be like to
deny them. I've identified the view that everything exists as one
of these.
Now that may sound presumptuous. After all, people have
published books and articles in which they raise the question
whether there might be things that don't exist. They usually con-
clude that the answer is no, of course; but they do raise the
question. Well, that's not my point; my point is that in the nor-
mal everyday functioning of philosophy it is taken as a truism
that there are no nonexistent objects.
Let me illustrate this with a very simple example. In a recent
article, Keith Donnellan discusses a certain principle about
names, and he speaks in favor of this principle as follows: "it does
not involve our theory of reference in any difficulties: . . . it has
no Meinongian implications, no overpopulation with entities
whose existence is being denied" ([SJV] p. 26). In other words,
Donnellan takes for granted that Meinongian theories are bad,
and that nonexistent objects are bad. And this is not a rhetorical
error of his-quite the opposite. You can't defend everythit:lg you
say, and he is aware that he's addressing an audience that agrees
with him in taking those things for granted. (The literature
abounds with other examples.)
Here is a different illustration of the same point. It is rela-
6 INTRODUCTION
tively common to teach elementary logic in a manner that pre-

supposes the orthodox view. In particular, when students are
asked to symbolize these sentences:
(a) Tables exist.
(b) There are tables.
the instructor expects the same symbolization, namely:
(c) (3x) Tx.
And this expectation is not presented as embodying a metaphys-
ical view that might be wrong; instead it is treated as a matter of
pure logic. But it is not pure logic. Symbolizing both (a) and (b)
in the same way amounts to equating the quantifier 'there is'
with the quantifier 'there exists', an equation which makes sense
only if what exists exhausts what there is; and that is the meta-
physical view I am now questioning. 6
This example from logic also illustrates another point: why,
from the orthodox point of view, it is hard to see as sensible the
question whether there are things that don't exist. If you sym-
bolize this in the customary fashion, it comes out synonymous
with 'there exist things that don't exist') and that is inconsistent
on anybody's view. But, of course, such a symbolization begs the
question ·at issue.
The matter is actually a bit more complicated than I have
indicated, for variations in word usage also enter in. In particu-
lar, I think that we sometimes use 'there are' to mean 'there
exist'; when this is done, the symbolization discussed does not
beg the question in any overt sense. But we also use 'there are' in
a broader sense, a sense roughly equivalent to that of the word
'some', or 'at least one', and this usage cannot be appropriately
symbolized in the same way as 'there exist's? For example, we are
inclined to say both:
6. Quantifiers are locutions from symbolic logic. There are usually two,
the universal quantifier '(x)', which means 'everything is such that . . . ' or
'for any object x, . . . ', and the existential quantifier '(3x)', which means
'something Is such that ', or 'there is at least one thing such that . . . ',
or 'for at least one thing x, ).
7. The literature now contains numerous references to this distinction
INTRODUCTION 7
(d) There are winged horses-Pegasus, for example.

and:
(e) There are no winged horses.
When we truly utter (d), we are using 'there are' in the broad
sense. When we say (e), we mean that there are no real winged
horses, and (e) is appropriately symbolized using a quantifier
that is read 'there- eixst'.
To avoid ambiguity, I will try always to use 'there are' in the
unrestricted sense, the sense of 'at least one'; when I want to as-
sert existence, I will use 'there exist' or some similar locution.
During certain periods of history (e.g., the Middle Ages) the
view that everything exists would have been regarded as abso-
lutely outrageous. But since the early 19005 it has become the re-
ceived view, very firmly entrenched and almost impossible to
refute" There are several reasons why it is almost immune to ref-
utation. First, as the received view, it has authority on its side; it
is endorsed or presupposed by those of our contemporaries
whom we most respect. Also, as the received view, it is intuitively
obvious (to many philosophers, anyway), our intuitions having
been shaped by years of experience with theories that embody
this view. These factors tend to throw the burden of proof on
those who might want to challenge the received view. But that is
almost impossible to do, for reasons that Kuhn and many others
have made clear: the view in question is a high-level theoretical
claim in our metaphysical scheme. And high-level theoretical
claims don't confront the data directly; they can be tested against
data only as interpreted by some method. And the 'orthodox view
contains within it a methodology that interprets the data so as to
preserve and protect the claim that everything exists. I've given
between 'there is' and 'there exists' (although different terminology is sometimes
used), with about half the authors pointing out the importance of the distinc...
tion (e.g., Russell himself in [ElP]) and the other half saying that they can't
see any difference. Some seem to resolutely see no distinction, and I have no
hope of convincing them; as for others, I hope that acquaintance with the
theory described in the body of the text will provide an explanatory illustra..
tion.
8 INTRODUCTION
one example above, the manner in which we use logic to sym-

bolize claims so as to presuppose that everything exists. More
important, we've all learned to use Russell's theory ofdescriptions
to analyze away apparent reference to nonexistent objects; those
beliefs that seem to require nonexistent objects for their troth we
instinctively paraphrase into other beliefs that do not. And we
retain our conviction that apparent reference to the unreal must
be capable of being paraphrased away even when we don't see how to
do it.
It may now come as a surprise that I have hardly any objec-
tion to this situation whatsoever. I think the orthodox view is a
fine view; it has been extremely useful. I don't object to its taking
things for granted, nor to its defending some of its central claims
by means of a methodology that biases the data I don't object
4
because I think that any fruitful philosophical theory is going to

do just that.
But I do think it's a rut, and I'd like to look over the edge
and see how things might be different.. To do this, we need to
encounter an actual theory about nonexistent objects. That will
be the task of the present work. The way has been paved by a
recent mood in logic according to which logic ought not to rule
out nonexistent objects (see Scott [AML]). But much of the moti-
vation here has stemmed from the desire to. preserve the neu-
trality of logic, and this very neutrality has prescribed silence
about what nonexistent objects are substantively like. The same
also holds for much work in "free logic":
Free logic validates certain reasoning containing words
suchas 'Pegasus'. But it does not follow from this fact that it
is committed to a realm of entities among which is included
a flying horse. To be sure, one could develop a philosophical
semantics for free logic that does recognize a realm of non-
actual but. possible beings. . . . But one need not develop
the semantics that way. . . . In our development, talk about
non-existents is just that-i-t'talk" is what is stressed. "Non-
existent" object, for us, is just a picturesque way of speaking
devoid of any ontological commitment.. 8
8. Lambert and van Fraassen [D&C] pp. 199-200. See also Leonard
[EA&P] pt. X, and Marcus [Q,&O] .
INTRODUCTION 9
My intent is to describe in some detail the ontological commit-

ment that these and other authors wish to avoid.
3. Methodological Preliminaries
I have had certain goals in mind when working on this project,

and it will aid the reader's understanding to be aware of them.
For they have often influenced what I have said in ways that
would not be apparent from my words alone.
One goal I have had is to try to develop a theory that is
understandable to those who, like myself, approach this topic
from what I have called the orthodox tradition. My techniques
will be familiar to those in that tradition, and my terminology
has been kept as familiar as possible; I have made efforts to
clarify the nonorthodox tenninology that I found it important
to. use (principal examples are the notions of nuclear and extra-
nuclear properties, and impossible and incomplete objects). This goal
has also guided me in producing a theory that is, in certain re-
spects, as detailed and specific as passible.. I have avoided many
alternative theories and many variants of the chosen theory-not
because they seemed to me to be wrong, but because I couldn't
see how to develop them in sufficient detail to gr-asp clearly how
they would go. I can't emphasize this point too much; although
I have often taken a given path in order to avoid error that I saw
elsewhere, I have much more often taken a particular approach
just because it was the only one I could develop to the point
where I felt comfortable with it. The reader who tries to find
objections to alternatives lurking behind my choices will often be
frustrated (though I do think it is often much more difficult to
develop an alternative approach than it seems oat first glance,
and that "objection" to alternatives is often relevant). I don't
mean to suggest, ofcourse, that I have completely avoided vague-
ness and unclarity myself; these are matters of degree, and I have
had as one of my goals to minimize them.
A second goal that I have had is to produce a theory that is
inconsistent with the orthodox view. One popular style in philos-
ophy is to take a position that initially appears outrageous, and
then to "interpret;" it in such a manner that it turns into some-
thing that we already believed. This is not what I am trying to do
10 INTRODUCTION
here. If I am successful in my enterprise, some people-who begin

with orthodox opinions will end up agreeing with the theory
presented here, but only because they have changed their minds,
not because the theory is "really" one they originally held.
There is a danger that the question of whether there are
objects that don't exist should turn out to be a semantic quibble
rather than a substantive matter of disagreement between Mei-
nongians and the orthodoxy.. There are at least two ways in which
this might happen. First, we could define 'exists' to mean some-
thing like 'has spatio-temporal location', and then defend the
claim that some things don't exist by pointing to numbers, classes,
ideas, or similar things. I want to avoid such a move . I am not
sure how to define 'exists', but I may be able to say enough about
my intended use of the word to forestall such a trivialization of
the issues..
First, I want to follow Meinong in separating abstract things
(e. g., numbers, properties, relations, propositions) from concrete
things (tables, unicorns, people).. Meinong held that abstract
things never exist (they are the wrong sort of thing to exist);
instead, some of them have a kind of being called subsistence. I
want to avoid this issue entirely. When discussing problems of
existence and nonexistence, I'll limit myself entirely to a discussion
of concrete objects. So when I say that some objects don't exist,
I mean that some concrete objects don't exist-I don't have in
mind propositions, or numbers, or sets..
With regard to concrete objects, Meinong held that some
of them exist and some of them don't, and the onf?s that don't
do not have some other kind of being-s-for example, subsistence
(see Meinong [TO] .sec. 4, where he considers and rejects an
argument that purports to establish a kind of being that all
objects have). Russell objected that if there are objects that do
not exist, they have to have some other kind of being (see [OD],
[EIP]) . I have never been able to find more thana terminological
issue here, If there is an issue about whether nonexistent objects
have some kind of being, I intend to remain neutral on the issue..
This also goes for a view (which may be the same one) that I
have often heard expressed in conversation; it is that "everything
has its own special mode of existence." For example,Pegasus
INTRODUCTION 11
exists in mythology, Sherlock Holmes exists in fiction, . Some

would even emphasize that these sorts of-existence can be more
important than the everyday sort of existence that I share with
my house and my automobile. Well perhaps, but that is not the
issue that I will be discussing in this book. There is a perfectly
ordinary sense of the word 'exists' in which Sherlock Holmes does
not exist, and that is the sense that I intend when I call Holmes
a -nonexistent object. This is also the sense in which orthodox
.philosophers claim that there is nothing except what exists. (This
does not commit them to the unimportance of literature; they
need only hold that the importance of literature does not depend
on the existence of its characters.)
Even given these provisos, there is still room for disagree-
ment concerning exactly what concrete objects exist-for ex-
ample, concerning whether there exist any living beings on other
planets. I am not concerned with these issues and so I will, as a
matter of policy, agree (or at least not disagree) with others on
these issues. Specifically, I intend to use the word 'exists' so that
it encompasses exactly those objects that orthodox philosophers
hold to exist. In particular, it includes all the ordinary physical
objects that we normally take to exist, and it does not include
unicorns, gold mountains, winged horses, round squares (round
square things), Pegasus, or Sherlock Holmes. The theory given
below will say that there are unicorns, there is such a thing as
Pegasus, etc., but that none of these exist.
For reasons of simplicity, I have avoided entirely dealing
with tensed properties in the theory.. When I give examples like
'being blue', it would have been better to give examples like 'being
blue at time t' or 'being blue sometime'. My 'exists' is always
meant tenselessly, so I take it to be true that Socrates exists (Le.. ,
Socrates is not an example of a nonexistent object), even though
it is perfectly correct English to say, 'Socrates once existed, but
he no longer does'. For those who prefer tenses here, read my
'exists' as short for 'existed or exists or will exist'.
A second way that the ontological issues might be trivialized
would be if I were to reveal in the last chapter that my quantifiers
are merely "substitutional,' that 'There are winged horses' is
true only because the sentence 'Pegasus is a winged horse' is true,
12 INTRODUCTION
and the latter sentence is true in spite of the fact that 'Pegasus'
doesn't refer at all. Don't bother skipping ahead; I won't do this.
Whenever there is a choice) my quantifiers are always intended
to be interpreted objectually.P
A third goal of mine is that the theory described be con-
sistent with the data. This is too vague to be of much help, but
it can be formulated more precisely in terms of the relation be-
tween the proposed theory and the orthodox view. This is that the
only point of disagreement between these views should be ex-
plainable in terms of what some call a "robust sense of reality,"
and what Meinong called the "prejudice in favor of the actual."
Namely, the views should agree on any issue which concerns only
existing objects. More specifically, there should be no disagree..
ment between them concerning the truth value of any sentence
whose quantifiers are all restricted to existing objects and whose
singular terms all name real objects. (The sentence 'some things
do not exist' is not one of these.) Then the orthodox view can be
seen to be a kind of special case of the more libertine one devel-
oped here; it is the libertine view with blinders on, blinders that
prohibit vision of the unreal. Another way of putting this is that
the more libertine view should "reduce to" the orthodox view
when applied to old and familiar (i.e . , real) objects in much the
same way that relativity theory and quantum physics reduce to
classical physics when applied only to slow-moving middle-sized
physical objects. This will be made clearer in chapter 6, section 2 ..
9.. Quantifiers can be read in different ways. The usual reading is the
cbjectual reading; here a sentence of the form 'There is an A' is supposed to be
true if, and only if, there is an object which is an A.. The substitutional reading
treats 'There is an A J as being true if, and only i~ there is some true sentence of
the form 'N is an A', where 'N7 is a name. In case every object has a name and
no names fail to refer, then the objectual and substitutional readings are
equivalent; otherwise, they need not be equivalent.. For example, some pro..
ponents of substitutional quantification assume that there are names which
fail to refer but which nonetheless appear in true sentences of the form 'N is an
A', and they then hold that some sentences of the form 'There is an A' are true
even though no object is an A. For example, Leonard (in [EA&PJ pt. IX)
suggests that 'For some x, x is fictitious' is true, on the grounds that 'Santa
Claus is fictitious' is true, even though the name 'Santa Claus' does not refer to
anything, and no object (e.g., no person) is fictitious. See also Marcus [Q&Ol
INTRODUCTIOJV 13
One last note: throughout this work I have always used

'real' and 'actual' as synonyms for 'existent', Both these words
have uses in which they mean something quite different (e.g.,
both are sometimes used to mean 'genuine, as opposed to coun-
terfeit'). I never intend these other meanings.
The book consists of three parts. In part I I give a simple
sketch of the main outlines of the theory, some discussion of
motivation, and a sketch of an application of the theory-an
application to fictional objects. Part 11 contains the formal de-
velopment of the theory. I place a great deal of importance on
this part, However, many readers will not be inclined to dwell
in such detail on technical matters, and so I have tried to write
part III in such a way that it can be understood, in general at
least, without having read part 11 at all. Part III begins with an
informal exposition of the results of part 11; I then discuss various
applications of the theory in some detail. The book concludes
with some general characteristics of, and difficulties with, the
theory.
I
Initia1 Exposition
I
Initial Sketch of the Theory
The purpose of this chapter is. to give a very simple, crude sketch
of a theory of nonexistent objects. I will confine myself here to
a description of the theory, Discussion of motivation and of
application will be postponed until later chapters, as will de-
tailed developments and discussion of subtleties.
I. Objects
I am going to assume that no two existing objects have exactly

the same properties.. This is not so much an assumption about the
paucity of existing objects as it is an assumption about the variety
of properties; in particular, I assume that for any existing object
there is at least one property (and probably many) that it has
and that no other existing object has.. Anyway, given this as-
sumption, there's a natural one-one correlation between real,
existing objects and certain nonempty sets of properties. For
example, Madame Curie is a real object, and. correlated with
her is the set of properties that she has:
Curie
X The set of Madame Curie's properties
Now, make a list of all existing objects. Correlated with each

one is a set of properties-i-the set of all the properties that it has:
Portions of this chapter have appeared in various forms in Parsons
[PMS], [MAFO], and [RNO].
17
18
INITIAL EXPOSITION
Real Objects Sets ofProperties

The set of OJ.'s properties
The set of ~'s properties
The set of Oa'8 properties

The left-hand list now exhausts the ontology of concrete objects
that people like Russell, Quine, Frege, and most of us find ac-
ceptable; the existing objects constituteall there is. But the theory
now being presented says that there is a lot more, and it goes
like this . .It is not clear how to continue the left-hand -list (that's
our goal), but you can easily see how to continue the right-hand
list; just write down any other nonempty set of properties.. For
example, write down:
{goldenness, mountainhood, . . .},
filling in whatever properties for the ellipses that you like. Now
the theory under discussion says that for any such set in the right-
hand list, there is correlated with it exactly one object. So write
in "0«+1" in the left-hand list:
°a+l {goldenness, mountainhood, . . .}

The object 0a+l can't be an existing object, because it has the
properties goldenness and mountainhood-s-it's a gold mountain
-and there aren't any real .gold mountains. But, as Meinong
pointed out, that doesn't mean that there are no. unreal gold
mountains; although certain narrow-minded people object
to this, that's just because they're prejudiced! (He called this
"the prejudice in favor of the actual" ([TO] sec" 2).)
It's clear how to extend the right-hand list: just include arry
set of properties that isn't already there. Corresponding to each
such set is a unique object, and vice versa; that is, each object
appears only once in the left-hand list . The two lists extend our
original correlation, so that it is now a correlation between all
objects and the sets of properties that they have.. !
1. This correlation is not one of identity; that is,. I am not saying that
INITIAL SKETCH OF THE THEORY 19
Actually we can dispense with talk of lists and correlations

and present the theory in a more direct manner in terms of two
principles. For reasons that will become apparent shortly, let me
call the properties I have been discussing nuclear properties. The
principles are:
(I) No two objects (real or unreal) have exactly the same
nuclear properties.
(2) For any set of nuclear properties, some object has all the
properties in that set and no other nuclear properties.
Principle (2) does most of the work; it's a sort of "comprehen-
sion" principle for objects. Notice that principle (2) does not
require that objects be "logically closed"; for example, an
object may have the property of being blue and the property of
being square without having the property of being blue.. and-
square.. This lack of logical closure will be important in certain
applications of the theory, particularly applications to fictional
objects and objects in dreams.
Many nonexistent objects will be incomplete.. By calling an
object "complete," I mean that for any nuclear property, the
object either has that property or it has its negation. This charac-
terization presupposes that it makes sense to talk of the "nega..
tion" of a nuclear property in a somewhat unusual sense. This
matter will require some discussion, most of which will be post-
poned until chapter 5, by which time most of the "logic" of
nuclear properties will have been developed. For the time being,
we can get along with the assumption that, for any nuclear
property p, there is another nuclear property q which existing
objects have if and only if they don't have p, and which I call
the nuclear negation of p, or just 'non-p' for short.. Notice that the
nuclear negation of a nuclear property p will not be a property
that any object has if and only if it does not have p, for no nuclear
property fits that description (by principle (2) any nuclear prop-
erty~q is such that some object has both p and q). So the reader
objects are the sets of properties that they have. The theory being developed
is not a "bundle" theory, according to which objects are bundles of properties
(though it may not be inconsistent with all such theories).
20 INITIAL EXPOSITION
should keep in mind that this is a somewhat unusual use of the

word 'negation'.
Given this account of nuclear property negation, all existing
objects are complete.f Some nonexistent objects' are complete,
too (see below); but some aren't. Consider the object whose
sole nuclear properties are goldenness and mountainhood. It
does not have the property of blueness, nor does it have the
property of nonblueness; I will say that it is indeterminate with
respect to blueness. That object will in fact be indeterminate
with respect to every nuclear property except goldenness and
mountainhood. (The object in question may be the one that
Meinong was referring to when he used the words 'the gold
mountain' ; whether this is so or not involves questions of textual
interpretation that I amunsure about.)
Completeness is different from logical closure.. Consider
the set of properties got by taking all of my properties and re-
placing "hazel-eyed" by "non.. hazel-eyed." According to princi-
ple (2), there is an object which has the resulting properties and
no others. This object will be complete but it will not be logically
closed. For example, it has brown-hairedness and it has non-
2. It is possible to develop a different notion of incompleteness which

does not depend on the notion of the negation of a nuclear property. Call p and
q complementary if, in fact, all and only existing-objects that have p lack q. Then
we ean call an object incomplete*' if it lacks both of a pair of complementary
nuclear properties. All incomplete objects are incomplete*.
It is not obvious that either of these leads to the best way to develop the
theory. There is a widespread view to the effect that existing objects fall into
categories, and that some properties that apply meaningfully to objects of one
category may not apply meaningfully to objects of other categories. For ex-
ample, it may be meaningful to say of me that I am complacent, but not mean-
ingful to say this of my car. Perhaps, then, there is a notion of property negation
according to which the negation of p is a property that is had by all and only
the existing objects of the appropriate category which do not have p, and then a
notion of completeness that is relativized to categories. I have avoided develop-
ing the theory in this way because (I) it would be considerably more com-
plicated than the theory I have developed, and (2) it would involve me in
making sense of what a category is-not an easy matter. The theory is already
cornplicated and involved enough as it stands. Readers who favor the category
approach should view the theory I am developing as an idealization.
INITIAL SKETCH OF THE THEORT 21
hazel-eyedness but it does not have the nuclear property of being

both brown-haired-and-non-hazel-eyed.~
To get an object that is logically closed yet incomplete, add
to "the gold mountain" all nuclear properties that are entailed
by goldenness and mountainhood. Then it will have) for example,
the nuclear property of either-being-located-in-North-America-
or-not-being-Iocated-in-North-America, but it will not have
either of those disj uncts; it will be indeterminate with respect to
being located in North America.
It may be appropriate at this point to allude to a remark
that Meinong was fond of making: perhaps no one could have
a practical interest in certain of these objects, but that should
not prejudice the philosopher against them.
Some objects are impossible. By calling an object x possible
I mean that it is possible that there exist an object which has all
of x's nuclear properties (and perhaps more besides). 4 All existing
objects are automatically possible objects by this definition. And
some unreal ones are, too-for example, "the gold mountain."
But consider the object whose sole nuclear properties are round-
ness and squareness (this may be Meinong's famous "round
square"). This is an impossible object because there could not
be an existing object that has both these properties. Still, as
Meinong pointed out, that doesn't prevent there from being an
impossible object that has them.
The reader might find it profitable to verify the following
claims. Not all of them follow with absolute certainty from what
3. I am assuming here that it makes sense to talk of "nuclear conjunc..
tions" of nuclear properties, "nuclear disjunctions," etc. This will be discussed
more fully in chapter 5, section 2.
4. There is some arbitrariness in terminology here, for several other
notions may have an equally good claim to be called possibility. For example:
(a) x is possible. if and only if it is possible that some"object have exactly
x's nuclear properties..
(b) x is possible?, if and only if it is possible that x exists.
Possibility2 is very different from either of the other notions; to call an object
possible, is to make a de re modal claim about it, which is not the case for
either of the others. The present theory is very neutral about de re modalities;
see chapter 5, section I.
I have said so far (e.g., I haven't given a precise definition of

logical closure, .and I haven't yet discussed what a "nuclear"
property is), but the fact that they are supposed to be consequences
of the theory being presented will help clarify the theory.
(a) Every object that is both complete and possible is-logi-
cally closed. _.
(b) Any object that is impossible and logically closed is
complete. (There is exactly one such object: the object
which has every nuclear property.)
(c) Except for -(a) and (b), all possible combinations of
completeness, possibility, and logical closure are man-
ifested; that is, there are objects that are complete,
closed, and possible; objects that are complete, closed,
and impossible; .etc.
(cl) There are objects that don't exist but that are complete,
possible, and logically closed. (The reader might want
to postpone this one until chapter 5, section 2.)
Principle (2) requires that there be a "null" object, that is,
an object that has no nuclear properties at all (and principle (1)
says that it is unique) . I am not at all sure whether this is desir-
able. It makes for a certain amount of theoretical simplicity,
and that offers some justification. But it would not make a great
deal of difference to the applications of the theory that I know of
ifit were omitted (by inserting 'nonempty' after 'any; in principle
(2)).
Principles (1) and (2) yield a theory that has an important
virtue: they not only tell us that there are nonexistent objects,
they also in part tell us what nonexistent objects there are, and
they tell us what properties they have. What nuclear proper-
ties they have anyway, which brings us to the next section.
2. Extranuclear Predicates
Not all predicates can stand for nuclear properties. Take 'exists'.
In the theory I've sketched, if we allowed 'exists' to stand for a
nuclear property, there would be trouble. For example, suppose
that it did stand for a nuclear property, existence. Now consider

this set ofproperties :
{goldenness, mountainhood, existence}.
If existence were a nuclear property, there would be an object
correlated with this set of properties; call it "the existing gold
mountain." Then the existing gold mountain would turn out to
have the property, existence; that is, the existing gold mountain
would exist. But that's just false .
Initially, we were troubled by there being a gold mountain;
Meinong placated us by .pointing out that it's only an unreal
object, it doesn't' exist. But in the case of the existing gold moun-
tain, this option doesn't seem open. Conclusion: 'exists,' at least
as it is used above, does not stand for a nuclear property.s I'll call
'exists' an extranuclear predicate, and, in general, I'll divide pred-
icates into two categories: those which stand for nuclear pro}r
erties, which I'll call nuclear predicates, and the others, which
I'll call extranuclear..
Which are which? First, here are some examples:
Nuclear Predicates .-
'is blue', 'is tall', 'kicked Socrates', 'was kicked by Socrates',
'kicked somebody', 'is golden', 'is a mountain'
Extranuclear Predicates •.
Ontological: 'exists', 'is mythical', 'is fictional'
Modal: 'is possible', 'is impossible'
Intentional: 'is thought about by Meinong', 'is wor-
shipped by someone'
Technical: 'is complete'
I'd like to emphasize that this division of predicates into
5. This argument parallels one of Russell's objections to Meinong; see

chapter 2, section 3. Meinong distinguished between existing (a "Sein" prop-
erty) and being existent (a "Sosein" property); it is the former property that is
under discussion in my argument. Strictly, the argument shows only that either
goldenness or mountainhood or existence is extranuclear. To prove that the
culprit is existence, suppose that existence were nuclear and consider the object
which has only existence. Since this object is incomplete, it doesn't exist, con-
tradicting the assumption that it does. (This argument was pointed out to me
by Dorothy Graver.)
nuclear and extranuclear is not peculiar to Meinong at all; it's

an old and familiar one. People such as Frege and Russell distin-
guish predicates that stand for properties of individuals from
those that don't. The extranuclear predicates listed above are
mostly ones that Frege and Russell have been telling us all along
do not stand for properties of individuals, For example, is 'exists'
a predicate? Some people say flatly no. Frege tells us that it is a
predicate, but not a predicate of individuals; it is a higher-order
predicate, a predicate of concepts. Likewise, w~ all know that 'is
possible' is either not a predicate at all, or a predicate not of in-
dividuals but of propositions or sentences. With the intentional
predicates we're not sure what to say, but we are sure that there is
trouble in supposing them to be properties of individuals. 6
Our historical situation yields a very rough kind of decision
procedure for telling whether a predicate is nuclear or extranu-
clear. I t is this: if everyone agrees that the predicate stands for an
ordinary property of individuals, then it is a nuclear predicate
and it stands for a nuclear property. On the other hand, if every-
one agrees that it doesn't stand for an ordinary property of in-
dividuals (for whatever reason), or if there is a history of contro-
versy about whether it stands for a property of individuals, then
it is an extranuclear predicate, and it does not stand for a nuclear
property. Of course, this "decision procedure" is a very imperfect
one. Probably its main virtue is to give us enough clear cases of
nuclear and extranuclear predicates for us to develop an intuition
for the distinction, so that we can readily classify new cases. I find
that I have such an intuitive ability, and that other people pick
it up quite readily; even those who are skeptical about the via-
bility of the distinction seem to agree about which predicates are
supposed to be which (except for comparatives, which will be
discussed in chapter 6, section 4) .
The theory itself will help by putting severe constraints on
what can be nuclear. For example, it is a thesis of the theory that
no nuclear property F satisfies the formula:
There is a set X of nuclear properties, not containing F, such
that every object which has every member of X has F..
6. Ryle says that these predicates are "status" or "quasi-ontological"
predicates (Ryle [ID]).
This is because if F is nuclear and F is not a member of X, then

the object which has exactly those nuclear properties in X has
every member of X without having F. For similar reasons, no
nuclear property F satisfies the formula:
There is a set X of nuclear properties, not containing F, such
that every object which has every member of X lacks F.
If we make some minimal assumptions about nuclear properties,
principles like these will show that lots or"properties are extra-
nuclear. For example, suppose that we assume that being a uni-
corn, being round, and being square are all nuclear.. Then we
can show, for example, that existence is extranuclear, by picking
as X the unit set of being a unicorn. Every object that has every
member of X-that is, every object that is a unicorn-lacks
existence. So existence is extranuclear, by the second schema."
Similarly, if we pick X to be {roundness, squareness} then the
first schema shows impossibility to be extranuclear, and the
second schema does the same for possibility.
In the .modalized version of the theory, we will have the
following principle: no nuclear property satisfies:
There is a set X of nuclear properties, not containing F,
such that it is possible that every object which has every
member of X has (lacks) F.
If we consider possible worlds in which (unlike our own) there
are no myths about unicorns, no stories about unicorns, and
Meinong never thought about a unicorn, then, using the unit set
7. This test for distinguishing nuclear from extranuclear properties will

not work if applied from a perspective of complete skepticism. For example,
you can't show that redness is extranuclear by using the second schema, be..
cause for any set X of nuclear properties we can (via the theory) produce an
example of an object which has all the properties in X and which also has red-
ness: ies the object that has exactly the properties in X plus redness. But
doesn't that beg the question, by assuming that redness is nuclear (for, otherwise,
the theory does not say that there is an object which has exactly the properties
in X plus redness)? In a sense, yes, but the point is, rather, that making this
assumption and then judging that the resulting object is red is, from the point
of view of the general theory, at worst dubious, whereas making a similar assump-
tion about existence, and judging that the resulting object exists, is (for some
choices ofX) clearly wrong.
of being a unicorn, this schema can be used to show that being

fictional, being mythical, and being thought about by Meinong
are all extranuclear.
Exercise for the reader: show that being complete is extra-
nuclear. 8
Ultimately, the distinction between nuclear and extra-
nuclear predicates and properties will gain viability by being
incorporated into a general theory of properties and objects.
This task will occupy most of the present book.
I think it is not very important whether you say that extra-
nuclear predicates stand for a special sort of property-s-extra-
nuclear properties-or whether you say that they do not stand
for any properties at alL In the theory I sketch, I will assume that
there are extranuclear properties; this allows me a freedom of
exposition that I would not otherwise have. But the important
point is to distinguish extranuclear from nuclear predicates, and
thereby to sharpen our intuitions about what nuclear properties
are. For it is by means of nuclear properties, not extranuclear
ones, that we individuate objects.
What has been said above about properties applies also to
relations. Some relations are ordinary nuclear ones: being an
aunt of, kicking, being kicked by. Others are extranuclear: the
relation of identity is a paradigm case. Some relations seem to be
nuclear with respect to their first place and extranuclear with
respect to their second place: thinking about, believing to be
blue. One way in which relations are important in the present
theory is that "plugging up') one place of a two ...place relation
can yield a property: being kicked by Socrates, being thought
8. Here is a solution. First, pick X to be the set of nuclear properties
possessed by Madame Curie: Notice that Madame Curie is complete, and so is
any object which has all of her properties. So if completeness is not a member
of X, the first schema classifies it as extranuclear. If completeness is a member
of X (it isn't, but we haven't proved that yet), then it is a nuclear property. So
now pick as a new X the result of replacing completeness by its nuclear negation
in the old X. By the way in which we have obtained the new X from the old
one, we know that any object that has all its members is complete and also that
completeness is not a member; so now the first schema classifies completeness
as extranuclear.
about by Meinong. This plugging-up process will be discussed in

chapter 3, its logic will be worked out in chapter 4, and applica-
tions will be discussed in chapter 7. .
3. Identity
One of the major sources of opposition to nonexistent objects is

the feeling that they are somehow "disorderly." I think that this
feeling comes primarily from lack of familiarity with a clear
theory about nonexistent objects. Within the present theory, the
realm of the nonexistent is a reasonably orderly one.
In addition to Russell's original objections to Meinong's
theory, which will be discussed in chapter 2, the main charge of
disorderliness is based on the claim that the concept of identity
does not apply to nonexistent objects. For example, after making
some aesthetic criticisms of nonexistent but possible objects,
W. V. Quine says:
Take, for instance, the possible fat man in the doorway;
and, again, the possible bald man in that doorway. Are they
the same possible man, or two possible men? How do we
decide? How many possible men are there in that doorway?
Are there more possible thin ones than fat ones? How many
of them are alike? Or would their being alike make them
one? Are no two possible things alike? Is this the same as
saying that it is impossible for two things to be alike? Or,
finally, is the concept of identity simply inapplicable to
unactualized possibles? But what sense can be found in talk-
ing of entities which cannot meaningfully be said to be
identical with themselves and distinct from one another?
These elements are well-nigh incorrigible. ([OWTI] p. 4)9
It is clear that the conclusion expr.essed in the last sentence is a

bit hasty, but the questions leading up to it just as clearly deserve
answers. I'll answer with respect to the theory developed in the
present work.
9- These difficulties with nonexistent objects are foreshadowed in Russell

[19°5 Review].
Identity is meaningfully applicable to all objects, existent

and nonexistent alike. All objects in fact obey the principle of the
identity of indiscernibles: if x and y have all the ~ame properties,
then x = y. But this much is trivial, since I am willing to count
"being identical with x" as a property, albeit an extranuclear
one. But objects also obey a much more powerful principle: the
principle of the identity of nuclear indiscernibles:
1Nl: If x and y have exactly the same nuclear properties,
then x =y.
This is our principle (I), used above in explaining the theory.I?
What, then, about those men in the doorway? Well, regard-
ing "the possible fat man in the doorway," there is no such thing,
for there are many possible fat men in the doorway. I am now
assuming that being fat, being a man, and being in the doorway
are all nuclear properties, but that being possible is an extranu-
clear one. There are many objects which have those three nuclear
properties, some of them possible and some of them impossible.
Similar remarks apply to "the possible bald man in the doorway."
So the question of whether "they" are the same possible man or
not does not make sense; this has nothing to do with "their"
nonexistence, it is simply the fact that no specific objects have
been singled out to ask the question about. Some possible fat men
in the doorway are also possible bald. men who are in the door-
way, and some are not.
How many possible men are there in the doorway? At least
as many as there are consistent sets of nuclear properties which
contain the properties of being a man and being in the doorway.
10. In a less simplified development of the theory, I would use "tern-

poral" nuclear properties instead of their "eternalized" versions-for example,
being blue, simpliciter, instead of being blue..at-time-t. The difference is that
temporal properties are not just had or not had, but rather had at given times,
or not had then, whereas the eternalized versions have the times built in, and
they are simply had or not had. In a developmen t of the theory using temporal
properties, objects would be correlated with functions from times to sets of
properties instead of with sets of properties, and 1Nl would be replaced by:
If x and y have exactly the same nuclear properties at all times, then
x =y_
INITIAL SKETCH OF THE THEORr 29
There will be at least a couritably infinite number of these. Are

there more possible fat men than thin ones? Probably not; there
are an infinite number of each, but parity suggests that the car-
dinality will be the same for each.
Would their being alike make them one? If "being alike"
means not differing with respect to any nuclear property, then
yes.. Are no two possible things alike? No two things are alike,
possible or impossible.
2
Meinong and .Motivation
I. Meinong's Motivation
The basic insight behind Meinong's theory of objects was that

every thought has an object-regardless of whether or not that
object exists. Bertrand Russell transferred this principle to lan-
guage, and attributed to Meinong the doctrine that every denot-
ing phrase stands for an object (Russell [OD]) . This principle,
when fleshed out in a natural way, yields a powerful and exciting
ontological view.. The natural fleshing out is that every denoting
phrase refers to the "right" object. For example, Meinong holds
that the phrase 'the gold mountain' refers to an object which is
golden and is a mountain, and that the phrase 'the round square'
refers to an object which is both round and square. The general
view seems to be that, in the case of definite descriptions at least,
any definite description refers to an object that satisfies the de-
scription. I will call this the unrestricted satisfaction principle. In spite
of the fact that Meinong specifically disavows certain applica-
tions of this principle (see the discussion in section 3, below), it
was attributed to him by Russell- (with some justice) and it re-
mains his most famous "contribution.. "
I do not think that the main evidence for unreal objects
comes from this principle, and I do not intend to endorse it
I. This involves an interpretation of Russell's argument that goes beyond
the actual text. For a slightly different interpretation see Lambert [10], RoutIey
[DIO], and Lambert [ODIO].
30
MEINONG AND MOTIVATION 31
myself. True, it does entail that there are objects (e.. g., the gold
mountain) of a sort that do not exist. But it entails too much;
the principle is in fact inconsistent. Russell's arguments against
Meinong almost show this, though not quite; they will be dis-
cussed in section 3. But a variant of one of his arguments does the
trick. We need only consider the definite description 'the object
which is such that it is golden and also such that it is not the case
that it is golden.' By the satisfaction principle, this refers to an
object, call it 's'. which satisfies the- .description.. That is, g satis..
fies the description 'it is golden ~nd it is not the case that it is
golden'. So the following is true: 's is golden and it is not the case
that g is golden). But that is a contradiction. Meinong mayor
may not have intended to endorse contradictions (the scholarly
evidence is unclear);2 but I do not, for contradictions are not
true.. Even though I am advocating a theory according to which
there are impossible objects, this will not lead to contradictions
(again, see section 3).
There is a parallel here with the paradoxes .of set theory..
The unrestricted comprehension principle of naive set theory
says that for any open formula there is a set consisting of exactly
those things which satisfy the formula . Taking the formula to be
'X ~ X' then yields "the Russell set", which is a member of itself
if and only if it isn't a member of itself. Some reacted to this
result by giving up sets altogether, but the majority view was to
search for some more modest principles of generating sets, Some
fairly natural principles of set theory had to be given up, but
others were preserved.
Similarly the unrestricted satisfaction principle of naive
object theory leads to contradiction, and must be given up.. The
majority view here has been to give up all but the existing ob . .
2. The principal evidence in favor of the view that Meinong was willing
to endorse contradictions comes from sec. 3 of [ USGS W] (p. 16):
Naturally I cannot in any way evade this consequence: whoever once
has dealings with a round square will not be able to stop when faced
with a square or some other sort of object which is simultaneously
round and not round. But one will also, as far as I can see, have
weighty reason hereupon to take the initiative: the principle of con..
tradiction is to be applied by no one to anything other than to reality
and possibility [my own translation; not to be highly trusted].
jects, a rather severe reaction. Certainly some of the natural "ob-

jects" of naive object theory must be given up, but I think there
is still hope that some of the nonexistents can be saved. (I'm sure
that this can be done consistently; the open question is whether
it can be done plausibly.)
2. Evidence for the Theory
I will discuss Meinong's restriction of the satisfaction principle in

section 3. As I have stated, I do not think that the major motiva-
tion for the endorsement of nonexistent entities should come from
this principle. The motivation should come instead from a host
of particular 'propositions which we believe and which seem to
commit us to unreal objects.
Consider, for example, the following remarks:
(i) Ironically, a certain fictional detective (namely, Sher-
lock Holmes) is much more famous than any real detec-
tive, living or dead.
(ii) Several of the Greek gods were also worshipped by the
Romans, though they called them by different names.
(iii) Any good modern criminologist knows more about
chemical analysis than Sherlock Holmes knew..
These are claims that we believe, and they seem, prima facie,
to commit us to a fictional detective (Sherlock Holmes) and to
Greek gods. Anyone can deny this, of course, but to deny it is to
make a claim that is primafacie implausible. The case can even
be strengthened by comparing (i)-(iii) with certain parallel
examples:
/)
(iIronically, a certain ancient philosopher (namely,
Plato) is much more famous than any modern phi-
losopher, living or dead.
(ii') Several of the Qreek cities were also occupied by the
Romans, though they called them by different names.
(iii') Any good modern criminologist knows more about
chemical analysis than Lavoisier knew.
We would normally say that (i')-(iii /) commit us to a certain
MEIJVONG AND MOTIVATION 33
Greek philosopher (Plato), Greek cities, and Lavoisier. An op . .

ponent of unreal objects must either find a difference between
the logic of the former and latter cases, or else deny the com-
mitment in the latter cases. The first move calls for a theory of
logical form that has not yet been developed." the second calls
for an explanation of commitment that differs from the normally
accepted ones.
I don't think that it is at all plausible to deny the truth of
(i)-(iii), nor to attribute to (i)-(iii) relevantly different logical
forms than to (i')-(iii'). But some people would deny that any of
these examples contain commitment to objects of any sort; I
will discuss briefly what I take to be the most popular variants
of this position.
Probably the most common way within the orthodox tradi-
tion to sidestep apparent commitment to objects is to hold that
the linguistic forms in question create "nonextensional" contexts.
No one thinks that 'It is possible that a unicorn is approaching'
commits one to there being unicorns; the prefix 'It is possible
that' creates a context that cancels the commitments of the sen-
tence that follows it . Similarly, one could hold that the locutions
used in (i)-(iii)-'is more famous than', 'was/were worshipped
by', and 'knows more about chemical analysis than'-create
such nonextensional contexts. Unfortunately, this is not only
implausible, it does not even fit accepted accounts of nonexten-
sionality,
There are two generally acknowledged tests for nonexten-
sionality: failure of substitutivity of identicals and failure of ex-
istential generalization. With regard to substitutivity ofidenticaIs,
the adverb 'necessarily' is shown to create a nonextensional con-
text by virtue of such failures of inferences as:
Necessarily anyone who flies to the Evening Star flies to the
Evening Star.
The Evening Star is the Morning Star.
3. Actually the theory developed. in this book could be interpreted as
attributing different logical forms to (i)-(iii) than to (i')-(iii t ) , since the former
contain extranuclear predicates in certain places where the latter contain
nuclear ones; but this theory does not provide a means for avoiding the commit-
ment to nonexistent objects apparent in (i)-(iii).
INITIAL EXPOSITION
Therefore, necessarily anyone who flies to the Evening Star

flies to the Morning Star.
But this test for nonextensionality indicates that all the contexts
mentioned above are extensional. For example, from:
Jimmy Carter is more famous than WaIter Mondale.
and:
Jimmy Carter is the present president of the U.S..
we can infer:
The present president of the V.S. is more famous than
WaIter Mondale.
Nor is the situation different in the case of nonexistent objects;
from:
Sherlock Holmes is more famous than any real detective.
and:
Sherlock Holmes is the principal character in Conan Doyle's

novels.
we can infer:
The principal character' in Conan Doyle's novels is more
famous than any real detective.
The same holds for 'worships'; from:
Samantha worships her queen.
.and:
Samantha's queen is the evilest person in the nation.
we can infer:
Samantha worships the evilest person in the nation.
Again the situation doesn't change where nonexistent objects

are concerned; from:
Francis, the talking mule, worships Pegasus.

MEINONG AND MO TIVA TION 35
and:
Pegasus is the winged horse of Greek mythology.
we can infer:
Francis, the talking mule, worships the winged horse of
Greek mythology.
The other test for nonextensionaIity is failure of existen-
tional generalization. For example, 'believes that' is shown to
create a nonextensional context by the failure of the inference:
Janes believes that the tallest spy is a spy_
Therefore, someone is such that Jones believes that he is a
spy.
Here) too, the idioms in question seem to be extensional; from:
Jimmy Carter is more famous than WaIter Mondale.
one can infer:
Someone is such that he is more famous than Walter Mon-
dale.
(Note: Of course, existential existential generalization may fail
in these cases; from the above premise alone we may not be able
to infer 'Someone who exists is such that he is more famous than
WaIter Mondale'. But this is not the inference in question.)
Perhaps I have belabored the nonextensionality issue un-
necessarily, for even if the idioms in question were nonextensional,
(i) and (ii) would still provide evidence for nonexistent objects.
For they seem to have the forms of de re statements, and these
are normally thought to reintroduce the commitment canceled
by the nonextensional idiom. For example, (i) has the ostensible
form:
(i") There is something such that it is a fictional detective
and it is more famous than any real detective.
Here the quantifier comes outside the scope of 'is more famous
than', and so (i") is explicitly in the "existentially" quantified
form that is commonly acknowledged to form a paradigm case
of commitment to objects.
. Another attempt to avoid the commitment to nonexistent

objects implicit in (i)-(iii) involves the appeal to substitutionaI
quantification. One can accept a sentence such as (i") above,
granting its explicit quantificational form, but insist that the
quantifier here is to be read as merely substitutional.. 4 This raises
a myriad of issues, most of which I do not have a settled opinion
about, and which I hope to avoid. My main reason for not tak-
ing substitutional quantification seriously here is that it can be
used just as well to avoid commitment to anything at all. If I
insist that there are cows, an anticowist can grant me the truth
of what I say, but hold that this does not really commit him to
there being cows, since the 'there are' may be taken as merely
substitutionaL Of course, if there were something peculiar about
the use of quantifiers to range over nonexistent objects, as
opposed to existent ones, then perhaps substitutional quantifi-
cation might play a role in explaining the difference. But the
question being discussed is not whether one could consistently
construe the data so as to avoid commitment to nonexistent ob-
jects, but rather whether there is some reason to do so. Substi-
tutional quantification may provide a mechanism for doing so,
but it doesn't provide a reason.
Probably the main orthodox reaction to sentences such as
(i)-(iii) does not appeal either to nonextensionality or to sub-
stitutional quantification, but rather to the need for some sort of
paraphrase of the sentences in question so as to avoid the ap-
parent commitment to unreals. Ever since Russell showed how to
do this with sentences like 'The king of France isn't bald', there
has been a widespread "faith in favor of the paraphrase," which
even at times tends to outrun the prejudice in favor of the actual,
The extent to which faith in the existence of an appropriate
paraphrase outruns the believer's ability to give such a para-
phrase is often quite striking, though if my analysis of the ortho-
dox tradition in terms of "normal science" is correct, this should
not be surprising at all; it is the sort of thing that is typical of
normal science. Again I have no objection, unless this faith is
4- See Marcus [Q&O] for a discussion of this view. For an explanation

of the term 'substitutional quantification', see the introduction, note 9~
directed at the plausibility or interest of the present exercise,

which aims at looking outside of the current normal science.
Without going into elaborate detail, I think there are two
reasons that the possibility of paraphrase does not diminish the
primafacie case for unreals offered by (i)-(iii). The first is that the
paraphrases are not at hand, and they are much more difficult to
produce than may' seem apparent at first glance. (There are
paraphrases at hand, of course-for example, Quine's develop-
ment of Russell's theory in [0 WTI]; but these paraphrases of
(i)-(iii) make explicit, rather than removing, the commitment to
nonexistent objects.) The second reason is that it is implausible to
paraphrase (i)-(iii) without also paraphrasing (i')-(iii /) in a
similar fashion; but the latter seem unobjectionable as they
stand. 5
There are lots of examples other than (i)-(iii); I have
avoided' discussing some of these because they have become so
familiar that philosophers have become used to biting the bullet
where they are concerned. For example, there is hardly any
orthodox theory that makes (iv) true and (v) false:
(iv) Pegasus is the winged horse of Greek mythology.
(v) Zeus is the winged horse of Greek mythology..
This is because terms that appear to refer to nonexistent objects
are treated as if they do not refer at all, and all terms that fail to
refer at all get treated on a par. The theory to be described will
treat 'Pegasus' and "Zeus' as terms that refer to different (non-
existent) objects; (iv) and (v) will be treated as ordinary identity
statements, the former being true and the latter false.
Viewed impartially, I believe that there is nothing within
the orthodox tradition to undercut the prima facie plausibility,
offered by examples like (i)-(v), that there are nonexistent ob-
jects. Of course, I have not proved that there are nonexistent
objects; I don't believe that anyone could do that in a non-
question-begging way. Instead, I have tried to show that there is
enough reason, primafacie, to believe in them, to make it worth-
5. There is a good discussion of this point and others that have been
made in this section in Howell [FO :HAHA] secs. IV and V. See also Chisholm
[BB&JVB] and, for a historical perspective, Urmson [PAl pp. 148-49-
while to try to develop a theory about them, with a reasonable hope

that it will turn out to be true.. Whether it is true or not will ultima.. .
tely be decided in terms of global considerations-s-how well it ac-
cords with the data and with other theories, and how widespread
and interesting its applications are. Only years of use and critical
examination" can answer such questions. Henceforth, I will not
argue in favor of the theory, but will concentrate entirely on its
development.
3. Russell"s Arguments against Meinong
I want to return to Russell's historical arguments against Mei-

nong, because although I think they are not decisive against the
theory, they do put severe constraints on an account of non-
existent objects, and Meinong's responses contain some provoca-
tive ideas.
The first objection has to do with impossible objects, such as
the round square. The problem, Russell says, is that such objects
"are apt to infringe the law of contradiction" ([OD]). Russell
does not say explicitly how this is supposed to happen, but based
on certain of his comments, we can reconstruct the following line
of reasoning if
-(I) The round square is round. Granted.
(2) The round square is square. Granted.
(3) (x) (x is square to '" (x is round)). ?
(4) ~ (The round square is round). From (2), (3),
(5) The round square is round &
"" (The round square is round). From (I), (4).
The argument, of course, is not good unless the quantifier in (3)
is construed broadly enough so as to include the round square in
its range. But in that case, why should (3) be accepted?
6. I am using here some terminology from symbolic logic, The sign ,~'
is
is read 'it is not the case that', so line (4) says 'It not the case that the round
square is round'. The '(x)' is the universal quantifier (see the introduction,
note 6), and '::>' stands for the English complex connective 'if . . . , then
. . .". So line (3) reads: 'Everything is such that if it is square, then it is not
the case that it is round', or, more simply, 'Everything that is square is not
round', The ampersand stands for 'and'.
MEINONG iJ·ND MOTIVATION· 39
-It has been suggested to me (in .conversation) that part of

the meaning of 'x is square' is 'it is not the case that x is round'.
If that were true, I would be inclined' to reject (J). But I don't
think it is true. Even if one could define a word, 'square', so as to
include 'not being round' in part of the definition, such a defini-
tion would be artificial and not in keeping with the meaning of
the word as normally used. If being square implies not being
round, it is because of some geometrical principles that connect
the two notions, and net simply because of the definitions alone.
(At least this is true of the customary definitions of' 'square' and
'circle' that usually occur in dictionaries and geometry books;
'round' is much vaguer.)
A more 'plausible suggestion is that. the properties of "being
round and being square are interrelated by the principles of
Euclidean geometry, and that (3) is a thesis of that theory. But is
that so?
Historically, Euclidean geometry developed as a formaliza-
tion of notions with empirical content. Eventually, the idea grew
that the interrelations of these notions that had been formalized
were not merely contingent, but were somehow necessary. And
finally, there was a move away from these issues by the innova-
tion of viewing Euclidean geometry (and other geometries) as an
uninterpreted calculus-a pure mathematical theory in its own
right, without regard for the interpretation of its primitive terms..
This most recent development is simply not relevant here.
If 'round' and 'square' are uninterpreted terms from an uninter-
preted formal theory, then 'there is a round square' is not even
an assertion, but only a string of symbols .awaiting an interpreta-
tion, and likewise for (3). An issue only arises when 'round' and
'square' are construed as symbols with actual content, as in the
earlier stages of Euclidean geometry. But it seems to me that the
early stages of Euclidean geometry did not concern themselves
with all objects, and so they are not relevant to the truth of (3),
with its broad-ranging quantifier. The original developments of
geometry, we are told, dealt with the practical considerations of
land survey following periodic floodings of the Nile; here it is
actual square and round areas that are in question. Philosophers
then tended to broaden the scope of its principles; but even here
there were limits. The most famous example is Kant, who at..
tributed the necessity of the principles of Euclidean geometry to
the structure of human spatial intuition; these. principles were
held to be somehow constitutive of this faculty, and therefore
must govern any object of which we can have a spatial intuition.
But even if he were right about this, the claim would only be that
any such object is, if square, then not round. As Meinong pointed
out, many objects are such that we cannot have an intuitive ap-
prehension of them; the round square is one of these. (In a .sense,
these objects are not conceivable.) Premise (3) would not apply
to them. (Note: Meinong granted Kant's claim that we cannot
intuitively apprehend a round square ([EP] P: 21), but I think he
may have been hasty about this. It seems to me that. dreams and
some drug experiences may provide examples, of spatial intuition
that defy description in terms consistent with the laws of Euclid-
ean geometry. The orthodox conclusion is that such intuitions
are not of objects at all, but one might equally well conclude that
they are sometimes of objects that do not obey the laws ofEuclid-
ean geometry.7)
Since there has not been a serious attempt to establish (3) in
its complete generality, I suggest that the evidence for it is
meager, and there is no need to accept it. Not- all square objects
fail to be round; the round square provides an example of this.
However, I suspect that this leaves the average reader somewhat
ill at ease. There is a feeling that one cannot understand a given
property (say squareness) without understanding how it excludes
other properties (e.g., roundness), and trying to view squareness
as not excluding roundness undercuts our understanding of what
these properties are. But perhaps the following observation will
help: being square does exclude being round for real objects
7. See R. L. Gregory's discussion of perceptual illusions in [E & B]. Here

is a sample quote (pp. 107-08):
If the after-effect from the rotating spiral is examined carefully, two

curious features will be noticed. The illusory movement may be para-
doxical: it may expand or shrink, and yet be seen not to get bigger or
smaller, but to remain the samesize andyet to grow. This sounds impossible,
and it is impossible for real objects, but we must always remember that
what holds for real objects may not hold for perception once we suffer
illusions .
(indeed, for all possible objects), and it is primarily with real

objects that we are normally concerned. There is a long tradition
according to which reality must be coherent in certain ways, the
exclusion of roundness by squareness being one of these. As for
why reality should be so coherent, that's a question I don't know
the answer to. In any event, since we primarily concern ourselves
with real objects, we instinctively make the judgments of exclu-
siveness of properties (such as round and square) that are ap-
propriate to real things. It is only natural that we should feel the
force of those instincts when we attempt to discuss unreal things
as well, and feel somewhat disquieted when we are tempted to
abandon them (such as in the recollection and recounting of
dreams).
There may be an analogy here to the situation in physics
when relativity theory was proposed. It was a theory which was
extremely difficult to understand in terms offamiliar experiences,
and it remains a theory which holds that there are physical events
which are, in a limited sense, not conceivable. The difficulty with
conceivability is that our "conceptions" or mental picturings
tend to take temporal relations, such as simultaneity, for granted
-a habit that is quite hard to break.
Exercise for the reader." ( I) Form a mental picture of four
events far removed in space. Event A happens before event B
from the point of view of event C, but after event B from the
point of view of event D. Both points of view are equally correct.
Make sure that you do not include in your picturing that A
happens absolutely before (or after, or simultaneous with} event
B. (2) Ifyou succeeded in carrying out (I), form a mental picture
of a round square; otherwise, just relax,
Of course, relativity theory kept in touch with ordinary
experience in .. this sense: when applied to neighboring objects
moving at medium speeds, for example, relativity theory approxi-
mates to classical mechanics. In much the same way, the theory
of unreal objects will agree with the orthodox view when applied
to. "familiar" situations-that is, to situations involving only
existing objects. The exclusiveness of roundness and squareness
is a special case of this; the general argument will be made in
chapters 5 and 6.
Before passing to Russell's second objection, let me sum-
marizethe view being taken of impossible objects, Let us call an

object which does infringe the law of contradiction (i.e., which
satisfies some formula of the form 'x is A & "" (x is A)') a con-
tradictory object. Then there are no contradictory objects (for if
there were, some contradiction would be true). The round
square is not a contradictory object; arguments that it is appeal
to some principle such as (3), which is true if its quantifier is
restricted to real objects, but not if its quantifier is read with
complete generality. The round square is impossible only in the
sense defined in chapter I : It has properties which no real object
could have. 8
Russell's second argument is this: consider the existent gold
mountain; by the satisfaction principle it is golden, is a moun-
tain, and exists. So some gold mountain exists, which is plainly
false.
Meinong's answer was surprising. The existent gold moun-
tain is existent, he said, but it does not exist (Meinong [USGSW]
sec. 3). The former notion (being existent) is part of the "so-
being" of an object, the latter part of an object's "being"; the
existent gold mountain has being existent as part of its so-being,
but it lacks existence as part of its being. (Strictly, it has no being.
at all.) Russell replied that he saw no difference between being
existent and existing ([lg07 Review]), and one can easily sym-
pathize with him here. But I think something can be made of the
distinction. Suppose that being existent is a nuclear property, and
that existing. is an extranuclear one (the one that 'exists' stands
for, and that was discussed in chapter I, section 2) . Then, ac-
cording to the theory discussed in the last chapter, there will be
an object which has exactly the three properties: being golden,
being a mountain, and being existent; but it will not exist, any
more than the gold mountain exists.
8. We might call the round square a geometrically impossible object, since

it is geometrical principles that preclude the existence of anything with its
properties. A question that. naturally arises here is' whether there are any
logically impossible objects, objects that have properties that it is logically im-
possible for an existent to have. That depends on what you count as logic. An
example might be the "nonsquare square," in which the prefix 'non-' signifies
nuclear predicate negation as defined in chapter 5, section 2.
ME/NONG AND MOTIVATION 43
This raises other questions: is there a nuclear property, being

existent? And if so, how is it related, if at all, to the extranuclear
property of existing? And further, couldn't the point of Russell's
objection be preserved by a simple reformulation of the example;
instead of 'the existent gold mountain', just consider 'the gold
mountain which exists'?
Let me quickly answer the last question with respect to the
theory being developed in the present work. On that theory, the
phrase 'the gold mountain which exists} does not refer to any
object at all; for ifit did, it would refer toa gold mountain which
exists, and there are none of those. (Strictly, the question has
partly to do with the nonontological, linguistic question of how to
interpret definite descriptions; this will be discussed in chapter 5.)
But what about the other questions? For example, is being-
existent a nuclear property? I'm not at all sure, but the following
answer seems natural, First, there is a nuclear property (probably
many) which belongs to all real objects. This is a consequence ofa
general assumption that I will make about nuclear properties-
roughly, that any class of existing objects is the extension of some
nuclear property. In defense of this general assumption, I can say
only that some such assumptions are needed, on pain of not
having a theory at all. And if we recall that, from the orthodox
point of view, nuclear properties are just ordinary properties of
individuals, and existing objects are all there is, the assumption
is just the "translation)' of the orthodox assumption that every
class is the extension of some property. Of course, many philoso-
phers reject properties altogether, but those who acknowledge
them typically adopt this assumption.
Given that there. are nuclear properties true of all existing
objects, are any of them denoted by 'being existent'? I think this
is mainly a linguistic question, and one that I have no definite
intuitioris about. But since the controversy over whether existence
is a property uses the word 'existence', and since both sides of the
dispute seem to have some claim to plausibility, it seems reason-
able to hold that the reason the dispute goes onis that both sides
are in some sense right. And this would be explained if 'exists' or
its cognates were ambiguous, standing for both a nuclear and an
extranuclear property, neither of which alone exhausts the way
in which we use the word. Ifso, it seems convenient to use 'being

existent' and 'existing' to mark the distinction, and I'll do this for
the remainder of this section (though later I'll just use 'is existent'
as a synonym for 'exists').
So according to the theory being developed, Meinong was
right in his reply to Russell: there is such an object as the existent
gold mountain, and it is existent (and golden, and a mountain),
but it does not exist. But what is the relation between being exist-
ent and .existing? Meinong attempts to answer this question in
ObeT Miiglichkeit und Wahrscheinlichkeit, where he calls the nuclear
property a "watered-down" version of the extranuclear one (see
Findlay [MTO V] chap. IV). Existing is not the only extranu-
clear property which has a watered-down version; being possible
also has one. This tempts one to wonder if all extranuclear prop-
erties have nuclear watered-down versions. That will depend, of
course, on what 'watered..d own' means. In Meinong's theory it is
not clear (at least to me). He speaks of the watered-down version
of a property as got by removing the "modal moment" from
"full-strength factuality." I am not sure what this means.
We know at least this much about the relation between a
property and its watered-down version: if p is a watered-down
version of P, then (1) P is extranuclear, (2) p is nuclear, and (3)
it's hard to tell the difference between p and P. We can be more
explicit about (3), by suggesting: (4) necessarily, any real"object
has p if and only if it has P. In the theory given below, for every
extranuclear property P, there will be at least one nuclear prop-
erty p which is related to P as in (4). Such nuclear properties will
be important in applications of the theory-for example, in ap-
plications to fictional objects.
4- Intentionality
So-called "intentional" idioms have been much discussed in the

recent philosophical literature (see Chisholm [P]). In this sec-
tion I will review those points on which there is the most agree-
ment, and say how such idioms will be used in the theory being
developed 4
Intentional idioms fall into at least two categories, First,

MEINOJYG AND MOTIVATION 45
there are those which take "prepositional" objects:

believe: Everyone believes that snow is white.
want: Smith wants people to be nice.
think: No one thinks Jones is a spy.
Second, there are those which take "ordinary" direct objects:
lookfor: Many people are looking for the cure for
cancer.
dream about: Smith is dreaming about a unicorn with
purple stripes.
conceive of: No one can conceive of a roundsquare.
Meinong devoted some effort to getting people to acknowledge
that in the first sort of case-cases involving belief, desire, etc.-
there are things that are the objects of the mental attitudes in ques-
tion. In the case of belief, he called these objects "objectives."
Objectives are somewhat similar to what we now call propositions
or states of affairs. Since Frege and Russell believed in similar
entities, and since they were much more influential than Mei-
nong, this view is not often thought of as one of Meinong's con..
tributions.
There is at least one difference between Meinong's objectives
and modern propositions: both true and false propositions are
supposed to exist, but only true objectives are supposed to exist;9
false objectives furnish another example of nonexistent objects.
At one time, Russell and Moore worried a great deal about how
something false could be said to exist, but this does not often
worry people nowadays.
The main evidence that there are propositions is that they
seem to be required for the truth ofsuch claims as :
Mary believes only some of the things she was taught as a
child.
9. I am using 'exist' here in a broader sense than Meinong; for compari-

son with his theory read this use of 'exist' as 'exist or subsist'. For Meinong,
objectives never exist, but some of them subsist-the true ones (see [TO]). Also,
he would not use the word 'true' in connection with objectives, but rather the
word 'factual'.
In the present work, I will suppose that there are propositions

(and that they are objects of belief). I t will not be necessary for
me to take sides on the issue of whether they all exist, or only
the true ones, or perhaps whether they all constitute examples of
nonexistent objects. I'm not sure whether this. is a substantive
question at all, and I don't want to divert attention from my.
major task.
Intentional idioms of this sort manifest a de re/de dicto ambi-
guiry.I? The sentence 'Agatha believes the round square is round'
has a de dicta reading in which it "is asserted that Agatha stands in
a certain relation to the proposition that the round square is
round, This reading does not require for its truth that there be
a round square. The de re reading can be paraphrased as 'The
round square is such that Agatha believes it is round'. This at-
tributes to the round square a certain extranuclear property: the
property of "being believed by Agatha to be round. It requires for
its truth that there be a round square (though not that there
exist a round square); at least, this is how it will be treated in the
present theory. The details of this treatment will be given in
chapter 5.
The second category of intentional idioms is less well under-
stood, despite the fact that these idioms figure centrally in the
10& The terms 'de re' and 'de dicto' translate roughly as 'of the thing' and
'ofwhat is said'} and they are used to mark certain aspects of meanings of certain
English sentences. For example, on its most natural construal, the sentence
'Agatha believes that everyone will be late' says that Agatha stands in a certain
relation (the belief relation) to a certain proposition (a certain "thing that is
said")-namely, the proposition that everyone will be late. The whole sentence
is called a de ditto beliefsentence. An example ofa dere beliefsentence is 'Agatha
believes ofJolm that he is unfaithful', in which Agatha is said to be related by
belief to John himself (to a H thing"). The utili ty of the terms is that they pro-
vide a brief way of resolving certain ambiguities. For example, the sentence
'Agatha believes that a man I know will help me out' is said to have two read-
ings: the de dicto reading, which may be paraphrased as 'Agarha believes that
the following proposition is true: some man whom I know will help me out'
(this reading does not require that Agatha have any idea at all regarding which
man will help me out); and the de re reading, which may be paraphrased as
'with regard to a certain man whom I know, Agatha believes that he will help
me out'. Other examples ofde re/de dictaambiguities will be discussed off and on
in the text.
MElNONG AND MOTIYATION 47
articulation of many famous philosophical issues. For example,

'conceiving' is used in the questions 'Is there-such a thing as that
than which no greater can be conceived?', 'Can one conceive of
an unconceived-of thing ?~.
Sentences involving intentional idioms of the second cate-
gory are typically-ambiguous, and many of the traditional philo-
sophical issues involving them seem to trade on this ambiguity.
Let me use 'looks for' as a philosophically neutral example, and
consider:
Samantha is looking for a cow.
The ambiguity is between the sense in which Samantha is looking
for some particular cow, perhaps the one she has just lost, andthe
sense in which she is just out cow hunting, perhaps because she
wishes to augment her herd andjust hasn't found a cow yet that
re
is suitable. I'll call the first reading the de reading, and the
second the de dicta reading (this may be a controversial use of
these terms). There is a tendency to call the first reading the
"definite" reading (since there is a definite cow Samantha is
looking for) and the latter the "indefinite"; but these terms get
confusing when the direct object of the verb is itself a definite
description, as in:
Samantha is looking for the best milk cow in the county.
The same ambiguity appears here as above. On the de re reading,
there is a specific cow that is being searched for, say Esmerelda,
who has already won five blue ribbons; whereas, on the latter
reading, she has no specific. cow in mind: she merely wants to end
up with the best, whichever it is.
Notice that, given the de re construal, the singular term is in
an extensional context, whereas, on the de dicta construal, it is not.
If the best milk cow in the county is Bossie, Samantha's own cow,
then, on the de re reading, it follows that Samantha is looking for
Bossie, while, on the de dicta reading, this does not follow. I am
not at all sure how to treat the de dicta readings of these idioms. I
suppose that, as with other nonextensional contexts, we should say
that the words in these contexts refer to their customary senses, or
something like this) but the logic ofsuch contexts is not at all easy
to work out. The best treatment I know ofis due to Montague in

his paper "The Proper Treatment of Quantification in Ordinary
English," but this treatment involves complications that I would
like to avoid here (see also Montague [PIL] p. 124). So I will
simply ignore the de dicto readings of these idioms. I will have
(extranuclear) relation words which represent the de re readings
ofsuch verbs.
One reason for fussing so much about these idioms is that
certain of them, particularly 'thinks about', 'dreams about", and
'writes about' stand for notions that get discussed under the title
of 'directedness toward an object'. Brentano is famous for the
view that mental acts can be "directed toward" objects even
when such objects don't exist. Although Brentano eventually al-
tered his views about the nature of this relation, Meinong took
over the doctrine wholeheartedly, with only an occasional qualm.
But the "doctrine" is subject to various interpretations. Meinong
is usually, I think, credited with the view that in any true sentence
of the form 'x is thinking about the <ft', even when construed de
dicto, the term 'the if/ must denote an object (though perhaps one
that doesn't exist). This view is, unfortunately, untenable, for
reasons that were given in section I of this chapter. It may be
true, de dicto, that Abigail is thinking about the gold mountain
that exists, but there is no such object. At least, that is the view of
the theory I am developing.
For related reasons, I have tried to avoid the phrase 'object
of thought'. I think it is customarily used to mean something like
'referent of the direct object of the verb 'to think". This phrase
tends to confla ~ the de re/de dicto ambiguity inherent in uses of the
verb, and also tends to smuggle in the presupposition that any
denoting phrase that can follow 'to think' must refer,
3
A Sketch oj a Theory oj Fictional Objects
In chapter I, I sketched a theory of nonexistent objects, and in

chapter 2 I discussed the motivation for that theory. Some of
that motivation concerned fictional objects; these form the sub...
ject matter of the"present chapter.! By 'fictional' I do not mean
'nonexistent', but rather 'occurring in fiction) ; as will be seen, I
think that real objects occur in fiction as well as nonexistent ones..
The theory offictional objects will be developed in this chapter only in a
crudeform; I will return to the subtleties of the theory in chapter 7.
The theory comes in two stages.. First, we need a stock. of
objects from which to select the fictional ones; this has been pro-
vided for in chapter I. But we need more than this; given a body
ofliterature, we need to know what fictional objects occur therein.
So far, no method has been given for telling-for example, which
nonexistent object of the theory of chapter I Sherlock Holmes is.
.The point of this chapter is to sketch such a method. But first, a
sligh t digression.
I. "Literary Theory
As ~ear as I can tell, most works in literary theory do not con-

cern themselves with the ontological issues concerning fictional
I. Fictional objects have always been problematic from the orthodox
poin t of view. As early as 1938, Wisdom noted (regarding the attempt to
paraphrase away reference to dubious entities): "The extra entities in the uni-
verse of discourse all went up-in smoke; though from the fictional entities there
lingered still a peculiar smell" (quoted in Urmson [PAl p. 150).
49
characters that I will be discussing.. Authors occasionally do

observe, in introductory passages, that characters of fiction do not
exist. For example, Wellek and Warren remark: "A character in
in
a novel differs from a "historical figure or a figure real life. He is
made only of the sentences describing him or put into his mouth
by the author. He has no past, no future, and sometimes no con-
tinuity of life" ([TL] p. 25)* (Their first sentence expresses the
substance of what I mean by saying that a character in a novel
does not exist.) These observations, however, are prefatory to
getting on with the real issues of literary theory: style, genre,
image, metaphor, myth, the grounds and nature of evaluation,
and so on. One of these issues is sometimes expressed using words
like 'rear: the issue of how "real" certain characters are (or
ought to be). But this notion of "real" is more akin to "lifelike" or
"vivacious"; it is a notion that, unlike existence, comes in de-
grees. (Certain writers tend to avoid the term, and speak instead
of the issue of realism, and of the "psychological truth" of the
behavior of fictional figures (e.g., Wellek and Warren [TL] p.
92; see also pp. 213-14) . ) By and large the going view seems to
be as follows:
(A) Fictional characters typically do not exist.

(This is often presumed, sometimes stated, and rarely taken to be
a controversial issue.)
(B) Nontheless, there are fictional characters.
(This is too obvious to require statement.) And:
(C) Their nonexistence does not diminish their literary
value.
The theory I am discussing agrees with (or is at least consistent
with) all of these.
I have sometimes remarked that fictional characters do not
exist, and had people reply: "But they do exist, they exist in
fiction !" I assume that some readers will have this same reaction,
and so let me emphasize that although the theory under discus-
sion denies that (many) fictional characters exist, it does not deny
that they exist in fiction. Existing in fiction and existing are quite
different things, and one may do the former without doing the
SKETCH OF A THEORY OF FICTIONAL OBJECTS 51
latter, just as one may commit adultery in the heart without

thereby committing adultery. (Matthew 5+~8 may indicate that
both are sinful, but not that doing the former entails doing the
latter.)
2. Immigrant and Native Objects
W~th regard to a given story, it's helpful to distinguish two differ-

ent sorts of fictional objects: objects native to the story versus
objects that are immigrants to the story. With regard to the Conan
Doyle series, Sherlock Holmes provides an example of an object
native to the story, whereas the city London provides an ex-
ample of an immigrant to the story. The distinction is, roughly,
whether the story totally· "creates)' the object in question, or
whether the object is an already familiar one imported into the
story. The word 'create' here is meant in the sense in which an
author is commonly said to create a character. It does not mean
'bring into existence', for such objects typically do not exist.
Perhaps 'create' is a bad word, but it is customarily used in the
sense I intend.
Whether an object is native or immigrant depends on the
story in question. Sherlock Holmes is a native object of the Conan
Doyle novels, 'but he is an immigrant to many other stories. For
example, if I write a story about a detective convention, and
discuss the attendance of Sherlock Holmes, Hercule Poirot, Miss
Marple, and others, then these objects will be immigrants to my
story. London, Richard. Nixon, and the present queen of Eng-
land, being real objects, will all be immigrant objects in any
ordinary fiction in which they occur.f
The point of my making the distinction between native and
immigrant objects is this: I'm only going to sketch a theory to
account for native objects of stories, objects "created" by stories.
Immigrant objects, such as London in the Conan Doyle novels,
do not automatically raise ontological difficulties anyway, and
when they do (as when Holmes appears in later stories by other
authors), it is better to deal with the problems at their source.
Portions of this section and the next contain material that appeared orig-
inally in Parsons [MAFO].
2. This is because objects created in ordinary fiction are always (or prac-
tically always) highly incomplete (see below in text).
But first, let me deal with a qualm that some readers may
have. Some people hold that there are no immigrant objects in
the sense in which I have used this term. They hold that when we
say, for example, that Sherlock Holmes lived in London, we are
not referring here to the real London, England, but rather to
some other city, a fictional city that doesn't exist.. (Presumably
they think this because the novels say things about London that
are not true, though it is not clear to me why one should generally
expect the sentences of a fictional work to be true.) I am inclined
to think that this view is wrong, and that when we say that
Holmes lived in London, we intend to refer (and do refer) to the
real London, England. But if I am wrong about this, it will make
very little difference to the theory being sketched. It will mean
only that I have provided a place in the theory for immigrant
objects which needs to be filled by a different kind of object, the
kind that will be discussed in section 4 under the title "Surrogate
Objects." With this qualification in mind, I will continue to speak
as if the account of immigrant objects given above is correct.
34 The Theory
As indicated above, the theory is already half-given: typically,

when we appear to be referring to fictional objects, we are doing
so. With an abundant ontology at hand, the fact that such ob-
jects don't exist needn't be a problem. But now we want to know,
for example, which object Sherlock Holmes is. As a help in an-
swering this question, suppose we begin by looking at some of the
things that are generally supposed to be true of him:
(i) .He's a detective.
(ii) He lived at 22 r B Baker Street.
(Hi) He doesn't exist.
(iv) He is a fictional detective.
(v) He has been thought about by many people.
(vi) He is admired by many real detectives.
The first two. of these claims are examples of nuclear predications
and the rest are extranuclear predications. I think that the dis-
tinction is important here, in that the extranuclear predications
provide more trustworthy data for a theory of fictional objects
SKETCH OF A THEORT OF FICTIONAL OBJECTS 53
than the nuclear ones. And this is because it is not at all certain
that the nuclear ones are true. They are, in fact, controversial,
and the reasons adduced on both sides of the controversy are not
very persuasive.
People who argue in favor of the truth of claims like (i) and
(ii) commonly cite reasons like these f
(a) If you are taking a literature test and are asked whether
or not Sherlock Holmes was a detective, the correct
answer is yes, despite the fact that both you and the
teacher know that Holmes doesn't exist.
(b) If I bet you that Sherlock Holmes lived at 22IB Baker
Street, and you bet he didn't, then we check the text
and I win, again in spite of the fact that we both agree
that he doesn't exist.
The trouble with these reasons is that they may not be directed at
the predications in (i) and (ii) at alL This is because we sometimes
use sentences like 'Sherlock Holmes was a detective' as mere short-
hand for claims that would be more literally expressed by'Ac-
cording to the story, Sherlock Holmes was a detective'. This is a
common linguistic practice, and it is not confined to discussions
of fiction. For example, in discussing Plato's theory of forms, we
sometimes lapse into saying things like 'The forms are self-predi-
cational', when we clearly do not mean to endorse the claim that
the forms are self...predicational, but rather to say simply that,
according to the theory, the forms are self-predicational. And in
the discussion of fiction, the practice is not limited to points about
fictional objects; in discussing Camus, we might say something
like "Human life is pointless," meaning that, according to
Camus's writings, human life is pointless. Unless we are sure that
literature tests and bets are addressed to the literal readings of
(i) and (ii), as opposed to other claims that are sometimes made
using such sentences, we cannot put too much faith in the reasons
cited above. And I think that we are not sure of this.
On the other hand, people who argue that (i) and (ii) are
3. See Blocker [TAFE] and Woods (LF]. Others-for example, Martin

and Schotch [MFN]-assume the truth of claims of (i)-(iii) without argument.
My criticisms of the arguments parallel discussion in Crittenden [TANB],
Devine [LOFJ, Lewis [TIF], Ryle [10], "Voiterstorff [WWA], and others.
not true usually cite reasons such as these: although (i) and (ii)
seem to be true, they can't really be true, since Sherlock Holmes
doesn't exist, and so he cannot have any properties at all. It will
be clear why I "find this reasoning unpersuasive: nonexistence
does not preclude the having of properties. I think it would be
nice to have a theory that preserved claims like (i) and (ii), but it
is not compelling.
There is no similar reason to mistrust (iii)-(vi). We certainly
don't believe these claims because we confuse them with what
is true according to the story. For example, we don't confuse
'Holmes doesn't exist' with 'According to the story, Holmes
doesn't exist', since we believe the former arid disbelieve the
latter. The same holds for (iv). And while (v) and (vi) agree with
what the story says, we might still believe them if the story had
said that Sherlock Holmes-was a hermit, unknown to most people
and disliked by all who knew of him..
Of course, none of this determines a theory of fictional
objects. Questions arise on two levels: what the data are, and how
to account for them. It's not even clear that (i) and (ii) are part
of the data, whereas it's pretty clear that (iii)-(vi) are. Still, even
if we accept all of (i)-(vi) as data, many different theories might
account for these. The theory I will discuss here will be the one
that strikes me as the most interesting one which can be given
within the framework of the theory of objects outlined in chap-
ter I.
The theory I propose is roughly what you get by supposing
that native fictional objects have exactly the nuclear properties
that we are naively inclined to apply to them. For example,
Sherlock Holmes is a detective, solves crimes, lives in London,
etc. Within the context of the ontology of chapter I, this deter-
mines uniquely which object Sherlock Holmes is. Specifically:
Sherlock Holmes = the object which has exactly those
nuclear properties which he is under..
stood to have in the Conan Doyle
novels.
This is actually a consequence of two proposals. The first identi-
fies Sherlock Holmes as a native of the Con an Doyle novels. This
is not a result of any particular theory, but just a matter of our
SKETCH OF A THEORr OF FICTIONAL OBJECTS 55
common understanding of the name 'Sherlock Holmes', (Holmes

appears in stories by other authors, but there only as an im-
migrant object.) The second proposal is a general schema for
identifying native fictional objects, given the relevant" body of
literature: .
The t/J of story s = the object which has exactly those nu-
clear properties that the t/J has in s.
The right-hand side of this identity is to be understood as con-
taining a de re ·predication; spelled out in more detail, it is as
follows:
(*) The ifJ of s = the object x which is such that for any
nuclear property p, x has p i~ and only if,
the <p of s is such that in s it has p.
The phrase following the last comma is intended, when filled in,
to be English, and to be of the form that English speakers are
able to understand and to judge. So grammatical changes that
are merely stylistic or that are required, say, for purposes of
gender agreement are understood as allowable. For example, if
(*) is applied to 'the detective of the Conan Doyle novels who
lived at 22 I B Baker Street' and if we are testing to see if cleverness
is one of its nuclear properties, then we will ask whether:
The detective of the Conan Doyle novels w-ho lived at 22IB
Baker Street is such that in the Conan Doyle novels he is
clever.
Notice that I am not here trying to define the locution 'the ifJ of s',
Instead, I take it that this is a form of words that we English
speakers understand; the point is rather to relate particular
phrases of this form to objects specified in theoretical terms.
Instead of putting a description for rp, we can also use a
name. For example, one application of (*) is to 'the Sherlock
Holmes of the Conan Doyle novels' ; given the de re reading on the
right-hand side of (*), this should give exactly the same results
as 'the detective . . .'; namely, both should yield an object
which is a detective, solves crimes, lives at 22 IB Baker Street, is a
friend of Watson's, and so on. So Holmes has the nuclear prop-
erties commonly ascribed to him. He will also turn out to be an
incomplete object, since he has neither of the nuclear versions of

these properties:
having a mole on his back;
lacking a mole on his back.
He lacks both these properties since neither of them is such that
he has it in the story. Being incomplete, we know for sure that he
doesn't exist.
One word of warning: when I speak of an object's having a
property "in a story," I mean something like having a property
according to a story I am not thinking of a story as a possible world,
4
with fictional objects being "in" it, as is sometimes discussed in

possible worlds theory. Nor should '.x has p in s' be read as '» has
p and x is in s', although this may be suggested by use of the word
'in'. Although, according to the theory given here, objects that
are native to a story have a nuclear property if and only if they have
it in that story, this is not also true of immigrant objects, nor is it
true (in the case of either native or immigrant objects) for ex-
tranuclear properties. In the Conan Doyle stories, the immigrant
object, London, has certain nuclear properties which in fact it
doesn't have-for example, the property of having had a certain
sort of crime committed in it. And in the story, Holmes has cer-
tain extranuclear properties, such as existence, which he in fact
lacks.
Not every way of filling in for <jJ and for s in (*) will give a
fictional object. For example, suppose we apply (*) to 'the purple
dragon of the Conan Doyle novels' ; then (*) identifies this as the
object which has exactly those nuclear properties attributed to
the purple dragon of the Conan Doyle novels by those novels 4
But there is no such object. My intent is that in such cases (*) not
be used. (Ifit is used, it is worded in such a way that ifit works at
all, then it identifies the purple dragon ofthe Conan Doyle novels
with the null object; this has a certain theoretical neatness, but I
prefer that instead of this it simply not be used at all. That is, (*)
is intended to presuppose that 'the ifJ of s' refers to an object;
when the presupposition fails, («) should not be used.)
Another more interesting case is this: suppose that a novel
says at a certain point "A crowd gathered,' and suppose that
nothing more is said about the crowd, Then none of the members
SKETCH OF A THEORr OF FICTIONAL OBJECTS 57
of the crowd will be identifiable as fictional objects of the novel.

Indeed, there is no way to apply (*) to any crowd members, for
no choice of 'if>' will pick one out. It seems to me that at an in-
tuitive level we don't think of any member of the crowd as a fie...
tional object, and that the theory is right to ignore "them." This
issue will be discussed more in chapter 7, as well as:
How do you tell what is true "in" a story?
How do you decide what story to usei'.
What happens when an author contradicts himself?
I've concentrated on two objects: Sherlock Holmes and London.

Holmes is native to the Conan Doyle stories, and is subject
to schema (*).. I have been assuming that London, the real
London, England, is an immigrant object of those stories. It is
not subject to schema (*); if it were, (*) would attribute to it all
of the nuclear properties which it has according to the story-
and some of these are nuclear properties which it does not have.
Now there are those who think that the real London does not
appear in those stories, but rather that another object does; it's a
fictional object, called 'London' in the story, and it is different
from the real London. Let me call it a "surrogate" of London. I
think these people are half right. That is, I think that the real
London occurs in the stories, but that sometimes in discussing the
stories we discuss its surrogate instead of it.
How do we tell what objects occur in a given piece of fic-
tion? Well, that depends in part on what we mean by 'occur in'.
I'm not sure that there is any consensus on this, but what I mean
is something like:
x occurs in s if, and only if, s says something about x.
If pressed to clarify the right-hand side of this, I would give the
following:
x occurs in s i~ and only if, there is some property P such
that, according to s, x has P.
Now the reason I am inclined to say that the real London occurs
in the Conon Doyle novels is that I think it is true that London is
such that, according to the novels, Holmes lived in it. Of course

this involves a de re predication about London, and these are
notoriously difficult to evaluate. I think I am inclined to accept
it because I see no difference in the referential situations:
(i) Telling a lie about Jimmy Carter.
(ii) Telling a lie about Carter which is very long (e.g.,
book length).
(iii) Making up a story about Carter which is not intended
to deceive anyone, and which contains falsehoods.
(iv) Writing a work of fiction in which Carter is a character.
It seems clear to me that there are cases of (i) which can be re ...
ported as "Carter is such that according to what Parsons said he
is P," and at least some cases of (ii)-(iv) seem just like (i) in this
respect.
So I think that there are immigrant objects. But in discussing
literature, we sometimes discuss other objects. A signal that this
is happening is that we no longer use the singular terms 'London',
'Carter', . . . , but switch to locutions such as 'the London of the
novels', or 'the London created by Conan Doyle', or'Sherlock
Holmes's London'. Here we do not refer to London, but rather
to what I have called its surrogate. The surrogate of London fits
into our theory as follows: it is the object that is picked out by an
application of (*) to 'the London of the Conan Doyle novels'. It,
like Holmes, is an object that is native to the novels; it is a city
"created" by Doyle (with the aid of our common understanding
of the real London). So the London of the novels will be an in-
complete object, and will also be a nonexistent object, Any im-
migrant object will have a surrogate; if'N' names an immigrant
object, then 'the N ofs' names its surrogate.
To summarize my view here: there are immigrant objects,
and they occur in novels. An example is London, in the Conan
Doyle novels. There are also surrogate objects, and we sometimes
discuss them, but they do not occur in stories (in the sense of
'occur in' discussed above). If 1 were convinced that I was wrong
about immigrant objects occurring in stories then I would prob-
ably hold that surrogate objects occur there. But it would defi-
nitely be wrong to hold that both an immigrant object and its
surrogate occur in a novel; when we say that, according to the
SKETCH OF A THEORT OF FICTIO}fAL OBJECTS 59
novels, Holmes lived in London, we refer either to the real Lon-

don or to its surrogate, but not to both.
5. Relational Properties
This is a natural place to discuss some intricacies of rela-

tional properties. Let us suppose, following Woods, that the
Conan Doyle novels specified that Holmes met and talked with
Gladstone. Then, according to the theory presented above,
Holmes had the property of having seen Gladstone. On the
other hand, it doesn't seem to be true of Gladstone that he had
the property of having been seen by Holmes (see Woods [LF]
Ill, 5-7, v. 7). And this contradicts certain assumptions that
we are normally inclined to make about relational properties.
First, we normally assume that, given a two-place relation R,
we can make one-place properties by "plugging up" either place
of R. Let me use 'x[Ry]' to symbolize 'x has the property that
you get by plugging up the second place of R withy,' and '[xR]y'
to symbolize )1 has the property that you get by plugging up the
first place of R by x'. Then the second assumption that we nor-
mally make is that for any x, y, and R: x[Ry] jf, and only if,
(xR]y. But now if we use 'h' to name Holmes, 'g) to name Glad-
stone, and '8' to stand for the relation of seeing, we seem to have
a counterexample to the second assumption: 'h[Sg]' is true, but
'[hS]g' appears to be false.
Actually this situation appears already in the ontology of
chapter I; it is not peculiar to fictional objects. By the theory of
chapter I, there is an object which has exactly these nuclear
properties: it is a giraffe, it is now in the corner, and it sees the
gold mountain (where "the gold mountain" is taken to be the
object which has exactly two nuclear properties: goldenness
and mountainhood). Tb"en the giraffe has the property: seeing
the gold mountain; but the gold mountain does not have the
property: being seen by the giraffe.
All this could be avoided by denying that there are any
relational properties, but this would lead to a theory of properties
that is probably too weak to be interesting. Instead, I am going
to suppose that there are relational properties, and also that the
equivalence mentioned above does not always hold. Of course,
if both terms of a relational statement are real objects, then the

equivalence of 'x[Ry]' and '[xR]y' will hold. This is in keeping
with my resolve that the theory given here will "reduce to" the
orthodox one in all cases in which only real objects figure. So
it will be a principle that if both x andy exist, then x[Ry] iff [xR]y
(this will be a theorem of the theory of chapter 4). I'll assume
that when we make an ordinary relational statement-saying
simply that x bears R to y-that this is true if and only if both
x[19'] and [xR]y. For real objects all three forms are equivalent,
but for unreal objects they are not.
Returning to the Holmes/Gladstone case, is it certain that
Gladstone lacks the property- of being seen by Holmes? Might
we not bear all sorts of interesting relations to unreal objects
without noticing it (perhaps because if their unreality)? Do real
objects ever stand in relations to unreal ones? I am not at all sure
what the best line is to take here. But among those writers who
have taken the issue seriously, there seems to be an impressive
consensus: real objects can never bear relations to unreal ones .
(Of course, the writers in question implicitly or explicitly restrict
themselves to nuclear relations here; real objects can admire
unreal ones, contemplate them) be more or less famous than
they, etc.) I am not sure why this should be true, but it seems
best to abide by the judgments of the majority here, and besides
it makes for a neat theory. So the theory to be given will require
that no real object ever has a relational property that is obtained
by plugging up one end of a nuclear relation with an unreal
object, and thus no real object will bear any nuclear relation to
an unreal one" This, too, will be a theorem of the theory of
chapter 4-
Returning to Holmes and Gladstone, then: if the novels.
had gone as indicated, Holmes would have the property of having
seen Gladstone, but Gladstone would not have the property of
having been seen by Holmes, and the bare statement: -
Holmes saw Gladstone,
would be false (though true in the story).
11
FormaI Development
4
The Language (f)
My plan in this chapter is to develop a very simple artificial

language within which some of the theses of the theory of objects
can be formulated. The use of such a language imposes a certain
kind of rigor: it allows some theses to be ve.ry clearly stated and
some to beproved from more basic assumptions, thereby increas-
ing our understanding of the basic assumptions. It also adds to
our understanding of some issues to see what cannot be proved
from the basic assumptions.
This chapter is devoted entirely to a study of the artificial
language (!). This is a very primitive language. It does not con-
tain, for example, any means of forming definite descriptions; it
does not contain any verbs of propositional attitude (such as
'believes'); and it does not contain any modal operators. All
these will be added in chapter 5.
The plan is as follows. In section I the syntax of {!} is de-
scribed. In sections 2 and 3 rules and axioms are stated, and the
notion of semantic interpretation is characterized and discussed.
In section 4a completeness theorem is stated and its proof briefly
sketched. In section 5 there is a sketch of an "orthodox" lan-
guage, which is then used in section 6 to show how to "rnodel"
the theory ofobjects within the orthodox conceptual scheme. This
leads to -a relative consistency proof for the theory of objects. It
also suggests a way in which the import of f!} might be captured
within the orthodox conceptual scheme; this is discussed in sec-
tion 7. If this chapter is being read for the first time, it might be
63
64 FORMAL DEVELOPMENT
best to skip ahead to chapter 6, section I, where this chapter

(chapter 4) is summarized.
I. The Symbolism of (!)

The language (!) will resemble the predicate calculus with pred-
icate quantification; the main innovations are the distinction
of nuclear from extranuclear predicates and the notation for
"plugging up" relation symbols with singular terms.
Primitive Vocabulary
There will be signs to stand for objects, for nuclear properties and
relations, and for extranuclear properties and relations; in each
case we will have both constants and bindable variables. These
are:
( I) Singular terms:
Constants: a, b, c, Variables: x, y, Z, Xl,
x2, X3, • • .
(2) n.. . place nuclear predicate terms:

Constants: P¥, p~, p~, . .. Variables: q¥, q~, q~,
(3) n-pl~ceextranuclear predicate terms:

Constants: Py, p~, PR, · ~. Variables: Qs, Q2' Q~,
The superscripts on predicate terms indicate how many places

the property or relation in question has. For example, Qf is a
variable that ranges over two-place extranuclear relations. We
will single out one of the one-place extranuclear predicate con-
stants, say P14> to stand for the extranuclear property of existence;
in order to conform to the historical literature, this symbol will
usually be abbreviated by the symbol EL In giving examples,
we will often omit the superscripts and subscripts on symbols
when this will cause no confusion.
In addition to the terms listed above, we also use:
(4) Connectives: &, V, :::J, =, r«;
(5) The existential quantifier sign: 3 .
THE LANGUAGE (f) 65
(6) The identity sign: = 4
(7) Parentheses: ( , ).
(8) The watering-down operator: w.
No other signs are primitive signs of f!}.l
T erms and Well-Formed Formulas (WjJs)

In chapter 3 I introduced some notation for indicating the result
of plugging up one or the other place of a two-place predicate
by a singular term: if a is a two-place predicate and if 't' is a
singular term, then [a-r] is the (complex one-place) predicate
got by plugging up the second place of a, and [~a] is the (com-
plex one-place) predicate got by plugging up the first place 9£
a. It then seemed perspicuous to write the argument of the for-
mer predicate in front of it, and the argument of the latter pred-
icate after it, so as to make wffs of the forms "1[a't'2] and ['t'la]
"2- I will continue to use this notation throughout the book, but
it is not part of the primitive notation; instead, it will be defined
in terms of other primitive notation. The reason for this is simply
that when we deal with 3-, and 4-, and 5-place predicates (this
is rare, but I want to present the system in complete generality),
the bracket notation becomes cumbersome, if not totally useless.
And I want to have a uniform manner of indicating plugging up,
for predicates of any number of places. The idea will be to have
a single rule for turning any n + r-place predicate into an n-
place predicate by placing the singular term directly after it,
with a superscript to indicate which place of the predicate is
being plugged up. Thus if we wish to turn a four-place predicate,
say, into a one-place predicate we will simply apply the rule three
times, being careful to correlate the superscripts properly with
the places that we want to plug up. The formation rule is:
·(9) If a is an n + I-place nuclear or extranuclear predicate
term (with 72 ~ I), and if ~ is a singular term, and if I ~
k s n + I, then ea" is an n-place predicate term (which
is nuclear or extranuclear, respectively).
I. I have omitted consideration of relations, such as worships, that are
nuclear at the first place and extranuclear at the second place. This is only for
simplicity; treatment of such mixed relations would cause no theoretical
difficulties.
66 FORitAAL DEVELOPMENT
The intent is that the 'k) indicates which place of a is being

plugged up. It is important to keep this in mind when a itself is
complex; the 'k' will not be indicating (directly) which place
of the simple predicate within a is being plugged up; this is only
indicated indirectly. For example, in p3x1y2 the y plugs up the
second place of p3xl, and this corresponds to the third place of p3,
not to its second place.f
Notice that rule (9) is only a rule for forming complex pred-
icates, not wffs. Rule (I I) will tell us that once we have got a
one-place predicate, either by starting" with one or by plugging
up all but one of the places of a many-place predicate, then we
can make an atomic formula by putting a singular term after
the predicate (and giving it a superscript of' I'; this superscript
is, strictly speaking, superfluous, but it is a notation which aids
in the statement of the axioms). This is the only way that rela-
tional statements may be made in (!). .
The bracket notation for two-place predicates is now defined
as follows:
['t"la]'t"2 abbreviates a't"}'rz-
t'l [a't"z] abbreviates a't"~'l"l·
Also, as in chapter 3, I will use the unbracketed form, without
superscripts, as an abbreviation for the conjunction of the above:
'Z"laZ02 abbreviates '!l[a~2] & ['t"la}rZ·
Incidentally, the reader may be wondering why I haven't
defined plugged-up predicates in terms of abstracts in the "nor-
mar' way-for example, by defining:
[pZx]
as:
[Ay(ypZx)].
2. There is a tendency to read this notation "backwards," that is, to
suppose that the linear positions of the singular terms correspond to the places
they are plugging up in the innermost (primitive) predicate, and that the
superscripts correspond to the order of plugging up. This is not what my
notation means. It seems more natural to read it this way, but if I had adopted
such an interpretation, it would have made the semantics more difficult to
state.
THE LANGUAGE Ij) 67
I have an autobiographical reason which still holds: I thought

up how to treat relations (in terms of plugging them up) when I
had no idea as to how or whether this notation could be analyzed,
and I am still not certain that it can be analyzed. This is some-
thing that interested readers may want to explore for them-
selves.s
3. Briefly, here are some of the problems involved, and a sketch of one
possible way ofdealing with them.
First, if abstracts are treated in the normal manner, then [Ay(yph)]a is
provably equivalent to [ly(apy)]b, and so the order of plugging up becomes
irrelevant; this would destroy the theory I am presenting. So if you want to
use abstracts to define the bracket notation in the manner suggested, you must
give a special treatment to abstracts, and the normalcy of the approach would
then be an illusion.
Second, you can't use abstracts in the manner suggested unless you have
some account of their contents. That is, [Ay(ypb)] doesn't make sense unless
ypb does. And normally,ypb is part of your primitive notation. But it is not a
part of the primitive notation of the theory I am giving. You see, in the enter-
prise at hand, relational statements are problematic; the goal is to account for
them in some way. I do have the notationypb, but it is defined notation, de-
fined in terms of the bracket notation, and so it can't be used (defined in that
way) to define the bracket notation. In short, using abstracts as proposed means
that you will have to invent a totally new treatment of relational statements.
For a long time, even though I was goaded by David Kaplan, I couldn't
see my way around these obstacles. Now, too late for detailed development)
I think I see a way it can be done. This isnJt quite the suggestion given above,
but it is similar. My explanation of it here presupposes notation that isn't
developed un til chapter 5, section 2, so the reader r.nay wish to return to this
footnote later. Let me confine myself to nuclear relations, since for extranu-
clear ones the order of plugging up is- irrelevant anyway. We allow relational
statements to be written in the normal manner, without superscripts: pn-r. 1 • • •
"n' And we give the customary semantic explanation to this; it's true just in
case the objects denoted fall in the extension of pn. But we require, by edict,
that only existing objects can stand in relations, and we add as an axiom:
(x) (PDt'l . • • t'jX'Ck • • • Z"n :)E!x).
Then, confining ouselves to two ..place relations for simplicity, we define:
[cer] as w[lx(axf")]
and:
[-ra] as w[h(a-rx)),
where w is the watering-down operator (see below in text) and the abstracts
We have one more type of complex predicate to add to l!J.

In keeping with the discussion of chapter 2, section 3, each ex-
tranuclear predicate has a nuclear watered-down version, which
is written with the operator w:
(10) If a is any n-place extranuclear predicate term, then
w(a) is an n-place nuclear predicate term"
This completes our discussion of terms.
The well-formed formulas (wffs) are given by:
( I I) Atomic: If a is a one-place predicate term and if 1: is a
singular term, then a-r1 is a wff.
As I indicated above, the superscript of ' I ' is a notational con-
venience, and will often be omitted where the context makes it
clear that we are dealing with a one-place predicate followed
by a singular term.
(12) Atomic": If a and {3 are both singular terms, or both n-
place nuclear predicate terms, or both n-place ex-
tranuclear predicate terms, then a = (J is a wff.
are extranuclear abstracts (see chapter 5, section 2). We keep all axioms and
rules the same except that we rearrange all singular terms in accordance with
their superscripts (which are now dropped), and we omit the PLUG axioms
because they have now become theorems. Let us call this revised system (f)*.
Then (f) seems, in a sense, to have the same theses as (f)*; this is a slightly subtle
point, since the two theories have different notations, but there is a correlation
between them (probably more than one). The one major difference between
them seems to be that (!J* says that existing objects can stand in relations to
one another in a way in which nonexistents cannot. For if a and b both exist,
we can have both:
Rab
and:
a[Rb] & [aR]b
true in (!J*, whereas if either a or b (or both) fail to exist, then only the latter
can be true. This is not in accordance with the intuitions that guided my
development of (9. (On the other hand, perhaps the best "translation" of the
former statement into (f) is E!a & E!b & a[Rh] & [aR]b, in which case maybe
it is in accordance with those intuitions.. ) I am not in a position yet to ap..
preciate the significance of these considerations.
THE LANGUAGE (9 69
(IS) Molecular: If if> and if; are wffs, then so are ~ifJ,
(ifJ & cjJ), (</J V ep), (ifJ ::> c/J), and (if> == <jJ).4
(14)_ Quantified: If if> is a wff and if a is any variable, then
(a)ifJ and (3a)<fi are wffs..
Nothing is a wff except as indicated by clauses (11)-(14). Any
wff with no free variables will be called a sentence.
It is apparent that, except for the distinction between nu-
clear and extranuclear predicate terms and for the peculiar
way in which predicate terms are combined with their singular
terms step by step to form atomic wffs, this language resembles
the ordinary predicate calculus with identity and with predicate
quantification. Symbolization of English sentences should pro-
ceed much as with the ordinary predicate calculus.f For example,
let:
h stand for Sherlock Holmes;
g stand for Gladstone;
pI stand for being clever;
p2 stand for seeing.
Then we symbolize 'Holmes is clever' as:
plk,
where the superscript on the singular term has been suppressed.
We symbolize 'Something is clever' as:
(3x)p1x.
We symbolize 'Holmes has the property of seeing Gladstone' as:
p2g2h.
In abbreviated notation this would be:
h[p2g ] .
We symbolize 'Gladstone has the property of being seen by
Holmes' as:
4. I will sometimes omit parentheses in keeping with the conventions
that c>and == dominate & and v , which in turn dominate e-.
5. In particular, the reader is temporarily asked to suppose that the
constant singular terms of (f) behave like English proper names. This over-
simplification will be rectified in chapter 5, section 4.
p2h1g,
which may be abbreviated by:
[hp2]g.
Quantification is relatively straightforward; we symbolize 'Every-
thing that has the property of seeing Gladstone is clever' as:
(x)(p2g2x ::> plx),
and we symbolize 'Every real thing that has the property of

be_ing seen by Gladstone is clever' as:
(x)(E!x & p2g1x ::> plx).
The last two sentences may be abbreviated as:
(x)(x[p2g ] ::l plx)
and:
(x)(E!x & [gp2]x ::::l pIx).
Last, we symbolize 'Anything which has all and only Holmes's
nuclear properties is Holmes' as:
(x) (q) (qh == qx) ::> x = h),
in which superscripts "and subscripts have been suppressed.
2. Rules and Axioms
One way to state a large portion of the theory of objects is to

specify some rules and axioms; the "theorems" which follow
from them are all intended t<? be theses of the theory in question.
This policy will be followed in the present section. I will state
the rules and axioms in four groups. The theorems will be those
sentences which follow from the axioms of groups 11, Ill, and IV
by means of the rules of inference from group I; these rules of
inference closely resemble those of the predicate calculus with
identity.
I. Ordinary Rules and Axioms

I will suppose that we have available a set of rules for construct-
ing derivations for the predicate calculus with identity, such as
those in Mates [EL], expanded so as to be applicable to the
THE LANGUAGE (9 71
symbolism developed in section 1. For example, if t/J and 4J appear

on lines in a derivation, and if X is tautologically implied by (t/J &
c/J), then we may write X on a later line. The "expansions" are
only needed to deal with predicate terms. For example, not only
can we derive p1h from (x)p1x, we may also derive it from (ql)q1h.
Likewise, the identity axioms ought to allow us to; derive pIA
from p~h, together with PI p~.=
Following Mates [EL], chap, 7, with slight modifications,
we define a derivation to be any sequence, each member of which
consists of a number (the line number), a wff, and a set of premise
numbers, where the sequence is constructed in accordance with
these rules:
AX: Any axiom may be written on any line with the empty set
as the set ofpremise numbers.
P: Any wff may be written on any line with the unit set of the
line number as the set of premise numbers.
T: If t/J is tautologically implied by some wffs on some lines in
the derivation, cjJ) may be written on a line if <jJ appears on an earlier
line. Take as premise numbers those of the line on which
ifJ appears, leaving out any number that is the line number
of a line on which ifJ appears.
US: If (a)ifJ appears on a line, <P~ may be written on a later line
with the same set of premise numbers, provided that neither
(3 nor any variable free in {3 is bound in ifJf. 6
VG: If ifJ appears on a line and if a does not occur free in any
premise of that line, then (a)ljJ may be written on a later
line with the same premise numbers.
E: If (3a)rf! appears on a line then (a) "'ifJ may be written
'""J
on a later line with the same premise numbers, and vice

versa.
6. The notation <Ms is an abbreviation for 'the result of replacing every

free occurrence of a in tP by ~'. In case a itself is a complex term, an occurrence
of a does not count as "free" for the purpose of this characterization if that
occurrence of a contains a variable which is bound by a quantifier (or other
variable-binding operator, as in chapter 5) which is outside of a in ifJ.
11 : a = a may be entered on any line with the empty set for its
set of premise numbers.
12 : If a = (3 appears on a line and if rjJ appears on a line, then
</>~ may be written on a later line, provided that neither [3 nor
any variable free in f3 is bound in ifJ~. As premise numbers,
take all those of both earlier lines.
(Within the context of the theory developed in this chapter, the

rules of existential· generalization and existential specification
(= existential instantiation) are available as derived rules; see
Mates [EL] for a sketch of the proof.) .
A theorem of the theory is any wff that appears on any line of
any derivation with the empty set as its set of premise numbers.
Incidentally, although these "ordinary" rules allow us to
prove things of the form (3x) (Ax V ~ Ax), which require that
there be objects, they .will not permit us to prove that any ob-
jects exist; that is, they do not suffice for (3x) (E!x & ~ .
4 ).
I I. Properties and Relations

There are two sets of axioms that tell us that there are certain
properties and relations. The first deals with extranuclear rela-
tions, and it says, roughly, that corresponding to anything we can
say in the language about objects, there is an extranuclear rela-
tion which relates arbitrary objects if and only if those objects
are related as we say..
AXIOMS OF ABSTRACTION FOR EXTRANUCLEAR RELATIONS
AB(E): For any wff 4> which does not contain Qy free, any
instance of the following is an axiom:
The ki's may be chosen any way you like, provided that I :::; k,
:$; n - i + I for every i. An example of this axiom would be
this: suppose that PI stands for being incomplete and p~ for being
admired by Meinong. Then the axiom says that there is an ex-
tranuclear property which holds of any object just in case it is an
incomplete object that Meinong admired:
THE LANGUAGE (g 73
The second set of axioms contains the watering-down ax-

ioms; these place constraints on the relation between an extranu-
clear relation and its (nuclear) watered-down version..
WATERING-DOWN AXIOMS
WD: (QY)(Xt) . · · (xn)(E!xl & · . · & E!xn ::>

(QYXfl .... x~n == w(QY)Xfl .... x~n)).
These watering-down axioms, in conjunction with axioms

AB(E), immediately yield a set of abstraction theorems for nu-
clear relations. These say that corresponding to anything we can
say in the language about objects, there is a nuclear relation
which relates real objects if and only if those objects are related as
we say.
THEOREMS OF ABSTRACTION FOR NUCLEAR RELATIONS
AB(N): For any wff ifJ which does not contain qy free, any
instance of the following is a theorem:
(3qr) (Xl) - . . (X n ) (E!Xl & . 4 • & E!Xn ::>

(qlx~l .... x~n == tjJ).
The ki's are as in AB(E).
For example, ifpI stands for being clever and if p2 stands for
seeing, then, since we can then say in the language that x is
.clever and seesy, there is a nuclear relation which relates any two
real objects if and only if the first is clever and sees the second.
The instance ofAB(N) which says this is:
Notice that AB(JV) is noncommittal about what unreal objects

stand in the relation which is being specified; this is be-
cause other axioms will deal with this, particularly OB] and
PLUG(N).
Our axiomatic policy here is to follow the outline of chapter
I, in that we "begin" with assumptions of nuclear properties
which. have "known" extensions among real objects, and then
we make other assumptions (corresponding to the "correlation"
of chapter I) which extend these properties to unreal objects.
I I I. Objects
There are two sets of axioms which deal with what objects there
are. The first contains a single axiom; it puts a~ upper bound on
what objects there are in terms of what nuclear properties there
are. It is the axiom of the identity of nuclear indiscernibles from
chapter I: objects cannot differ at all unless they differ with
respect to at least one nuclear property.
AXIOM OF THE IDENTITY OF NUCLEAR INDISCERNIBLES
1Nl: (xl) (X2) «ql) (ql X1 == qlxZ) :::> Xl = xz).

The second set of axioms puts a lower bound on what ob-
jects there are: for anything we can say in the language about
nuclear properties, there is an object which has exactly those
properties that are as we say.
AXIOMS FOR OBJECTS
OB]: For any wff <p which does not contain Xl free, any
instance of the following is an axiom:
(3Xl) (ql) (qlx1 == ifJ).
So, for example, we get the gold mountain as follows: ifpi stands
for being golden andp~ for being a mountain, then we have:
(3Xl)(qI)(q1xl == (ql = pI V ql = P~)).
Uses of OB] following this pattern will yield objects correspond-
ing to any finite set of nuclear properties which are nameable
in the language (e.g., the round square). There are other uses,
too; for example, we produce an object just like Sherlock Holmes,
except that it has a mole on its back, by:
where pi stands for having a mole on one's back.

Notice that it is an immediate consequence of OB] that
coextensive nuclear properties are identical:
(ql) (q~) «x) (qlx == q!x) :::> ql = q2)-

For if ql #: q2 then, by OBJ, there is an object which has ql with-
out having q2- But this result is not as surprising as it might seem
THE LANGUAGE (f) 75
at first sight.. It is generally held that properties with the same

extension need not be identical, but in this conventional wis-
dom the notion "same extension" is generally held to mean
"same extension among-real objects." And OB] does not say that
properties which agree on the reals need to be identical. The
notion of nuclear property being used here is something like
C . I. Lewis's notion of the comprehension of a predicate (see Lewis
[MOM] p . 244). Lewis identified this with the set of all possible
objects to which the predicate is applicable. Our identity condi-
tions for nuclear properties are actually even tighter than this,
for nuclear properties are identical only if they coincide for
impossible objects as welL (This may make them closer to Lewis's
notion of the intension of a term; see Lewis [MOM] pp. 121-22.
Lambert refers to my notion as the distention of a term (personal
communication).)
IV. Plugging Up Relations

There are two sets of axioms dealing with plugging· up n-place
relations with objects to get n - r-place relations.. When stated
in complete generality, they are slightly difficult to read and
understand, and so it may be best to start with an example, using
the abbreviations introduced earlier for two-place relations.
Suppose that r is a two-place nuclear relation. Then a special
case of the first axiom (which will be given below) will be:
E!xl =' ([xl r]X2 & E!X2 == Xl[rx2])·
It may be best to break this into two claims, by analyzing the
inside biconditional into two conditionals.
The first says:
E!xl :J ([xlr]x2 & E!X2 :::> Xl [rxz]).
This says, essentially, that if both terms of a relation exist, then
the order of plugging up the relation makes no difference at all;
this was discussed in chapter 3. (Actually, it only says that if the
statement generated by plugging up the first place of the relation
first is true, then the corresponding statement made by plugging
up the second place first is also true; the converse of this comes
from another instance of the axiom being discussed.. )
The other conditional is:
E!Xl ::> (Xl [rXZ] :::J ([Xl r ]X2 & E !Xz) ).

This says that if Xl is real and it has a relational property, then
its relatum exists (and again the order doesn't make any dif..
ference). Notice that this conditional entails:
E!xl & ~E!X2 :::> ~xl[rx2];
that is, no real object has a relational property got by plugging

up a nuclear relation with an unreal object (again recall the dis-
cussion of chapter 3). But it does not entail :
E!xl & ",E!xz ::J ~[xlr]x2;
in other words, it does not say that an unreal object must lack
relational properties made by plugging up nuclear relations with
real objects. In fact, it is a theorem that if there are any real ob-
jects, then some unreal objects have such relational properties.
In virtue of OB] it suffices to prove that there is such a relational
property that no real object has, and this is obtained quickly
from the following instance of AB(N):
(3 q2) (xl) (xz) (E!Xl & E!X2 ::> (qZxlx2 == Xl =/; Xl))·
Following is the general statement of the first axiom set.
AXIOMS FOR PLUGGING Up NUCLEAR RELATIONS
PLUG(N): Any instance of the following is an axiom:

(qf)(Xl)"· . (Xn)(E!Xl & · ... &E!Xj_l
& E!Xj+l & . . . & E!xn :::> (qYXfl . . . x~n & E!xj
=
-
1 • • . xmj-1xmjO+l
qnXmJ' xm1
1 j j-l ..• xn ) ) ,
mn
j~l
where I s j :s; n and where the ki's and mi's must be properly
chosen. The proper choice is, intuitively, that each Xi fill up the
same place of q~ on both sides of the biconditional. This can be
made precise as follows (if you trust me, you may skip the next
few lines): if a is a simple n-place predicate term and if -rrr~ .... ,
't"rn are singular terms, then with respect to a formula if> of the
form a't~t .... 'L'"~~ we define the place of "q' written Pl<jS(~q), as
follows:
PI,('t" fI) = k 1•
Pl,(-t"ri+l) = the ki + 1th member of {I, ... ,n} -., {m. I there
is somej s i such that Pl.p{1: q ) = m}.
THE LANGUAGE (J) 77
Then the requirement on the ki's and mi's in PLUG(N) is that

for any i, Pl;l(xj) = Pln(xi), where 4;1 is the formula on the left
of the biconditional sign and <P2 the formula on the right.
The last axiom set is for extranuclear relations, and it simply
says that for extranuclear relations the order in which the places
are plugged up makes no difference to the truth of what results
(in any given case, reiterated applications of the rule may be
necessary in order to show this).
AXIOMS FOR PLUGGING Up EXTRANUCLEAR RELATIONS
PLUG(E): Any instance of the following is an axiom:

(Qj) (Xl) . . . (xn ) (QYX~l . . . x~n
j-l xjT!l . . . xn ) ,
- Qnxmjxml mj + l rnn
= 1 j 1 . . . xmj-l
where I ~ j s n and the ki's and mi's must be related as in

PLUG(N).
This completes the system of rules and axioms. The axioms
give a special status to existence, and they allow us to prove, for
example, that there are objects that don't exist (AB(N) gives us a
nuclear property that no real object has, and OB] says that some
object has it). There is no corresponding special treatment of the
notions of incompleteness and impossibility. The notion of in-
completeness rests on that of nuclear property negation, and will
be discussed in chapter 5; the notion of an impossible object
presupposes modal notions, and it, too, will be discussed, though
briefly, in chapter 5-
Actually we can already define a weak surrogate of incom-
pleteness and prove some illustrative theorems about it. Suppose
we say that x is complete" if:
Then we can prove trivially that all existing objects are com-
plete*, and that some objects are not complete* (the null object,
yielded by OBJ, will do). And if there are at least two existing
objects, then there are nonnull objects which are incomplete * 4
Let Xl and X2 be distinct existing objects. Then AB(N) gives us

a property PI that Xl has and all other existing objects lack, a
property P2 that Xl "lacks and all other existing objects have, a
property Ps that all existing objects have, and a property P4 that
78 FORMA.L DEVELOPMENT
all existing objects lack. Since Xl -:F X2' these four. properties are
distinct. Then axiom OBJ says that there is an object which has
exactly P3 and P4' and that object will be incomplete* because it
lacks both Pt and P2" '
Here are a couple of theorems that are basic to many ap..
plications of the theory:
(ql) (3Ql) (x) (Qx == qx).
The proof is immediate from AB(E). On the other hand, we
have:
~ (Ql) (3ql )(x) (Qx =qx).
Axiom AB(E) tells us that there is an extranuclear property
possessed by no object at all; axiom OB] tells us that no nuclear
property is coextensive with that extranuclear property, since for
every nuclear property there is an object which has it.
3· The Semantics of(f)

The language (!) is intended to be interpreted in a certain way;
for example, it is intended that the singular-term variables
range over all objects, that E! stand for the extranuclear property
of existence, and so on. In this section I will lay out certain con-
straints on any interpretation of the language; such constraints
will embody many metaphysical assumptions, but only those
which are put forth a priori. For example, although I believe that
there are real objects, lots of them, the constraints will not pre-
suppose this, for I don't envisage this as a priori true, And al..
though the intent is that certain variables range over, say, nuclear
properties, the a priori constraints will only require that these
variables range over things which behave like nuclear proper-
ties-they will have extensions which include subclasses of real
objects, they will be possessed by unreal objects, etc.
Officially, an interpretation of (!) is any nine-tupIe I which
satisfies conditions (I )-( I0) :
(I) The first member of I is to be a nonempty class OB~
which will be called the class of objects..
(2) The second member is a subclass EX of OB. EX is
called the class of existing objects.
THE LANGUAGE tJ 79
(3) The third member is a sequence of nonempty classes

NI' N2, N3, • • • • Each N n is caned the class of n-pltue
nuclear relations•
. (4) The fourth member is a sequence of nonempty classes
El' E 2, E 3 ) • • • • Each En_ is called the class of n-plaoe
extranuclear relations.
(5) The fifth member is a function ext which maps each
re N n to a class of ordered n-tuples of members of EX, and
maps each R e En to a class of ordered n-tuples of mem-
bers of OB. We call ext(X) the primary extension of X. (The
reason for the adjective "primary" is that, whereas the
primary extension of a nuclear property is a subclass' of
the real objects, such a property-will also be possessed by
certain unreal objects as well. The class of all objects
which possess a property will be called the full extension
of that property. For extranuclear properties, the full
and primary extensions coincide.)
(6) The sixth member is a one-one function f which maps
each member of OB to a subclass of NI. We call f the
correlation function. We require that for any XE EX,!(x) =
{r E NI I x E ext(r)}.. (The function f determines which
objects "have" which nuclear properties; i.e . , for rENt
and x E OB we define '» has r' as 'r ef(x)'.)
(7) The seventh member is a three-place function PLUG
which is such that for any re N n+1 and any x E OB and
any k between I and n + I, PLUG(r, x, k) e N n , and
>
ext(PLUG(r, x, k)) = {(Xb · . · ,xn I (xb ...... , Xk-b
>
X, Xk, . . . . , xn E ext(r)}; and likewise for extranuclear
relations (i.e., with N~ and N n+1 replaced by. En and
E n+1) ·
(8) The eighth member is a one-place function W which
is such that for any R E En' .W(R) E N n, and ext( W(R))
= ext(R) n EXn.
(9) The ninth member is a function, A, defined on the
terms of @. A is called the assignment function. A should
assign the right sorts of things to the terms of (!), namely:
(9.1) If" is a constant singular term, then A(1:) E OB.
(9.2) If a is a primitive n-place nuclear predicate constant,
then A(a) E N n -
(9-3) If a is a primitrve n-place extranuclear predicate

constant, then A(a) E En"
(10) Last, a global requirement: the members of! must be
such that the axioms OB] and AB(E) of the last
section are true in I. This requirement presupposes a
definition of 'true in I') which is given below..
Truth in an Interpretation
As a preliminary to defining truth in an interpretation, we char-
acterize what is meant by an "extended assignment." An ex-
tended assignment is a function g, defined on all the terms of {f}, and
satisfying (i)-(vi):
(i) g(a) = A(a) for any primitive constant term of (0..

(ii) If'C is a singular term variable, g(-'!) E OB.
(iii) If a is an n-place nuclear predicate variable, then
g(a) E N n -
(iv) If a is an n-place extranuclear predicate variable, then
g(a) E En.
(v) If a is of the form {Jt'k, then g(a) = PLUG(gC8),
g(r), k).
(vi) If a is of the form w({3), then g(a) = W(g(f3)).
Now we can define the notion 'ifJ is true~g' ('ifJ is true in inter-
pretation I with respect to assignment g') by the following re..
cursive definition:
(DI) If a is a one-place nuclear predicate term and if ~ is a

singular term, then az- 1 is tru.eI.g if and only if g(a) E
j(g('C) ).
(D2) If a is a one-place extranuclear predicate term and if
1: is a singular term, then a-r 1 is trueYtg if and only if
g(t') E ext(g(a)).
(DS) a = f3 is trueI,g if and only if g(a) = g({J).
(D4) ~ rp is truel,g if and only if ifJ is not truel,g"
(DS) (f/J & 4') is trueI,g if and only if both ifJ and cjJ are
truel,g·
(D6) (ifJ V cjJ) is truel,g if and only if at least one of ljJ, c/J is
trueI,g.
THE LANGUAGE ([) 81.
(D7) (ifJ == cjJ) is trueLg if and only if !/J and cjJ are both
truer, g' or neither is.
(DB) (if>:::> <jJ) is true I, g if and only if at least one of ,...., <p, </J
is .trueI~g.
(Dg) (3a)tjJ is trueI,g if and only if if> is trueI.g* for some
extended assignment g* which agrees with g for every
variable of (J) except possibly for (X.
(DIO) (a)<jJ is trueI,g if and only if ......, (3a)"""'1> is trueI,g-
Last, we define 'f/J is true in interpretation I' to be: '<jJ is trueI,g
for every extended assignment g' ~
There are three aspects of the characterization above that

call for some comment. First, the conditions are all fairly neutral
about the "identity conditions" for properties and relations.
They are consistent with the loosest identity conditions: that
r1 = '2 if and only if ext(rl) = ext(r2). And they are also con-
sistent with fairly tight identity conditions; for example, they
allow that A(al) #:- A(a2) even when a1 and CXz are logically
equivalent. The same is true of the rules and axioms of section 2
as well. I think that in
the best application of the theory the
identity conditions for properties should lie somewhere near the
tight end of these two extremes, but so far no condition has been
imposed that would require this. This issue will be discussed
further in chapter 5 in the light of the relevance of verbs of
prepositional attitude, and in chapter 9-
Second, let me call attention to the treatment of the exten-
sions of nuclear properties. The presentation here mimics that of
chapter I, and it is designed to conform to the methodological
policy of catering to the orthodox conceptual scheme. The idea
is that we begin with nuclear properties which have "primary"
extensions among the real objects. The theory then "asserts"
(via its use of the correlation function f} that there are also
certain unreal objects that have these nuclear properties, too.
This assertion appears embedded in clause (DI) of the truth
definition, together with clause (10). 7
7. It isn't necessary to do the semantics in this way. For example, we

may drop ext altogether as a primitive function in the case of nuclear prop-
Third, clause (6) is actually a weakening of the theory as

presented in chapter I .. In chapter I it was asserted that for any
class of nuclear properties there is an object. which has exactly
those properties in the class. But clause (6) does not require this,
for it does not require the correlation. function f to be onto the
power set of NI. What does (6) require? Well, in the presence of
clause (10) it requires that for any class of nuclear properties
which is describable within (J) there be an object which has exactly
those nuclear properties that are in the class. And there are
interpretations in which the classes of nuclear properties de-
scribable within (f) do not include all the subclasses of NI- When
I say that a class X of nuclear properties is describable within f!} in
an interpretation I, I mean that there is some wff ifJ of (9 which
contains q1 free such that relative to some assignment to the
other free variables, X is exactly the set of nuclear properties
which satisfy ifJ. This is a very liberal sense of 'describable', since
it allows, for example, for the "description" of an uncountable
number of classes of nuclear properties, even though the language
is countable. For example, if we are working with an interpreta-
tion in which OB is uncountably big, then the wff ql Xl can be
used to describe any class of nuclear properties that is the class
of properties possessed by some object; given any object x, the
class of nuclear properties that satisfy ql Xl relative to the assign-
ment of x to Xl is just the class of nuclear properties possessed
by x, Given such a liberal notion of describability, some might
hold that it doesn't make sense to hold that there might be
undescribable classes. I think that it does make sense, but if the
reader disagrees, the present discussion may simply be ignored.
The reason for formulating the semantics in the weak way it
has been formulated is simply this: so formulated it is possible to
erties and relations, and just do everything in terms of the function f We may
then reintroduce ext by means of the recursive definition:
If T E Nu then ext(r) = {x Ix E EX & 1 ef(x) J.
If rE JVn+h then ext(r) = {<x u • • . , x n +1) I (x 2, ••• J x n + 1) e
ext(PLUG(r, Xl) 1)}.
If we do this, the third sentence of condition (6) becomes true by definition.
The condition regarding ext in condition (7)) however, is not true by defini-
tion, and it must be retained in order for the PLUG axioms to· turn out true.
THE LANGUAGE (f) 83
prove a completeness theorem for the theory. That is) those wffs
which are provable from the rules and axioms of section 2 are
precisely the wffs that are true in every interpretation (see section 4
for a sketch of the proof) . Thus there are two independent ways of
characterizing the theorems of the theory-by means ofrules and
axioms, as in section 2, or semantically, as in the present section.
If clause (6) were strengthened by making f be onto the
power s~t of NI' then there would be fewer interpretations and
there would be more sentences true in every interpretation. The
axioms and rules ofsection 2 would then no longer be adequate to
prove all of these.. Indeed, it can be shown that no recursively
specifiable set of axioms would suffice to generate as theorems all
those sentences.f
Which formulation of the theory is better? I am not sure.
Fortunately, the choice doesn't matter much. For most purposes,
the completeness theorem given in section 4 is just a nicety.
Hereafter, almost nothing I have to say will depend on whether
the theory is formulated weakly or strongly; when I think there
is a difference, I'll mention it. So the reader may pick his or her
preferred version and go with it.
Those readers who are not formally inclined may want to
skip sections 4-6. If you do this you should take on faith:
CA) The system of rules and axioms of section 2 is consistent;
no contradiction may be derived within that system..
(B) The language (f) does have models of the sort described
in the present section; that is, it has interpretations in
which all the axioms are true.
(C) The sets of rules and axioms of section 2 a.re complete;
the theorems of that system consist exactly of those wffs
that are true in every interpretation.
8. This is because arithmetic would then be formulable within the theory.

More specifically, we can write analogues of Peano's postulates within (f) in
which nuclear properties play the role of numbers, and objects are treated as
sets of these "numbers." Ifclause (6) were strengthened, then the object quanti-
fier in the Peano induction postulate would have to range over all "sets" of
"numbers," and the theory would become categorical; so the valid sentences
would include analogues of all first-order truths of arithmetic. By Godel's
theorem such a theory cannot be recursively axiomatized,
4. Theoremhood Equals Validity
Suppose we say that a wff is valid if and only if i.t is true in every
interpretation (where an interpretation is understood to satisfy
clauses (1)-(10) of section 3). And suppose we say that a wff <p
follows validly from a set r of wffs if and only if c/J is true in every
r
interpretation in which every wff of is true. It is then straight-
forward to show that every theorem of the system of section 2 is
valid, and also that if a closed wff if> is derivable from a subset of
I', then <p follows validly from I',
The converse is also true: all valid wffs are derivable, and
if <p is closed and follows validly from F, then there is a derivation
of <p from a subset of r. This completeness result can be proved
by using a standard Henkin technique. It suffices to show that
r
if is any set of wffs from which no contradiction .is derivable,
then there is some interpretation in which every wff of r is true..
r
The technique for showing this is as follows: given a set ofwffs
from which no contradiction is derivable, one can extend to a r
r
superset /71 of from which. no· contradiction is derivable and
which is both com-plete. and cv..complete (where complete =df
for every closed· wff tjJ, either <p or '" <p is a member of F'; and
oi-complete =df for every wffin F' of the form (3a)tP(a),. there is
another wff in I" of the form r!J({3).) .. This much of the proof is
standard. .Then ther.e is a recipe for constructing an interpreta-
tion I such that every wff of F' (and thus every wff of () is true
in I, as follows:
First, for any constant term a of I!J, let [a] be the set of all
terms {3 such that a =. {3 is a member of I": Then we make the
members of I as follows:
OB = {[r] It'- is a constan t singular term of (9}.
EX = {[-r] IE!z E F'}.
N n = {raJ la is "an n-place constant nuclear- predicate
term].
En = {[a] fa is an n-place. constant extranuclear pred-
icate term}.
ext = the function such that for any n-place constant
nuclear predicate term a, ext([aJ) = {< Iril-
THE LANGUAGE (9 85
• • . , (ZOnJ >la7:} · .. 1:~ E F' and E!-rr E ["

and . . . and E!'t"n E r/};. and for any n-place
constant extranuclear predicate term a', ext([a])
>
= {<('Cl], . . ., [1:"n] la'l;} · . . 'C~ EF'}.
i = the function which maps each [~] to
{[a] I [a] E JV1 and a't"l EF'}.
PLUG = the function which is such that for any [a] E
N n + 1 or E n+1 and any [1:'] E OB and any k be-
tween I and n + I, PLUG([a], [r], k) = [a-rk] .
W = the function which maps each [a] E En to [wa].
A = the function which maps each primitive constant
term a to [a].
It is then straightforward to show that I is indeed an interpreta-
tion in which every wff of l" is true.
5. An Orthodox Theory fe
In this section I will formulate a theory that, with one reserva-

tion, I take to be a portion of the orthodox conceptual scheme.
My purpose is to allow a precise comparison of the theory (!)
with orthodox opinion. The reservation is that the orthodox
theory formulated here supposes that there are classes and prop-
erties, and this supposition runs counter to the nominalists
and extensionalists who form a significant subgroup of the ortho-
doxy. Perhaps in the end there is a way to make the theory
palatable to this subgroup, say by "identifying" properties and
classes with the predicates of an appropriate language; but that
is an issue I will not pursue here.
My development of this theory is parasitic on that of Henkin
in his paper, "Completeness in the Theory of Types." The theory
may be considered by some not to be a metaphysical theory at
all, but rather a part of logic; at least it is usually discussed in
works oflogic. I won't need to take sides on this issue; fortunately,
there is nothing in the part presented here that is at odds with
the theory (9. The theory, in fact, is a variant of what is some-
times called the fourth-order predicate calculus. It assumes that
there is a nonempty class of individuals. For purposes of compari-
son with l!J, these can be thought of as the same as the existing
objects. Second, the theory assumes that for any n there are certain
n-place relations among individuals; these have extensions which are
classes of ordered n-tuples of individuals. Third, there are classes
ofproperties of individuals (where a property of individuals is just
a one-place relation among individuals). And last, there are
certain n-place relations among the classes of properties of individuals
that have extensions that are classes of ordered n-tuples of those
classes. Normally, the theory of types assumes much more than
this, but these assumptions are all that I will be using, so I'll
leave out the rest.
I'll call the language of this theory ftJ. I'll begin with a pre-
cise account of its syntax, then I'll state rules and axioms for it,
In section 6, !l' will be used to provide a "model" for {f}.
The Syntax of.P .

.ft' contains four sorts of primitive terms:
( I) Individual terms:
Constants: i 1, i2, is, . '" Variables: i 1, i2, i g, .
(2) n-Place individual-predicate terms:
Constants: Pl, p~, p~, ... Variables: q~, q~, q§, . · .
(3) Class terms:
Constants: AI' A 2, A 3, .. • .. Variables: X b X 2 , X 3, .
(4) n... Place class-predicate terms:
Constants:P~,P~,P3'" ... Variables: Q¥, Q~, Q3' . · ·
In addition to the primitive terms there are also the following

primitive signs:
(5) Connectives: &, v, :::>, ==, r-«:
(6) The existential quantifier sign: 3.
(7) The identity sign: = .
(8) Parentheses, brackets, braces: ( , ), [ , ], { , }.
(9) Abstraction signs: tt, I.
(10) The membership sign: E.
Then there will be certain complex terms:
(I I) If tP is a wff and if aI, . . . , an are all individual
varia-bles or all class variables, then [AaI ... an<;S] is
THE LANGUAGE (!) 87
an n-place individual-predicate or class-predicate

term, respectively.
(12) If tP is a wffand if q~ is a one-place individual-predicate
variable, then {q~ I <p} is a class term.
The atomic wffs are straightforward:
(13) If Cl is an n-place individual-predicate (or class-predi-
cate) term and if {31' .... , f3n are individual (or class)
terms, then af31 . " · f3n is a wff.
(14) If a is a one-place individual-predicate term and {j
is a class term, then a E (3 is a wff.
(15) If a and {J are both terms of the same kind, then 0: =
f3 is a wff.
The complex wffs are the usual ones :
(16) If a is a variable and if f/J and </J are wffs, then '" 4>,
(</J & cjJ), (ifJ V cjJ), (ifJ ::J ifJ), (ifJ == cjJ), (3a )1>, and (a)ifJ
are all wffs.
Rules and Axiomsfor !f

The rules and axioms come in three groups:
I. ORDINARY RULES AND AXIOMS: Again we suppose that we

have available ordinary predicate-calculus-type rules which stem
from the logic of the connectives, quantifiers, and identity"
I I. ABSTRACTION AXIOMS: There are three sets of abstraction

axioms; they say, essentially, that the class and relation abstracts
work properly. In the following formulations the notation:
"~P:: : :p~" abbreviates the result of replacing every free occur-
rence of f3i by an occurrence of aj in ifJ, for every i from I to n.
(The notation will also be taken to presuppose that in the replace-
ment none of the new occurrences of ai's which are variables
become bound by any quantifier or abstract in ifJ.)
The universal closures ofeach of the following are axioms:
Ind. Rel.: [Ail" · . in~] 't"l .. • . ~n == 4Jfl: : :f:
Class Rei.: [AX1 · .. • Xn~] ~1 • •• Z'n == 4>~l: : :~~
Class: t E {q~ I ifJ} == <P~~
Ill. CLASS EXTENSIONALITY: This is an axiom:
6. Modeling (9 Within .2
It is possible, from the orthodox point of view, to form a

kind of picture or model of the theory of objects. The purpose of
discussing this model here is to establish the consistency of the
assumptions that underlie the theory of objects. (Later we will
discuss how to model the orthodox theory within the theory of
objects, with a different end in mind.) I'll begin with an informal
description of the model, and then I'll exploit its central ideas in
establishing precise syntactic and semantic relations between "P
and {f}. ~~
Basically, the idea is to picture or represent objects by means

of the sets of nuclear properties that they have. From the ortho..
dox point of view, these sets of properties exist, and so talk about
them is not metaphysically suspect. We can then view talk about
nonexistent objects as elliptical for talk about sets of properties that
are exemplified by no existent thing. For example, saying that the
golden mountain is golden can be represented by the orthodox
paraphrase: '{goldenness) mountainhood} includes goldenness',
and saying that the golden mountain is incomplete can be rep-
resented orthodoxly as 'for some p, {goldenness, mountainhood}
fails to include both p and the negation of p' .9
The particular picturing given here is based on the assump-
tion that there is no overlap between nuclear and extranuclear
properties, but it is easy to alter it to another which assumes that
the nuclear properties form a subset of the extranuclear (that
modeling will not be pursued here). The modeling of f!} within
2 goes as follows:
Recall the resources of the orthodox theory that will be used:
there are individuals, relations among individuals, classes of
9. In [P MS] I tried to describe a theory of objects indirectly by means

of such a picturing. But so many people mistook the picture for what was being
pictured that I have abandoned this as an expository device. I am using it in
this section solely as a portion of a consistency proof.
THE LANGUAGE (!J 89
properties of individuals, and relations among those classes. To

begin the model, we will represent or picture nuclear relations by
relations among individuals; specifically, we will now designate
any relation among individuals a nuclear relation representative.
Next, we will represent objects by classes of properties of indi-
viduals, that is, by classes of nuclear property representatives. In
particular, we will now call any such class ·an· object representative.
We can picture an object's having a nuclear property as follows:
we say that an object representative has" a nuclear property
representative if and only if the latter is a member of the former.
Notice that we have now followed the strategy of chapter I in
finding a way to correlate object representatives one-one with
the classes of nuclear property representatives that they have*;
in this model the correlation is the trivial one of identity..
Which object representatives represent existing objects?
Well, those that correspond to orthodox individuals in a natural
way. We will say that an object representative exists" if and only
if there is some individual i such that the object representative is
the class of (orthodox) properties (of individuals) that i has.
Finally, we will designate relations among classes of nuclear
property representatives (i.e., relations among object representa-
tives) extranuclear relation representatives. And we will say that an
object representative has* an extranuclear property representa-
tive if and only ifit has it (in the orthodox sense of "has") .
So far, what we have is:
Things Doing the Picturing Things Being Pictured
Relations among individuals Nuclear relations
Classes of properties of Objects
individuals
Classes of properties of Existing objects
individuals that correspond
to individuals
Relations among classes of Extranuclear relations
properties of individuals
This gives most of the picture; the only remaining items have to
do with how to picture the plugging up of nuclear and extra-
nuclear relations, and the watering-down operation. Roughly,
it goes like this: for any n + I -place nuclear relation-representa-

tive r and any object-representative X, the nuclear relation rep-
resentative which represents the result of plugging up what r
represents in its kth place by what X represents is that relation
among individuals which holds of any i b . . . , in if and only if
there is some individual i such that X is the class of properties
<
that i has, and r holds of i 1, • . . , i k- h i, i k , • • . , in (i.. e., i l ,
. . . ) i k- 1, i, i k , • • • , in> ~ ext(r)). For extranuclear relation
representatives, things are slightly simpler: for any n + I-place
extranuclear relation representative R and any object representa-
tive X, the extranuclear relation representative which represents
the result of plugging up what R represents in its kth place with
what X represents is that relation among classes of properties of
individuals which holds of any Xl' . . . , X n if and only if R
holds of Xl' - - . ,Xk - 1, X, X k , A • X n • (Actually, the character..
• ,
izations just given do not uniquely single out relations, they just
constrain the extensions of the relations being chosen; but that
won't matter for purposes of the picturing.)
To picture the watering-down operation, we just need this:
if R is an n-place extranuclear relation representative, then W R
is to be an n-place nuclear relation representative such that W R
holds of any i1, •• - , in if and only if R holds of {r I i 1 has r},
. . . , {r I in has T}.
Syntactic Modeling of (f) Within!l'

The syntax of 2 was chosen primarily so as to facilitate formu-
lation of the modeling within it just discussed. First, let us in-
troduce certain notational abbreviations into .P.
(I) If a is a one-place individual-predicate term, then:
ax~ abbreviates a E Kn-
aa~ abbreviates a E An.
This abbreviation technique takes a sentence of fLJ which "pic-
tures" a certain claim of (9, and replaces that sentence of ::£' with
the sentence of (!) which says in (!) what is being pictured in fL/..
For example, the sentence pI E A of It' says that a certain property
is a member of a certain class; but the property represents a
property of (!), and the class represents an object of (9. The ab-
breviation of th~ sentence, namely p1a1, says within (f) that a has
THE LANGUAGE o 91
the property pI, and this is the relationship that is pictured by the
original sentence of ft'. Similar remarks _apply to the other ab-
breviations.
(2) If a is a one-place class-predicate term, then:

aXa abbreviates aXn~
aa~ abbreviates aA n *
(3) If a is an n + I ..place individual-predicate term, with
n > 0, then a.x~ abbreviates:
[Ail ... in (3i ) (ql) (ql E X j == q1i ) & ail . . .
ik-1iik · · . in)], .
and aaf abbreviates the same thing, with Xj replaced
byAj .
(4) If et is an n + r-place class-predicate term, with n > 0,
then aXf abbreviates:
[JlX 1 • • • XnaX1 • • • Xk-1XjXk • • . X n] ,

and aaf abbreviates the same thing with X j replaced
by A j •
(5) E!x~ abbreviates (3i) (ql) (ql E X n == q1i).
E!a~ abbreviates (3i)(ql)(qI E An == qli)~
(6) wa abbreviates [Ail . . . ina{ql I q1i 1} . . . {ql I q1in } ] .
(7) X n = X m abbreviates X n = X m •
X n = am abbreviates X n = Am·
an = Xm abbreviates An = X m -
an = am abbreviates An = Am.
(8) If ifJ abbreviates tjJ* and <jJ abbreviates cjJ*, then '" ~
abbreviates .--1>*, and (</> & </1) abbreviates (4)* & cjJ*)'J
and similarly for the other connectives, and (3x n )lft
abbreviates (3Xn ) if1*, and similarly for the universal
quantifier.
Given these abbreviations, it is straightforward to establish that

every theorem of (!) is an abbreviation of a theorem of 2. But
2 is universally regarded as being consistent.l? and thus rg is
also consistent." For if (!) were not consistent, it would have a
10. See Quine [STL] sec. 36 for a kind of proof of consistency for a system
of type theory of which 2 is a fragment.
theorem of the form <p & ~ </1, and this would be an abbreviation
of a theorem of £'. But any abbreviation of the form if! & ~ <P is
an abbreviation of something which is also of this form, and so
this would mean that !l' was inconsistent, too.
7.· Uses of the Picturing

At the beginning ofsection 6, I gave a rough description of a way
that one might "picture" {f} within .fil; objects of (f) are pictured
by the sets of properties that they have, and talk about objects
is then pictured by talk about those sets of properties. Then, at
the end of that section, I showed how a certain syntactical version
ofthat picturing could be made precise, and this offered a relative
consistency proof of {f} with respect to 2. Specifically, one can
produce a purely syntactic correlation (let me call it momentarily
a "translation" from f!) into 2) which translates theorems of (f)
into theorems of 2. Then if (!) were inconsistent, it would have
a contradiction as a theorem, the translation of this contradiction
into !t' would also be a contradiction, and so-!l' would also have
a contradiction as a theorem; so ~ would be inconsistent, too .
So mis consistent if Sf i~.
The question that I want to discuss here is: why doesn't
this. translation procedure accomplish the task that philosophers
have failed at for decades, the task of finding a systematic way
to paraphrase apparent talk about nonexistent {concrete) objects
into talk about existing things and their properties? The idea is
that "obj-ects" are _pronounced to be logical constructions (out of
classes of properties); one c.an speak the language (i.e., use the
syntax) of the. theory of objects, endorsing the theses of that
theory, at least verbally, while always meaning things that would
never disconcert the staunchest member of the orthodoxy. For
example, the radical-looking "Some object is. both golden and a
mountain" can be endorsed-with the intent that it really refers to
the class of properties out. of which the "object" is constructed,
and what it really says is "Some class of properties contains as
members both goldenness and. mountainhood.'
Before I discuss this issue let me insist on a point that
should be apparent" by now: this paraphrase of talk about objects
into talk about classes of properties is not "the intended interpre-
THE LANGUAGE (!) 93
tation of (9. I don't think that concrete objects are sets of proper-
ties. For example, I don't think that you are a set of properties.
And I think the same about Sherlock Holmes, the gold mountain,
and the round square. Occasionally, people have said to me (in
an effort at friendly compromise), "When you say 'object', you
really mean 'set of properties', don't you?" No, I don't.
However, someone else might want to use (!), but mean
"object" as "set of properties"; have I now provided them the
means to do so? This will take some careful looking at. First, I
assume that when we are comparing (!) and 2 within the context
of the proposed "elimination" of nonexistent objects, we are talk-
ing about interpreted theories, not uninterpreted symbolisms.
And second, I assume that the task of the speaker of!l' is to agree,
verbally, with the speaker of (9, but without thereby committing
himself to there being any (concrete) nonexistent objects. Be-
cause if verbal disagreement is allowed, the task is trivial; the
orthodox philosopher can simply speak the language of @ but
deny that there are any nonexistent objects, and all that this
entails within {f}. For example, he can deny that Sherlock Holmes
is more famous than any real detective, deny that Pegasus is the
winged horse of Greek mythology, deny that Zeus is Jupiter, etc.
This isn't the problem; the problem is to agree with all these
"data" without thereby committing yourself to nonexistent ob-
jects. To be more precise, let us suppose that there is a speaker
A, of (9, who believes everything about nonexistent objects that is
discussed in this book, and articulates those beliefs in the sym-
bolism of (j). Then the task is to find a way to agree with those
sentences, by systematically paraphrasing them into orthodox
truths of cP.
Here is the difficulty: it is straightforward enough to cor-
relate certain words of (!) with certain words of 2, as described in
section 6, so that the correlation will preserve the syntactic prop-
erties of theoremhood and nontheoremhood. But now we are
trying to find a method of correlating certain words of (fJ with
certain already interpreted words of !l' in such a manner that
this results in paraphrasing certain sentences of (!) into certain
sentences of ft' which have the semantic property of truth, Success
in the former task does not automatically guarantee success in
the latter task, and this is the main point of this section.. Since the
point is somewhat abstract, I will comment further on where the

difficulties lie.
Consider a speaker B, of 2, who is trying to mimic A's
sentences. There is some reason to think that this can be partly
done. For example, there are some primitive nuclear predicates
in (f). Each of them is correlated with a predicate of English; for
example, suppose that p~ is supposed to stand for the same prop-
erty as the English predicate 'is a horse'. Then perhaps B can
simply adopt the same interpretation for the predicate of his that
he intends to translate p§ into; that is, B insists that the predicate
that he will correlate with p~ also means the same as 'is a horse'.
Of course, A and B will have certain disagreements concerning
the property that this predicate stands for; for example, B will
think that only existing individuals have this property. But this
needn't mean that they are not talking about the same property.
Then perhaps a great deal of the job will already be accomplished
by correlations of this sort. The sentence of A's 'There is a gold
mountain' gets paraphrased just as 'There is a class of properties
which contains both goldenness and mountainhood'; and A's
sentence 'No gold mountain exists' gets paraphrased into B's
'No class which includes both goldenness and mountainhood is
the class of properties of some individual'. Things seem to be
working fine.U
The trouble comes when we try to do the same for the extra-
nuclear predicates. Certain special cases may be easy (e.g., de...
fined predicates such as 'is complete'), but most of them will be
I I. Actually, I am skimming over several potentially serious impediments

to the translation. One has already been hinted at: it is not at all obvious that
the nuclear predicates can be translated as "themselves" without change of
meaning (by dint of change of theoretical context). A second possible im-
pediment is that the "translations" will not 'preserve ontological commitment;
it is not clear whether this is a virtue of the translation or whether it proves it
incorrect. I am also skipping over the problem of correlating the domain of
objects within (j) with the domain of sets of properties of individuals within 2,
which presupposes a prior agreement concerning the ranges of the (nuclear)
property variables. Likewise, there will be difficulties concerning the transla-
tion of names between the theories. And, of course, I am ignoring episternic
con texts altogether.
THE LANGUAGE (f) 95
problematic.. I will discuss the de re reading of 'thinks about' as an

example, though practically any other case would work as well.
(Please be careful to avoid interpreting any of the following as
involving the de dicta reading of 'thinks about'.. ) Now although A
may again say that a certain predicate of (f), say P2, means the
same as the English 'thinks about', this cannot be said by B con-
cerning the predicate of .2 that 'thinks about' will translate into..
The reason is very simple rP B's translation of.P! is going to be a
predicate that holds, not between individuals and things, but
between sets of properties and sets of properties. And typically,
thinking about something (de re) does not involve thinking about
a set of prope~ties.
Let's compare this again with 'is a horse' _B can insist that a
certain individual predicate of !e means the same as the English
'is a horse', because B believes that it is individuals that are
horses-it is individuals that have the property of being a horse.
When B translates A's remarks about horses into ~, those re-
marks will translate into certain remarks about the membership
of this property in certain sets. This doesn't commit B to thinking
that horses are sets; it only commits B to construing A's remarks
which are apparently about horses into remarks which are really
about sets of properties. But in the case of extranuclear predicates,
things are different. Here the translation format of section 6
requires that A's extranuclear predicates be correlated with
class predicates of 2, predicates which stand for properties that
sets, not individuals, have. In fact,' the goal of the translation is
to find a predicate of .2 that relates two classes of properties,
just in case the first is the set of all and only the properties of an
object Ob and the second is the set of all and only the properties
of an object 02, and 01 thinks about 0z- And this relation won't be
the same as 'thinks about' for, even ignoring troubles with the
first place of the relation, it's just not true that you think about
an object if and only if you think about the set of nuclear proper~
ties that it has .
Is there, perhaps, vocabulary within !l' in terms of which
12. The discussion here parallels Church's discussion of a similar point in
the Introduction, note 20, to Church (IML].
this relation could be defined? I doubt it. If I am correct, this is a

task that has occupied philosophers for some time now: the task
of trying to produce, by definition, terminology which would
permit us to mirror this sort of talk about nonexistent objects.
For a recent attempt in this direction, see Chisholm [TIR] and
Harmon [HUP].
Of course, if we begin with an uninterpreted version of 2,
then there is reason to suspect that there is an interpretation of
that symbolism that would do the trick (we can suppose this if we
are Platonists about meanings and if we believe that there are
enough of them of the right sorts). But there is a big difference
between there being a language or a theory of a certain sort, and
our possessing such a language or theory.. If our goal is to show
that an orthodox paraphrase of (!) is in principle possible, then the
efforts of the last section have gone a long way toward making
that plausible. But very little has been accomplished by way of
actualizing that possibility.
It is not even clear at this time what the significance would
be if this task were accomplished. I suspect that it would depend
a great deal on how it was accomplished. Suppose, for example,
that it were somehow established that we had the terminology
within !£ to mirror (J), somewhat as in section 6. Then notice
that this mirroring would eliminate talk of all objects in terms of
sets of properties. IS But if all reference to objects can be avoided,
the existent as well as the nonexistent, what significance should
this have for ~he special case of the nonexistent ones?
I suspect that this issue is similar to a problem that can be
formulated entirely within the orthodox scheme: can existing
objects be totally eliminated in terms of properties? Sure, it can
be done, at least in principle; just replace each object" x by the
property of being identical with x, and replace claims about x
by cleverly chosen claims about that property. Or, since there
13. This is not automatic, since the procedure from section 6 correlates
existence claims of (!) with claims of .fLJ that say that there are such and such
individuals; it seems plausible, however, that if the resources could be found
within .flJ to mirror all the other extranuclear predicates of f!), then this could
be done in the case of existence as well.
THE LANGUAGE @ 97
are at least as many numbers as there are concrete things, and

there is a correlation between existing things and certain numbers,
talk about the former can, in principle, be replaced by talk about
the latter. This possibility hasn't made Pythagoreans out of very
many ofus, and I suspect that the situation regarding nonexistent
objects ought to parallel this.
5
Further Developments of the Language ([)
The language (f) developed in chapter 4 is a relatively austere

one. In this chapter I will expand it by adding various things to
it, principally modal operators, property abstracts, verbs of
propositional attitude, and definite descriptions and names.
These will be important in -applications of the theory later in
the book. I will devote sections 1-4 to an informal discussion of
these additions; the explicit formal treatment will. be given in
section 5. (Note,: This chapter is summarized in chapter 6,
section 3.. )
I. Modalities
I will begin with the addition of modal operators, So let me

suppose that we have added to our .vocabulary a pair of
sentential operators, 0 and 0, representing 'necessarily' and
'possibly', respectively. The idea behind their semantical treat-
ment is as follows. Suppose we are working with one interpreta-
tion of f!), the "intended)' interpretation, 10- For example, 10
assigns the "right" properties to the predicates of the language,
and it also assigns the "right" extensions to those properties. (I
am thinking here of {f} as representing English (as sketched in the
appendix). The "right" assignment to a predicate is the property
that the corresponding word of English stands for, and the
"right" (full) extension of that property is the set ofobjects which
indeed have the property.) Now, for modal purposes, we want
to consider various ways in which the world might have gone
98
FURTHER DEVELOPMENTS OF THE LANGUAGE (j) 99
differently. This will be accomplished by considering interpreta-

tions other than 1o, interpretations which agree with 10 about
which predicates stand for which properties (for we are not
envisaging changing the meanings of the words of our language)
but which differ from 10 concerning what extensions those prop-
erties have. We can think of these alternative interpretations
as descriptions of alternative possible worlds. In fact, for present
purposes, it is convenient to simply identify possible worlds with
interpretations, and so hereafter when I say 'possible world' j
I will take this to be a heuristically chosen synonym for 'inter-

pretation'.
Now let us define a modal structure to be any set of possible
worlds which are exactly the same as one another except for:
(i) their second members (EX); that is, they may differ
with respect to the specification of what objects exist;
.and:
(ii) their fifth and sixth members (ext andf); that is, they
may differ with respect to which objects have which
properties.
Then we can handle modalities as follows. First, we have to sup-
pose that we are given not just one interpretation, but rather a
whole modal structure X. Then, given X, we say that a formula
of the form Deft is true in the possible world w of X with respect
to an assignment to the variables g just in case ifJ is true in every
world in X with respect to assignment g.
Here is a simple example. Suppose that we have a possible
world w which is a member of a modal structure X, and suppose
that w assigns to e the property of being an emerald, and to g
the property of being green. Then consider the sentence (neces-
sarily, all real emeralds are green'; that is} 0 (x) (E!x & ex =:J
gx). This will be true in w just in case (x) (E!x & ex :::> gx) is true
in every world in X. (With closed sentences we don't need to
worry about assignments to the variables.) Well, will (x) (E!x &
ex ::J gx) be true in every possible world in X? That depends on
what is in X. Since (x) (E!x & ex ~ gx) is neither a theorem nor
the negation of a theorem of our theory, the semantical rules
leave it open to us to choose what to have in X. (This is like

choosing which of the purportedly possible worlds ofpossible worlds
theory are genuinely possible ones.) If we think that it is a neces-
sary truth that all real emeralds are green, then, for modal pur-
poses, we will only consider using modal structures in which
every world in the structure makes the (primary) extension of
emeraldness a subset of the (primary) extension of greenness;
otherwise, we will ensure that some world in the structure violates
this condition.
Let us call a formula modally valid if it is true in any possible
world in any modal structure (with respect to any assignment to
the variables). Then all the theorems of (9 are modally valid, of
course.. But also, if ~ is a theorem of (f), then O·~ will also be
modally valid. Likewise, all theorems of Kripke's quantified Ss
are modally valid (see Kripke [SCML]). For example, any forrnu-
la of the form O(f} ::> <jJ) ::J CO</> ::J Dif;) will be modally valid.
We will also have all of the S5 reduction theorems; for example,
O~ == O·O<fi. This is not an intended result, in the sense that I
think that 85 is the right modal system, or that I prefer it to
others, It'sjust that Ss is simpler than other systems to formulate;
a slightly more complicated semantics is necessary to yield these
systems. If this were an essay in modal logic, I would explore
these options, but it is riot, so I will not.
There are also some modally valid formulas that do not
correspond to any of Kripke's. For example) (x)O(3y)(x = y)
is modally valid, and it is not a theorem of Kripke's system. But
the significance of this sentence is different in the two systems.
Kripke supposes that his quantifiers range only over what exists
(in each world), and so he reads this sentence as the objectionable
'Every existing object necessarily exists'. But within f!) it is to be
read as 'Every object is necessarily identical with something',
which is no more objectionable than the claim that every object
is necessarily self-identical (' (x) 0 (x = x)', which is also modally
valid). In (!), to say that every existing object necessarily exists we
need to write (x) (E!x ::l DE!x), and this is not modally valid.
The Barcan formulas all hold in (!): the necessity sign
commutes with the universal quantifier and the possibility sign
commutes with the existential quantifier. These are very useful
FURTHER DEVELOPMENTS OF THE LANGUAGE (f) 101
principles to have at hand, since they considerably simplify the

choice of axioms. They are normally thought to be objectionable
since, for example, they let us infer (x) 0 (3y) (x = y) from O(x)
(3y)(x = y). But, as was just pointed out, this is not objectionable
within the present context.!
Once we have modalities, we are only one step away from
being able to define 'possible object'. The further step involves
enriching ([) even more. Let us add to (9 a new style of variable to
range over sets ofnuclear properties; -l)11 use the Greek letters Xl' ;(2'
X3> • • • for this purpose. We also add the sign E for set member-
ship, and we suppose that set membership is rigid within modal
structures. That is, we suppose that in any modal structure if a
property is a member of a set in one possible world in the struc-
ture, it is a member of that set in all such worlds. This encapsu-
lates the idea that a set is constituted by its members, and it
validates the new axiom (ql) (X)(ql E X == D ql E X). Then we can
define:
x is possible = df (3 X)((q) (q E X === qx) &
<:; (3y)(E!y & (q)(q EX::> qy))).
This just puts into symbols the definition from chapter I: an
object is possible just in case it is possible for some real object to
have, all the properties that it actually has~2 Then we can prove
quite trivially that all existing objects are possible. And we can
also prove that there are impossible objects as well. For axiom
I. There is one respect in which some may find the modal semantics
objectionable: if rp is a sentence which says something about the cardinality of
the set of objects, then 4J ~ Olft will be modally valid. For example) if 4J says
that there are at least 64 objects, then it is a theorem that <P is either false or
necessary. I don't find this particularly objectionable since I think it plausible
that it is necessary that there are an infinite number of objects (since it's neces-
sary that there are an infinite number of nuclear properties). The important
thing is that the system does not say anything about how many objects exist.
Probably there are ways to complicate the semantics so as to achieve more
neutrality about the cardinality of the set of objects.
2. Recall chapter I, note 4- It would be possible to define "possible
object" without using set variables if we had an "actuality" operator in the
language (see Segerberg [TDML]); then the formulation would be: x is
possible =df 0 (3y)(E!y & (q)(Actually (qx) => qy)).
OB] tells us that there is, in theactualworld, an object x which

has all nuclear properties. But it is easy to show that necessarily
no existing object has all nuclear properties, so- no object in any
world has all the properties that x has here, and so x is, by defini-
tion, not possible.
Ironically, we cannot prove (as far as I can see) that there
are any "impossible" properties; in particular, we cannot yet
prove that there are any extranuclear properties which no object
could possibly have, and we cannot prove that there are any
nuclear properties which no object could possibly both exist and
have.. The reason for this is fairly easy to see. We can show only
that there are properties ofsuch and such a kind by means of the
abstraction axioms.. Being axioms, they are necessary, and hold
in all possible worlds. And, for example, in the case of extra...
nuclear properties, this means that in every possible world there
is a property that no object has (O(3P) (x) ",Px, which follows
from the version of the abstraction axiom that says 0 (3P) (x)
(Px == x :;6 x)). But we cannot switch quantifiers here; we do not
have that there is a property which, in every possible world, no
object has: (3P)O(x) -s Px.. 3 As far as I can see the semantics
also fails to require this, for we have so far placed no special
constraints on how the extensions of any properties might vary
from world to world. This is not the way in which we normally·
think about properties; we expect there to be a property, say, of
being-both-existent-and-not-existent, which has the same, null,
extension in every possible world, but we have not guaranteed
3. One must be careful here to avoid illicit use of the "rule" of existential
instantiation. For example) one might try to prove (3P)O(x)-Px as follows:
(3P)(x)(Px == x :1= x) by AB(E)

(x)(Px =: x :1:- x) by existential instantiation
(x)~Px by predicate logic
O(x)A.tPX by necessitation
(3P)O(x) ~Px by existential generalization
All but the second of these steps are permissible.' In the official set of rules and
axioms we do not have a rule of existential instantiation; although I have used
this "rule" in examples occasionally, it is used there as a shortcut rule, a rule
that does not work in all contexts. I have done this for illustrative purposes
only.
this. We have only guaranteed that in each world some property

or other will have null extension. In section 2, we will strengthen
our axioms so as to accommodate this intuition, and in section 5,
we will strengthen the semantics in the corresponding way (by
placing additional constraints on our modal structures).
2. Abstracts and Complex Properties
Given any wff ~ of @ there is an extranuclear property which is

true of exactly those objects that satisfy <P (axiom AB(E) says
this). It would be handy to have a means of referring to such a
property within (!), and for this purpos~ I will introduce property
abstracts. They are formed by preceding a wffwith the symbol 'A'
and a variable, and enclosing the whole thing in square brackets,
with the understanding that [Ax~] is to be a (complex) extranu-
clear predicate which is true of exactly those objects which satisfy
ifJ. And, in general, [Ax-I ... xn~] is to be a (complex) n-place
extranuclear predicate which is true of exactly those n..tuples of
objects which satisfy t/J.
These abstracts are meant to mirror certain devices of
English. In English we have a host of simple predicates: 'is
round', which stands for roundness; 'is a mountain', which
stands for mountainhood; etc. But we also have various complex
predicates; the phrase ~is a tall woman who supports her mother'
stands for the property of being a tall woman who supports her
mother. Within (!) we have, say, t, which stands for tallness; ta,
which stands for being a woman; and so on; but without ab-
stracts we have no means of forming a complex predicate out of
these to stand for the property of being a tall woman who sup-
ports her mother. With abstracts we can write [lx{tx & wx &
(3y)(myx & sxy))] to represent that property. This is the intent,
anyway; we will need some extra axioms to guarantee that the
abstracts work as intended.
One obvious axiom is simply the symbolic version of ,exactly
those objects which satisfy ~),
and in general:
AB(E) *
Hereafter, AB(E)* will replace AB(E), which it entails.. Since
AB(E)* is an axiom, its modalized form will be a theorem, But
then, since [Axl . . . xn~] is now considered to be a genuine
predicate term, it is subject to existential generalization, and we
can now prove that there are "impossible" properties. For ex-
ample, the necessitation of one instance of AB(E) * is:
D(y) ([Ax(E!x & ~E!x)]y ;: E!y & ",E!y).
By existential generalization this yields:
(3P)D(y)(Py == Ely & "'E!y),
from which one can easily prove r'
(3P)O(y) ~Py.
Abstracts are equally appropriate for nuclear relations, pro-

vided that we ensure that they work properly. One way to ac-
complish this would be to have a nuclear abstraction operator,
AN, which satisfies:
AB(N)* (Yl) . · · (Yn)(E!Yl & . . . & E!Yn
::> ([ANxl . · . xnq)]yl · . · y~ == ~{f:: :{~)).
We could introduce such notation as primitive, but -it is simpler

to regard [ANxI . . . xn<PJ as an abbreviation for w[Axl · · · xntP],
the watered-down version of the extranuclear abstract. For then
AB(N)* follows immediately from AB(E)*, together with the
watering-down axiom, WD, from chapter 4:
WD: (pn)(Xl) . . . (xn)(E!Xl & .. ~ & E!xn
:::> (Pxl .... Xn == wPXl · . 4 Xn »).
One nice use of abstracts is to define certain properties. For
example, if P is an extranuclear predicate, I will write P for "the
4.. The proof is easier if done in this order:

(Y)([Ax(E!x & I'IE!x)]y == E!y & "'VE!y) by AB(E)*
oi- [Ax(E!x & f'WE!x)]y by predicate logic
rx» .-v [h(E!x & ~E!x)]y by necessitation
(3P)D(y)-Py by existential generalization
FURTHER DEVELOPMENTS OF THE LANGUAGE @ 105
negation of P," and define this as (itx '" Px]. Then, necessarily,
P will be true of any object just in case P is.not true of that object;
that is, 0 (x)(Px == Px). Notice however that the latter formula
I'V
does not by itself uniquely single out P. That is, it is not generally
the case that:
(~) (QJ(D(x)(Qx == ",Px) ::> Q = p).
More accurately, this formula is not a theorem, and I do not in-
tend it to be. For I do. not believe that necessary eoextensiveness
between properties is a sufficient condition for the properties be-
ing identical. For example, I think there is a difference between:
being a (real) villager
and:.
being a (real) villager of a village in which no (real)
villager shaves all and only those (real) villagers who do
not shave themselves,
even though it is logically necessary that they are coextensive..
Likewise, their negations should be different properties, and
this would not be possible if (~) were true. (1 don't want to
misread here. The official theory given leaves it open whether or
not necessary coextensiveness constitutes a sufficient condition
for the identity of properties. One may take either side of this
issue. I am taking the negative side, in part for reasons having to
do- with the presence of predicates within the scope of proposi..
tional attitude verbs. This will be discussed fu·rther in section 3-)
We can define complex nuclear .propeFties in an analogous.
manner; for example:
P =df [ANX I'V pxl·

p& q =df [itNx(px & qx)].
But some care must be taken in interpreting these notions, For

example, although jJ is true of exactly those existing objects that p
is not true- of, it is not true of exactly those objects that p is nee true
of. Indeed, there is no nuclear property which fulfills this role?
.no q such that:
(x)(qx == '" px).
106. FORMAL DEVELOPMENT
This is easily seen from considering axiom OB]: for any p and q
there is an object which 'has both. Ifwe insist that the negation of
a nuclear property p be true of all and only those objects that p is
not true of then we will either have to conclude that p has no
negation, or that its negation is extranuclear (perhaps [Lt px]). "'J
This is one reason why I expressed reservations in chapter I about

talking about the negation of a nuclear property. Another reason
is that there is no reason to suppose that the negation of the nega-
tion ofp is identical with p, though, .of course, the negation of the
negation of p will necessarily apply to the same real objects as
does p itself. Nonetheless, I think that the notion defined by p is a
useful one, even though it might not be fully appropriate to call
it negation. I will call it negation, but the reader should keep the
above reservations in mind.
We now have the means to define what it means for an
object to be complete: 5
x is complete = df (q)(qx V qx).
It is then easy to prove that all existing objects are complete, that
some objects aren't complete (just take the null object), and that
some unreal objects are complete (take the object which has all
nuclear properties). It is a good deal harder to show that some
objects which don't exist are complete, possible, and logically
closed (this was the "exercise" in chapter I .that was deferred to
the present chapter). First, we need to define 'logically closed' .
As a preliminary; we define:
xentailsp =dfO(y)(Ety& (q)(qEX::J qy) ::>py).
Then we have:
x is logically closed = df (p) ( (3 X) ((q)(q E X :J qx) & X
~ . entails p) :::> px).
That is, x is logically closed if and only if it has every property
which is entailed by some properties it already has.
There is no way to prove from the assumptions made so far
that some nonexistent object is complete, possible, and logically
5. This is the definition from chapter L In this connection recall note 2
from that chapter.
FURTHER DEVELOPMENTS OF THE LANGUAGE Ij) 107
closed. This is because of our excessive neutrality about what is

modally true. For example, nothing has been said that would
rule out modal structures which consist of nothing but a single
world. In such a structure, necessity coincides with truth, and
the complete possible objects consist exactly of the existing ob-
jects.
But for the intended modal structure, we may reasonably
make some more assumptions. I will make this one: there is at
least one nuclear property which no real object has, but which is
such that it might have been the case that some real object had
it; that is:
(3q)( ~ (3x) (E!x & qx) & 0 (3x) (E!x & qx)).
Examples abound: being a unicorn, being a centaur, being a
woman over 17 feet tall) being a pink frog, . ~ ~ , so I think this
is a plausible assumption. Now any object which is possible and
complete is also logically closed, so we only need to produce an
object which is both complete and possible.f Here is a proof
strategy. Given that p is a property that no real object has but
that some might have, look at a world in which this happens..
Let Xo be the set of nuclear properties that some p-exemplifier
has in that world, and let Xo be the object that, in this world, has
exactly the properties in Xo. Then Xo will be an object which, in
this world, is nonexistent, complete, and possible (and logically
closedj.?
6. It will be a theorem of the extended (!) that every object which is com-
plete and possible is logically closed ..
7. This proof is formulable within the extended eJ; it goes roughly as
follows:
(I) -(3x)(E!x & px) & <> (3x)(E!x & px) assumption,
(2) (3x) <> (E!x & px) from (1) and Barcan formula.
(s) <> (E!x o & pXo) existential instantiation
(4) 0 (E!x o & (3X)«q)(q EX == qxo) & pe X» set theory.
(5) c (3X)(E!x o & (q)(q EX=: qxo) & pE X) logic.
(6) (3X) 0 (E!xo & (q)(q E Xo == qxo) & p ex) Barcan formula.
(7) <> (E!.~o & (q)(q E Xo == qxo) & pe Xo) existential instantiation.
(8) (3y)(q)(qy E q E Xo) axiom OB].
(9) (q)(qyo == q e If)) existential instantiation.
Line (9) gives us the desired object Yo. We then need to show:
3. Propositions and Prepositional Attitudes
It is hard to avoid dealing with verbs of prepositional attitude-

they are too important. I have in mind here verbs that take
animate subjects and that take that clauses or whether clauses as
direct objects, verbs such as believes, doubts, wonders, says, tells.
I will add these to (!), representing them by the symbols B, B 1,
B 2 , B 3, • • • . Semantically, I suppose that they ·stand for rela-
tions that relate people to propositions. For example, 'Samantha
believes that Herman is unfaithful' says that Samantha is related
by the belief relation to the proposition that Herman is unfaith-
ful. The subject place of such a sentence seems to behave like
the first place of a nuclear predicate, and I'll treat it this way. So
if X stands for a proposition and if B n stands for a propositional
attitude relation, then Bn{X} will be a one-place nuclear predi-
cate; it will stand for -a nuclear property whose (primary) ex-
tension is the set of those people who are related by the relation
in question to what X stands for, that is, ext{I(Bn{X})) =
{xl<x, I(X) E ext(I(Bn) ) } .
All this supposes that we have something in the language to
stand for propositions. I'm going to suppose that the sentences of (!)
do this. This sounds slightly peculiar, since traditionally sentences
either stand for nothing or else stand for truth values. There are
two ways to dispel the peculiarity. First, one could regard it as a
mere matter of terminology; perhaps my 'stands for' is just an
idiosyncratic misspelling of 'expresses'.. Second, one could sup-
pose that it is really that clauses that stand for propositions) and
within (9 the word 'that' is always written in invisible ink. There
(IO} Yo doesn't exist; suppose otherwise. Then -PYo, by (1). Then

~b ~ l<u by (9). But then, by rigidity of set membership, OP ~ lo)
which con tradicts (7).
( t 1) Yo is possible : conjoin (9) with (7) and existentially generalize, to
get the defini tion of ':'0 is possible."
( I 2) Yo is complete: Otherwise q0 '$ Xo & q0 ~ Xo for some. q0) and
necessarily so (by rigidity of set membership, and using (9»)·
This, together with (7), yields 0 (E!x o & (q)(qe lo == qxo) &
qott Xo & qo' Xo), and then 0 (E!x o & "'qrro & ,.." qoxo), which
con tradicts an easily proved theorem that (necessaril y) all
existing objects are complete.
are more complex proposals that are fairly natural, but I'll avoid
them because of their complexity.
Saying that sentences stand for propositions suggests treating
them as degenerate cases of predicates. Our treatment of pred-
icates has been this: when a is an n + r-place predicate (with
n > 0) and 'C is an object word, a~k is an n-place predicate which
stands for the result of plugging up the kth place of I(a) with
1('C), that is, which stands for PLUG(I(a), 1('C), k). We can par-
allel this for sentences: when a is a one-place predicate and
'C is an object word, a-r- 1 is a sentence which stands for the proposi-
tion got by plugging up lea) with I('C), that is, which stands for
PLUG(l(a), I('!), I). Then, just as there were conditions on the
extensions of complex predicates so produced, there will be con-
ditions on the truth values of the sentences so produced. Let me
call the truth value of a proposition its extension; then we have
the condition that ext(I(a~l)) must be T just in case a~l is true
according to the truth definition given in chapter 4, section 3,
and F otherwise.
We will also expect sentences of other forms to stand for
propositions, and for their truth values to turn out right. For
example, I(rp & </1) will be a proposition such that ext(/(ifJ & cjJ)
= T if and only if ext(I(ifJ)) = ext(I(if;)) = T. The same holds
for quantified sentences. For identity sentences, we should
extend the PLUG relation so that PLUG(X, Y, iden) is a prop-
osition such that ext(PLUG(X, T, iden) = Tif and only if X
= T, and then we add that a = (3 stands for PLUG(I(a), I({J),
iden). (This 'iden' is just a dummy symbol to preserve the re-
quirement that PLUG is always three-place.) These conditions
are all spelled out in detail in section 5.
We are now in a position to introduce our strongest version
-of the abstraction principle for extranuclear properties. We can
now say that x's having the property of being famous is the same
proposition as x's being famous, and that this sort of identity holds
for all such property descriptions :
AB(E)**
All earlier versions of AB(E) follow from this one. (Notice that
this does not entail a similar strengthening of AB(N) *.. ) Axiom
AB(E)** coheres best with a view of extranuclear relations as

propositional functions, functions which map n-tuples of objects
to propositions. This view has mainly convenience to recommend
it; it allows us to give some very natural identity conditions for
extranuclear relations. That is, two such relations are identical
when they map the same objects to the same propositions:
ID..E (xl) ~ · · (X n ) (PXl ~ · · Xn = QXl · .. Xn ) ::> P = Q.
(We may also wish to strengthen PLUG(E) by replacing its
biconditional by an identity sign . )
The axiom ID-E allows us to establish certain identities that
have not yet been provable. For example, we can show that, for
any extranuclear predicate P, the following principle holds:
P = [AXPX].
For AB(E)** tells us:
(Y)([AXPX]Y = Py),
and ID..E lets us infer the desired conclusion.
One result of this treatment is that substitutivity of identity
still holds saloa oeritate in all contexts. This sounds initially quite
surprising, but it turns out to be innocuous:
( I) In case a and f3 are predicates, we have that if a = {3,
then a and f3 are intersubstitutable everywhere; this was why I
resisted making necessary coextensiveness a sufficient condition
for property identity. Agatha may believe that George is a (real)
villager without believing that he is a (real) villager of a village
in which no -villager shaves all and only those villagers who do
not shave themselves. Interestingly, if a and (3 are one-place
nuclear predicates, and if (x) (ax == (3x), then a = {3, and a and {3
are everywhere intersubstitutable, But recall that the variable
x must range over all objects, and so the standard sorts of coun-
terexamples to such cases do not apply (e.g., if a stands for
being a creature with a heart and (3 for being a creature with a
kidney then the quantified biconditional is not true),
(2) If a and f3 are sentences, then a = {3 only if a and ~
stand for the same proposition, so their intersubstitutability is
to be expected.
FURTHER DEVELOPMENTS OF THE LANGUAGE (f) III
(3) If a and (J are object words, then it looks as if there will

be trouble.. Suppose that Cicero is Tully; this shouldn't let us
infer that Agatha believes Tully is Roman just because she be-
lieves Cicero is. Right, and this is why I am not going to use
object constants to symbolize English proper names, names such
as 'Cicero' and "Tully'. In fact, I am not going to use object
constants for anything in the representation: of English, for I
don't think there is anything in English that resembles them. In
my view, we do not yet have any means within (J) to accurately
represent English proper names (letalone definite descriptions);
this topic will be taken up in section 4. As for object variables,
they behave something like certain anaphoric pronouns of Eng..
lish, certain pronouns which "refer back" to antecedents. This
can be illustrated by a standard example of a de re-de dicto
ambiguity.. The sentence, ']oan believes something is in the
corner' is .ambiguous; it can mean either:
(A) Joan believes that at least one thing is in the corner.
or:
(B) Something is such that Joan believes that it is in the
corner.
Using '1' for 'is in the corner', we can symbolize the first as the
de dicto:
(A') jBn{(3x)Ix},
and the second as the de re..
(B') (3x)jB n{Ix}.
In (B') the x inside the curly brackets plays the role that the 'it'
does in (B).
4. Singular Terms
In this section I want to add to (!) some singular terms: proper

names and definite descriptions. They are to mimic proper
names and definite descriptions of English, and so they will
Portions of this section are also in Parsons [RNO].
behave differently from the object words that are already in @.

For one thing, they will manifest something like scope ambigui..
ties, to be discussed later in this section.. For another, some of
them may simply fail to refer to anything. Of course, the object
words of (f) may already fail to refer to things that exist; that is,
they may succeed in referring to things that don't exist, but this
is quite different from failing to refer at all. Most Anglo-American
philosophers haven't believed in nonexistent objects, and so they
have generally tried" to treat singular. terms which refer to non...
existent objects on a par with those which fail to refer to anything
at alL This has generally resulted in theories that don't do justice
to either sort of term (though occasionally one sort would receive
a fairly "nice treatment at the expense of the other). This situation
has been fostered by the interchangeable use of two paradigms:
'Pegasus' and 'the present king of France'. The former refers to
a nonexistent object, whereas the latter was (apparently) origin-
ally used as a genuine nondenoting phrase; but because of
decades of use as a philosophical example, one can now almost
make a case for the view that the phrase 'the present king of
France' refers to a character of recent philosophical mythology
(see Linsky [R&R] p. 85). I think that this overuse has corrupted
our intuitions, and made the phrase practically worthless as a
test case for a theory.. So I'll avoid it and use other examples.
I want to take as a guide certain observations about how
English speakers distinguish failing to refer from referring to the
nonexistent. I think that people behave quite differently when
they fail to refer than when they refer to something that doesn't
exist; that is, they react differently when they realize what
they've done in each case.f Here are two illustrative conversa-
tions; you are supposed to find speaker A's reactions normaL
8. Throughout this chapter when I discuss "referring,' it is the de re
7
construal of "refers to" that is at issue, not the de dicta sense. In the de dicta
sense I take it that it is quite uncontroversial that we can easily "refer to"
nonexistent objects. For example, I refer to round squares in this de dicta sense
just by saying, "Round squares have always troubled philosophers." I suspect
that some of the antiorthodox arguments in the literature have only directed
themselves to establishing that we can refer to nonexistent objects in the de
dicta sense of "refers to," and, although the logic of these contexts is problem-
atic, its blatant nonextensionality shows that it need not disconcert the ortho-
doxy.
FURTHER DEVELOPitJ:ENTS OF THE LANGUAGE (9 113
A.. The man in the doorway over there looks pretty silly.
C. But there is no man in the doorway. over there.
A. (Looks again.) Oh! I thought there was; I was wrong.
B. Does he look anything like your department chairman?
A. Who?
B . The man in the doorway over there.
A. There isn't any man there; I was mistaken about that..
B. Well, he doesn't exist, but he's there, isn't he?
A. (Exasperated.) Look, I was talking about a guy who
exists; that is I thought I was, but I was wrong, I wasn't
talking about anybody. I can't tell you what "he" looks
like because there's no "he" to look like anything!
Now that was supposed tobe a case of failure of reference;

the speaker was trying to refer to someone, but he just made a
mistake and failed to do so. (The reference to existence is not
essential to the example; speaker A could have equally well
failed to refer to "the dragon in the Sherlock Holmes novels.")
When" confronted with questions about the object he was refer-
ringto, he treats the questions as spurious (i.e., he does this once
he has realized his mistake).
. Now here's another case:
A. The unicorn I dreamed about last night looked pretty

silly.
C. But there are no unicorns.
A. So what?
C. Well, there aren't any unicorns, so there couldn't be
any" such thing as the unicorn you dreamed about last
night. So "it" couldn't possibly have looked silly.
A. Come on, it's not a real unicorn, it's one I dreamed
about.
B. Did it look anything like your department chairman?
A. No, actually it looked a little bit like my hairdresser.
In this conversation speaker A rejects the contention that he has

failed to refer to anything, though he grants that what he is
referring to doesn't "exist. And he treats questions about it as
perfectly reasonable and straightforward. .
With these conversations' as a guide, let me turn to the
official development of definite descriptions. To begin with, I'm

going to write them in the customary manner; namely, if you
have a formula ifJ, you can put an iota and a variable in front
of it like this: (Jx)ifJ; and you read it 'the thing such that <jJ', or
words to this effect. For example, you read (7X)(WX & Hx) as
'the thing such that it's winged and it's a horse', or just 'the
winged horse'. (Although, actually, if I were to symbolize most
normal uses of 'the winged horse', I would symbolize the literal
sense of something like 'the mythical winged horse' in order to
do justice to the intent of the speaker.) And the semantical
account of these definite descriptions is to be about as common..
sensical as possible; namely, (7x)ifJ refers to the unique object that
satisfies rp, if there is one, and, otherwise, (7X) t/J just doesn't refer
at all.
Now we ought to be able to make sentences using these
things, and here I'm going to do something just a bit unortho-
dox; in order that these new singular terms not be confused with
our old object words, I want to put definite descriptions infront of
the predicates theycombine with, as we do it in English. The rule
is: if a is a one-place predicate (either nuclear or extranuclear),
then (')x)<pCt is a wff. So if we want to write 'The man in the
doorway is clever', we can say:
C'x) (Mx & Ix)C.
Again, though, that isn't the best way to symbolize that English
sentence if someone were to use it in an ordinary real-life situa-
tion. After all, there are going to be lots of unreal men in the
doorway, and it is unlikely that the speaker would be referring to
any of them. Furthermore, their presence makes the definite
description written above be nonreferring (there is no unique ob-
ject which satisfies the matrix), But, as in the first conversation
above, our utterances are normally subject to something like the
"prejudice in favor ofthe actual" alluded to by Meinong. Namely,
we mean something like 'the existent man in the doorway';
we don't bother saying 'existent', because the context makes it
clear (just as the context may make it clear in the winged horse
case that we mean a mythical winged horse). So the right way to
FURTHER DEVELOPMENTS OF THE LANGUAGE (!) 115
symbolize the most natural use of the sentence in question would

be:
(7X) (E!x & Mx & Ix)C.
And now, if I'm lucky, my definite description will refer to some..
body, and that will happen just in case there is one and only one
real man in the doorway. Then maybe what I say will be true.
when will it be true? This much is clear: if (1X)<p refers to an
object, then (7x)ifJa will be true if the object referred to has the
property that a stands for, and false if it lacks the property that
a stands for. But what if (Jx)<jJ fails to refer? Then the sentence is
untrue; but is it untrue because it's false, or is it untrue because
the sentence lacks truth value altogether?
I really don't know, and I suspect that the linguistic data
won't decide for us. In particular, I can think of three different
theories., each ofwhich might be viable:
(I) When (Jx)ifJ fails to refer, (Jx)ifia always lacks truth
value.
(2) When (7x)ifJ fails to refer, ('Jx)<fia is always false.
(3) When (Jx)if> fails to refer, then sometimes (Jx)</Ja IS
false, and sometimes it lacks truth value.
The standard· objections to (I) are that certain uses of nonrefer-
ring definite descriptions yield sentences with truth value. Some
purported examples are:
(a) The man in the doorway doesn't exist. True
(b) I had dinner with the king of France. False
(c) The present king of France is a monarch. True
I find these examples inconclusive. There are two troubles with
(a). First, ifsomeone used (a) in the knowledge that the definite
description was nonreferential, they would tend to put scare
quotes around the definite description. This suggests that in such
uses the sentence has an unusual logical form. (For example, per..
haps the scare quotes suggest that a metalinguistic claim is being
made, such as "'The man in the doorway' doesn't refer"; see
Matheson [SST].) Second, the sentence contains a negation, and
the part being negated ('The man in the doorway exists') could
lack truth value here if the negation were "wide" or "exclusion"
negation.. With regard to (b) and (c), I think they trade on our
tendency to treat the hackneyed phrase 'the king of France) as if
it referred to a nonexistent object. If we change it to a real-life
example, and put it into a real-life context, the claimed truth
values become much less plausible, For example, change the
definite description into 'the man in the doorway' (and change
'monarch' to 'man'), and try inserting these examples into the
first conversation given above; I think the advertised truth values
will lose their obviousness.
The trouble with account (2) is that it needs to be supple..
mented by an account of why people treat questions as spurious
that utilize definite descriptions that are known to fail to refer.
But that isn't too hard; we normally assume that our singular
terms refer, and to simply answer yes- or no to such a question
would be extremely misleading, at best. So maybe speaker A
won't say no when B asks if the man in the doorway looks like
A's department chairman because (i) it would give the impres..
sion that the man in the doorway looks different from A's depart-
ment chairman, and (ii) it would give the impression of convey-
ing substantive perceptual content, rather than saying something
which is "automatically" true because of failure of reference. If
this explanation were correct, then it would be all right for A to
answer no, just as long as he went on to explain that there was no
such man, And I think that this would be a natural thing to do
(but this doesn't establish account (2), because we often say no
just .as a kind of generalized protest reaction).
I am going to adopt the sec·ond account; if (?x)tjJ fails to
refer, then (~x)rjJa is automatically false. This is just because I
want to have the simplest theory that is consistent with the data.
Once we allow truth-value gaps) then all kinds of complications
creep in. For example, we need to say what happens to complex
sentences whose parts lack truth value, and there is no consensus
as to how to do. this. My enterprise is complicated enough without
burdening the reader with additional complications such as.
these.
My choice allows a nice axiomatization, namely:
FURTHER DEVELOPMENTS OF THE LAJVGUA"GE (f) 117
DD D«7x)sba == (3y)«(x)(<ft == x = y) & ay)).

This is one of the "standard" elimination axioms for definite
descriptions, though of course the "elimination" only works for
extensional or modal contexts, (In other words, if we were to
excise epistemic operators from the language, we could always
view (7x)<fia as an abbreviation of (3y)«x).(<j> == x = y) & ay).
This would be very similar to Russell's theory of [OD].)
What happens to all those scope distinctions that are typical
of theories of definite descriptions? Well, as Ron Scales has
pointed out (Scales [A & E]; see also Thomason and Stalnaker
[M&R])} we can get much the same effect by the use of our
abstracts. For example, the sentence:
Agatha believes that the tallest spy is a spy
is ambiguous. Its de dicto reading can be symbolized:
aB{ (7x)ifJS},
where (:Jx)ifJ stands for 'the tallest spy', say (Jx)(E!x & Sx &
(y)(E!y & Sy & y :F x ~ Tjx)). But the sentence also has a
1'"0.1
de re reading: Agatha believes of the tallest spy that he or she is a

spy. We can write this as follows: first, we use abstraction to sym-
bolize the de re property of being believed by Agatha to be a spy:
[AxaB{SX}],
and then we say that the tallest spy has that property:
(Jx)p[AxaB{Sx}].
This gives the effect of descriptions having scope, but without
forcing us to consider them as incomplete symbols. And that, in
turn, allows us to "solve" one of Russell's problems (from Russell
[OD]), a problem that he himself failed to solve (at least to many
people's satisfaction). Namely, how can we write 'George IV
wondered whether the author ofWaverley was F' so that this is
distinguished from 'George IV wondered whether Scott was F'?
Russell proposed that we write it:
g wondered whether (3y)((x)(x authored Waverley ==
x =y) &y is F).
But most people don't believe that George IV was wondering

this at all (see Linsky [R] chap. V). We have available an alter-
native symbolization:
g wondered whether ('x) (x authored Waverley)F.
Although the propositions wondered are necessarily equivalent,
they are not the same. (The difference between the latter sym-
bolization and the symbolization of 'George IV wondered
whether Scott was F' will be simply that a different singular
term, 'Scott', will replace 'the author ofWaverley'; these have dif..
ferent senses, and it is the sense of a singular term which contrib..
utes to the proposition being expressed, not just its referent.. For
the treatment of names, see below.)
Our use of abstracts to represent the de re readings of various
propositional attitude constructions may look somewhat ad hoc;
the order of the symbols departs enormously from that of their
sources in the English sentence being represented. This makes one
wonder whether there is any systematic way to treat these con-
structions. The answer is yes, at least for one fairly significant
fragment of English. Such a fragment is given in the appendix,
along with a systematic way to relate formal system sentences
with sentences of the fragment of English.
I think that the treatment given of definite descriptions is a
fairly reasonable representation of English usage. But there is
another usage that it doesn't capture, and that has been of some
interest to us. This is perhaps Meinong's usage when he speaks
of "the round square" or "the gold mountain." The trouble is
that there are lots of round squares and lots of gold mountains,
and so (JX) (x is round & x is square) and (7X) (x is golden & x is
a mountain) both lack reference. Nor does this seem to be a case
where filling in some extra predicates will help; filling in things
like 'imaginary', 'possible', 'existent', etc., all fail to produce
descriptions which refer uniquely.? Ifwe want to allow such de..
scriptions to refer, some other treatment is necessary. Let me
9. This problem at the level of language mimics a phenornenological

problem with which Meinong struggled; see Findlay's discussion of Meinong's
notion of a "completed incomplete object" in Findlay [l'drO V] chap. VI.
See also Grossman (M] p. 2IO.
FURTHER DEVELOPMENTS OF THE LANGUAGE f!) 119
suppose that this is a special use of 'the', the "Meinongian" use,

and let me symbolize these definite descriptions as (mx)q,. What
is to be their semantics?
One approach, due to Routley [EM]], does not see it as a
special use of the word 'the); but it won't hurt if we so view it.
The idea is that every such definite description, no matter what it
looks like inside, refers to an object. And every such object has
the property attributed to it by the description; in other words,
the object referred to by (rx)t/J always has the property [Ax~] (I
am using 'r' instead of 'm' or ',' so as to avoid confusion with
other accounts). The difficulty with this proposal, from the point
of view of the theory already developed here, is not that it says
anything incorrect, but rather that it does not say enough. It
does not say enough to identify which object is being referred to by
the definite description. Since Routley does not permit unre-
stricted abstract elimination, one cannot even hem in the possible
choice of a referent to those objects which satisfy the description..
(The restriction on abstract elmination is intended to avoid the
inference from '(rx)(E!x & x is golden & x is a mountain)
[).x(E!x & x is golden & x is a mountain)]', which Routley
accepts, to '(rx)(E!x & gx & mx)E! & (rx)(E!x & gx & mx) is
golden & (rx)(E!x & gx & mx) is a mountain', which he rejects.)
This is not necessarily a criticism of Routley's proposal within
the context of his own approach, which is quite different from
the approach taken here; but it is not a proposal that is definite
enough to fit into the semantics adopted in the present work.
My proposal, instead, is to let (mx)tjJ refer to the "minimum"
of the objects which satisfy t/J. Specifically:
If some object satisfies rp, then (mx)rjJ refers to the object
whose nuclear properties are exactly those shared by all
objects which satisfy 4>, and if no object satisfies ifJ, then
(mx)rp refers to the null object.
In case ifJ is satisfied by exactly one object, then (mx)tj> refers to
that object, just like ('Jx)<jJ. But if fjJ is not uniquely satisfied, then
(mx)p still refers, though not necessarily to an object which satis-
fies 4>. "For example, on this account, 'the real giraffe) refers to
the object which has exactly those nuclear properties shared by
all real giraffes; this object itself will be a giraffe, though not a
real one. Similarly, if 'the whale' is treated as elliptical for 'the
existing whale', then the whale is a mammal, (However, it will
not be true that the whale nourishes its young with milk, since
only some existing whales do this-the female ones with chil-
dren.) On this proposal, 'the round square' refers to the object
which is round and square and has no other nuclear properties;
the same is true for 'the gold mountain'. In fact, if PI' · . · .r;
are nuclear properties, then (mx) (PIX & . . . & Pox) will refer
to the object which has exactly Pb ~ . . )Pn as nuclear properties.
Probably this is not very important; I don't think it conforms to
any widespread English usage, but it is nice to have some account
of this terminology here, especially since it is terminology that I
myself have used in earlier chapters.
Finally, what about proper names? Let me symbolize them
with capital letters : A, B, C, ~ . . ; they will appear before one-
place predicates, just like definite descriptions. So we write
'Pegasus flys' just as PF. Semantically, some refer and some
don't, just like definite descriptions, except that there is no telling
from their parts what they refer to, since they have no parts. I'll
suppose that a sentence of the form Aa is true or false depending
on whether or not the object that A refers to has a or not, and
that if A fails to refer, then such a sentence is automatically false.
This makes names behave just like definite descriptions, and it
justifies the following axiom:
NAMES O(Aa == C'x)A[Ay(y = x)]a).
This essentially identifies, say, 'Fido' with 'the object which Fido
is', at least within all extensional and modal contexts.P
.,This account of names is meant to be inconsistent with
many things that are being said of them nowadays. For one
thing, it is intended that some names refer to objects that don't
10. This axiom would be important in certain inferences-for example,

in inferring (3z)az from Aa. The proof would go:
Aa
(1x)A[Ay(y = x)]a by NAMES
(3z)«x)(A[AY(Y = x)J == x = z) & az) by DD
(3z)az by predicate logic
FURTHER DEVELOPMENTS OF THE LAJVGUAGE {f} 121
exist. In particular, it is intended that 'Pegasus' refers to a certain

object, to the winged horse of Greek mythology. This has been
denied.l!
The view that such names cannot refer has been associated
with (e.g., thought to be entailed by) the view that whether or
not a name refers depends on whether our use of it can be traced
back by means of a causal chain to something like a dubbing
that takes place in the presence of the referent. This is a version
of the. "causal" or "historical" theory of names, and I think that
it is mistaken. Though I think that it only makes a small mistake
{maybe), that -is-n't really relevant to the spirit of the causal
theory. The mistake is that the referent of the name must itself
be a causal .agent in the chain. I don't think that that is right
even in the case of certain existing things. For example, the
novel The Wind in the Willows has a certain name (namely, "The
Wind in the Willows"); but if we trace back our present use of.
that name causally, we don't come to the novel, but rather to
a 'copy of the novel. The novel itself is not a physicalobject, and
doesn't enter into causal relations. But coming to a copy of the
novel is good enough; we need one more link in the chain, but
it's not a causal one; rather, it consists of something like ex-
emplification or tokening. I think that reference to Pegasus
or Sherlock Holmes is like this. We trace our use of the name
causally to the novel, or to a telling of the myth, but then, in-
stead of encountering what .Donnellan calls a "block,"12 which
is something like a break in the chain, we make one more non-
causal step to its referent. (What this step is like is discussed in
chapter 7.) If we couldn't reach Holmes or Pegasus through the
novels or myth in this way, then perhaps we wouldn't be in a
position to use these names to refer to them.
I have heard it suggested that proper names, unlike definite
I I • For example, cc Pegasus does not exist, and ~ Pegasus) does not denote.
. Some rascal just made up the name 'Pegasus', and he then pretended, in
what he told us, that the name really referred to something. But it did not"
(Kaplan [B&C&T&A]).
12. The notion of a "block" is from Donnellan [SN]. Donnellan only uses
this notion to analyze names that originate from stories which are mistakenly
believed to be true (an analysis with which I agree); he does not attempt to
show that reference to fictional objects is impossible.
descriptions, do not manifest de dicta-de re scope ambiguities. This

is sometimes thought to follow from the view that they are rigid
designators, where a rigid designator is defined to be a name that
names the same object with respect to every possible world.l''
But all that follows from this is that proper names do not mani-
fest de re-de dicta ambiguities with respect to modal notions.
Propositional attitude words are another question entirely.
Agatha may believe, de dicta, that Plato was a great philosopher,
without having any de re beliefs about him at all. And she may
believe de re that Tonto has befriended her without believing this
de dicto, if she doesn't know his name ("Who was that Indian?"
~ . . the thundering hoofbeats of the great horse, Paint . . . )..
Are proper names rigid designators? I don't know. If you
think so, then add that provision to my semantics, and add the
axiom scheme (X)(A[AY(Y = x)] :::> OA[Jly(y = x)]).
Last, I want to say that names have sense. Or at least they
are as good candidates for having sense as any other kind of
word in our language. Their having sense would explain how it
is possible for Agatha to believe, de dicto, that Cicero was Roman
without believing that Tully was, even though Cicero is Tully..
That is, it explains the invalidity of the argument:
C[,:{xT[AY(Y = X)]].l4
AB{CR}.
Therefore, AB{ TR}.
The point is that the proposition that is said to be believed in
the premise is different from the one in the conclusion, despite
the fact that C and T have the same reference; it is their senses
that contribute to the propositions in question.
The main reason given for proper names lacking sense is
that people have thought that if names had sense, then they
must be synonymous with certain definite descriptions. But
13. There is another use of the term "rigid designator" according to
which a term is a rigid designator just in case it contributes nothing to the
proposition except its referent (see Kaplan [D]); in this sense, names are
clearly not rigid designators, since coreferential names are not interchangeable
in epistemic contexts, saloa ueritate.
14. Notice that "Cicero is Tully" ·is not symbolized as C = T; indeed,
this is not even well..formed in (f). For a systematic account of how to symbolize
sentences containing the "Is" of identity, see the appendix.
FURTHER DEVELOPMENTS OF THE LANGUAGE ([) 123
there is no good reason to think this, any more than you should
think that if definite descriptions have senses, they must be
synonymous with certain names.
5. Formalization
In this section I will give a precise formulation of the syntax and

semantics of the expanded version of (f) as developed in sections
1-4.. ·. Much more could be added to the language than is given
here (e.g., predicates of propositions, such as 'is true'), but the
language is already becoming excessively complicated.
Syntax
The primitive symbols of (f) are:
( I) Object variables: x, y, Z, Xb X2' .

(2) n-place nuclear predicate terms:
Constants : p~, p~) .. . Variables: q¥) q~, .
(3) n-place extranuclear predicate terms:
Constants: p~) p~, . .. Variables: Qy, -Q2, .
(Recall that P14 is abbreviated by EL)
(4) Names: A, C, D, . . ..
(5) Propositional variables: Sh $2' . . ....
(6) Set variables: Xl' Xz, • · . .
(7) Prepositional attitude constants: B, Bb B 2, •
(8) Connectives: ~, &, V, =', =.
(9) The existential quantifier symbol: 3.
(10) The identity sign: =.
( I I) Punctuation signs: ( , ), [ , ], { , l-
(12) The watering-down operator: w.
(13) T,he abstraction sign: A.
(14) Definite description signs: " m.
(15) The set membership sign: E.
(16) The modal operators: 0, O.
The complex terms and wffs are given by:
(17) Plugging-up: If a is an n + I-place nuclear or extra..
124 FO"RMAL DEVELOPMEJVT
nuclear predicate term with n ~ I and I S k S"

n+ I, then axJ is an n-place predicate term (which
is nuclear or extranuclear, respectively). (From chapter
4, section I.) .
(18) Watering-down: If a is an n-place extranuclear predi-
cate term, then w(a) is an n"'place nuclear predicate
term.
(19) Abstracts: If 4J is a wff and if Xkl' ~ • • , xkn are distinct
variables, then [Axk1 .. • .. Xknrp] is an n-place extra-
nuclear predicate term. (Recall that [it Nx k1 • . • Xkn~]
abbreviates W{[AXkl'O ... Xkn~]); see section 2, this
chapter.)
(20) Definite Descriptions: If <P is a wff, then (7x n )ifJ and
(mxn)ljJ are definite descriptions..
(21) Prop-Att: If t/J is a wffand B n is a propositional attitude
constant, then Bn{p} is a one-place nuclear pred-
icate term.
(22) Atomic: If sn,is a propositional variable, it is a wff..
(23) Atomic: If a is a one-place predicate term, then ax~
is a wff. (The superscript is often- omitted for simplic-
ity.)
(24) a
Atomic (Subject-Predicate}: If {3 is name or a definite
description and a is a one-place predicate, then "{ja
is. a wff.
(25) Atomic (Sets): If a is a one-place nuclear predicate
term and if Xn is a set variable, then a E Xn is a wff..
(26) Atomic (Identity): .If a and f3 are both object variables,
or both n-place nuclear predicate terms, or both n..
place extranuclear predicate terms, or both" set vari-
ables, or both wffs, then a = {3 is a wff
(27) Molecular, Quantificational~ Modal: If cP and ifJ are "both
wffs and (3 is a: variable, then these are all wffs :
"J~, (if.> & ep), (tjJ V <j;), (4)::> 4;), (ifJ == <jJ), {3{J)if>,
(f3)<jJ, Dp> 0 tfJ·
Axioms and Rules

The axioms and rules have- all-been indicated previously; they are
restated here for convenience. They come in six sets.
I. First, we have some "ordinary" rules for a higher-order,
FURTHER DEVELOPMENTS OF THE LANGUAGE (9 125
sorted predicate calculus with identity. They haven't changed in

form from chapter 4, although with our expanded vocabulary
they now have a broader significance. For example, we now have
propositional variables which are wffs, and so we may apply
the universal specification rule to them. For example, from
(s)(Os :::;) s) we may infer DAPl :::> AP1. Also, wffs may flank
identities, and the laws of identity apply to them. So, for example,
from AP~ = GP! and <) AP2 we may infer 0 GP!, by rule [2. The
rules and axioms are restated below:
AX: Any axiom may be written on any line with the empty set as
the set of premise numbers.
P: Any wff may be written on any line with the unit set of
the line number as the set of premise numbers.
T: If <P is tautologically implied by some wffs on some lines in
the derivation, cP. may be written on any later line, with its
set of premise numbers being the union of the sets from
the lines containing the wffs in question.
GP: (if> ::::> ifJ) may be written on a line if cjJ appears on an earlier
line. Take as premise numbers those of the line on which
cjJ appears, leaving out any number that is the line number
of a line on which <p appears.
US: If (a)<jJ appears on a line, <p~ may be written on a later line
with the same set of premise numbers, provided that neither
{J nor any variable free in ~ is bound in ifJ~.
VG: If f/J appears on a line and if a does not occur free in any
premise of that line, then (a)ifJ may be written on a later
line with the same premise numbers.
E: If (3a)tjJ appears on a line, then >- (a) "'tP may be written on
a later line with the same premise numbers, and vice versa.
/1: a = a may be entered on any line with the empty set for
its set of premise numbers.
/2: If a = {3 appears on a line and if ifJ appears on a line, then
lfig may be written on a later line, provided that neither f3
nor any variable free in {3 is bound in <p~. As premise numbers,
take all those of both earlier lines in question.
11. Second, we have axioms governing properties and rela-
tions and their plugging-up (chapter 4, section 2, and sections
2 and 3 above):
126· FORMAL DEVELOPMENT
AB(E)** (Yl) · . · (Yn) ([Axkl · · . xkn,]yl · · · y~ =

~l: : :"'ikn), where 4> does not contain any arYl'
· · . ,Yn·
PLUG(E) * (Q?)(Xl) .•. (xn)(Qnx~l . . . x~n = Qllxjj~~l
m· m- m)
• . • Xj.1rlXj+~+l • • • X n •
n
ID-E (Q~)(Q~)«(Xl) · .. (xn)(Q¥xl · .. x~ = Q~l
· · . x~) ::> Q¥ =
Qi).
WD (Q)(Xl) . . . (X n ) (E!Xl & .... & s»; ::> (QX~l
. . . x~n == w(Q).x~l .... x~n)).
PLUG(N) (q)(Xl) . . . (Xo ) (EL~l & · . · & E!Xj_l & .
E!Xj+l & . . . & E!xn ::> (qX~l . . . x~n & E!Xj ==
qXjjX~l . . . XjJflx~tl ••. x~n)..l5
Ill. Third, we have the axioms governing objects (chapter

4, section 2) :
1Nl (Xl) (X2) «ql) (qlx1 == qlX2) ~ Xl = X2).
OB] (3x n )(ql ) (ql xn == ifJ), where tP does not contain Xn free.
IV. Fourth, we have the axioms governing singular terms
(section 4 above) :
DD ('x)</xx == (3y)(x)(~ == x = y) & ay).
NAMES Aa == (ix)A[Jy(y = x)]a.
V. Fifth, there are the normal axioms governing sets (section
I above):
EXT (q)(q E Xl == q e 1,2) :;:, Xl = X2·
ABSTR (3X) (q)(q EX == if»} where if> does not contain X
free.
RIGID (X) (q)(q EX::> Dq EX)·
VI. Last, there are the rules and axioms governing modal-
ities (section I above):
NECC If 1J appears on a line of a derivation with the empty
set of premise numbers, then Dif; may be written on
a later line, also with the empty set of premise num-
bers.
15. In PLUG (E) * and PLUG(N) the ki's and m/s must be related as
described in conjunction with the presentation of PL UG(N) in chapter 4,
section 2.
FURTHER DEVELOPMENTS OF THE LANGUAGE ~ 127
DEF ot/J == "'O~,.

PROP D~ ::> ifJ·
DISTR O(q> :::> cjJ) ~ (Oq> :::> D~).
8-5 c ifJ ::> 0 0 ~.
BF (x)O<;> == O(x);.16
As far as consistency is concerned, these rules and axioms are
pretty safe. For we can regard all the new notation of chapter 5
as abbreviations of notation that is already present in the system
of chapter 4. This is not the intended interpretation, of course,
but unintended interpretations are relevant where questions of
simple consistency are concerned. In particular, we may suppose
that:
(I) If 1> ([Axkl · · . Xkn 4']) is an atomic wff, it is an abbre-
viation of (3pn) ( (Xkl) · • • (Xkn ) (PDX k1 • .. .. xkn == cjJ) &
_tjJ(pn)) .
(2) == ifJ, whenever ifJ and </J are wffs.
(3) (7x)ifxx abbreviates (;V) «x) (rjJ == x = y) & £{Y).
(4) Ana abbreviates ('x)A:xa, where A: is a new primitive
extranuclear one-place predicate added to the vocab-
ulary.
(5) Bn{cfi} abbreviates B:, where B: is a new one-place
nuclear predicate added to the vocabulary.
(6) 0 and <> each abbreviate double negation..
(7) qn E Xj abbreviates qnXj' and Xn = Xtn abbreviates Xn =
xm • 17
16. Actually BF is derivable from the other axioms and rules; it is listed
here for reference only. The proofin the right..to..left direction is easy:
(I) (x)tP ::J tP by predicate logic
(2) O«x)<ft :::> t/J) from (1) by NECC
(3) O(x)~ ;) D O~) from (3) by UG
(5) O(x)q) => (x)DtP from (4) by predicate logic
The proof left-to-righ t is considerably more complicated; it is given in
Prior [MQSsl-
17.. This isn't quite right; the idea is to replace set variables occurring in
a wff by object variables which do not themselves occur in the wff; in case
Xn and Xn both occur in a wff, a different x"
must be chosen to replace Xn. If
derivability is in question, Xn may have to be replaced by some Xj which does
not occur in the derivation.
Then it is straightforward to show that every rule and axiom of

the system of this chapter is a rule or theorem of the system of
chapter 4, to which has been. added a thesis of extensionality for
extranuclear relations.
Semantics
The semantics is somewhat complicated because of the presence
of propositional attitude contexts, which require that necessarily
equivalent things may still be distinct. Thus we must have a
large number of independent clauses to require that the meaning
ofa complex is a function of the meanings ofits parts, even within
propositional attitude contexts, and that the extensions of com-
plexes (strictly, the extensions of their meanings) are functions of
the extensions of their parts when they contain no modal or
propositional attitude constructions.
I begin with a definition of a modal structure. The first thir-
teen parts of a modal structure can be most easily thought of as
constituting an account of some Russellian "meanings," where
the meanings are propositions, propositional functions, objects,
sets, etc. (1 call these "Russellian meanings" because it was
Russell who committed the sin of calling concrete things "mean-
ings.") The fourteenth part of the modal structure assigns "ex...
tensions" to certain of these meanings in various possible worlds
(and the fifteenth part consists of a distinguished world-the
"actual" world). The sixteenth part is an assignment ofmeanings
to the primitive constant terms of the language. This results in a
natural definition of truth in a structure and a definition of
validity.
A modal structure M is a sixteen-tuple which satisfies condi-
tions (I )-( 17) below:
(1) The first member of M is a nonempty set OB, called

the set of objects.
(2) The second member of M is a sequence Xv N 2, •••
of nonempty sets. Each No is called the set of n-place nu-
clear relations.
(3) The third member of M is a nonempty set P, called
the set of propositions.
(4) The fourth member of M is a sequence El' E 2, • • • of
FURTHER DEVELOPMENTS OF THE LANGUAGE@ 129
sets of propositional functions. Each member of En is

defined on OBn and takes values in P; En is called the
set of n-place extranuclear relations.
(5) The fifth member of M is a function W which maps
members of En to members of N n .. W is called the water-
ing-down operation.
(6) The sixth member of M is a set X of subsets of Ni. It
is required that X be of the same cardinality as OB.
(7) The seventh member of M is a set S of functions from
NI U El to JP. ·Sis calied the set of singular-term senses..
(8) The eighth member of M is a seQl:1ence NI' N 2, ..... of
sets of prepositional functions. Each member of Nu
is defined on No and takes values in P. (These
functions will be used solely for the ·purposes of defining
quantification over nuclear relations; similarly for
clauses (9)-( I I).)
(9) The ninth member of M is a sequence El' E 2, • . • of
sets of propositional functions. Each member of En
is defined on En and takes values in JP.
(10) The tenth member of M is a set C of propositional
functions.. Each member of C is defined on X and
takes values in P.
(I I) The eleventh member of M is a set P of propositional
functions. Each member of P is defined on P and
takes values in P.
(12) The twelfth member of M is a set B of functions from
P to NI. B is called the set ofprepositional attitude senses.
(13) The thirteenth member of M is a function PLUG
which satisfies the following conditions. (The condi-
tions are numbered so as to correspond with certain
of the syntactic constructions. Occasionally various
dummy symbols appear in the argument places of
PLUG; these are spelled heuristically; they may be
taken to stand for any distinct things that are not
otherwise parts of (parts of) M. Sometimes the num-
ber I is also chosen to fulfill this role.)
(13.17) If x E OB, reNn + h n ~ I, and I ~ k $ n + I,
then PLUG(r, x, k) E Ha; and if x E OB, rE E n+1,
n ~ I, and. I ~ k ;S; n + I, then PLUG(r, x, k) E En.
(13.20) If rEEl' then PLUG(iota, r, I) E Sand

PLUG(mein, T, I) E S..
(13.23) 1fT ENt UE1 and x E OB, then PLUG(r, x, I) E JP.
(13.25) 1fT E NI and X E X, then PLUG(epsi, T, X) E P .
(13.26) If c, d are both members of OB, or of Nu' or of En'
or of X, or ofP, then PLUG(iden, c, d) E P.
(13,,27) If 51' $2 E JP, then PLUG(neg, SI, I) E P, and
PLUG(and, SI, S2) E P, and PLUG(or, Sb s2)
E P, and PLUG(i~ SI, 52) E JP, and PLUG(eq, SI,
S2) E JP, and PLUG(nee, SI' I) E P, and PLUG
(poss, SI, I) E P.. Further, if Z E El U No U En U
C UP then PLUG(all, z, r) E P and PLUG(some,
z, I) E P.
(13-00) A global requirement: the order of plugging up of
an extranuclear relation should make no difference
to the resulting proposition: if re En' then PLUG
« · · ,,(PLUG(PLUG(r, Xl, k1) , X2' kz) · · " ), Xn ,
kn ) = PLUG« .... (PLUG(PLUG (r, Xj' mj ) , Xl'
ml) . · . ), X D , mn ) , where the rn's and k's are chosen
as in axiom PLUG(E)*.
(14) The fourteenth member of M is a set G, called the set
ofpossible worlds. Each possible world, w e G, is a triple
of the form (EXw , lw' ext.,') such that EXw £ OB, fw
is a one-one function from OB onto X, and ext.,
satisfies the following conditions :
(14.. 2) IfreNn, then extw(r) ~ EX:, and if rE NI' then
extw(r) = {x I x E EXw and T E fw(x)}.
(14.3). If rE En' then ex1:w(r) £; OBn and extw(A(E!)) =
EXw } and extw(r) = {(Xl' .. · , xn ) I extwCr«xb
. . . , xn » ) = T}. (See clause (16) regarding A.)
(14.4) If a E S, then extw(a) is either the unit set of some
member of OB, or is undefined.
(14-5) If SE P, then ex~(s) E {T, F}.
(14. 1 7) If 1ENn + 1 U E n+1, n ~ I, I ~ k ~ n + I, and
x E OB, then extw(PLUG(r, x, k)) = {<Xb · . · ,xn ) I
>
(Xl' • . • , Xk-l' x, Xk' . - • , xn E extw(r)}.
(14_ 18) IfreEn , then extw(W(r)) = EXw n extw(r).
(14.20) If rE El' then extw(PLUG(iota, T, I)) = the
member x of OB such that extw(r) = {x} if there is
FURTHER DEVELOPMENTS OF THE LANGUAGE(f) 131
one, otherwise undefined; and extw(PLUG(mein,

T, I)) = the member x of OB such thatf(x) = {rllr}
E N l and for every yE extw(r), r1 Ef(y)}, provided
that extw(r) =1= A; otherwise, extw(PLUG(mein, r,
I)) = the member x ofOBsuchthatj(x) = A.
(14.23) Ifr E NI and x E OB, then extw(PLUG(r, x, I)) = T
if and only if r Efw(x); and if rEEl and x e OB,
then extw(r(x)) = T if and only if x E extw(r).
(14.24) If a E Sand rEEl' then extw(a(r)) = T if and only
if extw(a) e extwCr). (It is assumed that if ext.,
Ca) is undefined, then extw(a(r)) = F,) Likewise,
if a E Sand r fi= Nb then extw(a(r)) = Tif and only
ifr efw(extw(a)).
(14.25) 1fT E NI and X E X, then extw(PLUG(epsi, T, X)) =
T if and only if rE X.
(14.26) extw(PLUG(iden, c, d)) = T if and only if c = d.
(14.27) extw(PLUG(neg, 5, I)) = T if and only if extw(s)
= F. ex~(PLUG(and, Sb s2)) = T if and only
if extw(sl) = T and ex~(s2) = T.
Similarly for the other connectives.
extw(PLUG(necc, s, I)) = T if and only if extw'(s)
= T for every w' E G; and similarly for poss, chang-
ing "every" to "some."
If r e Es, then extw(PLUG(some,·r, I)) = T if and
onlyif extw(r) =I: A, and extw(PLUG(aIl, r, I)) =
T if and only if extw(r) = OB4
If r E En' then ext.w(PLUG(some, T, I)) = T if and
only if extw(r) ~ A, and extw(PLUG(all, r, I)) =
T if and only if ext(r) = En.
If rE N n , then extw(PLUG(some, r, I») = Tifand
only if extw(r) :F A, and extw(PLUG(all, T, I») =
T if and onlyifext(r} = N n •
If rE C, then extw(PLUG(some, r, I)) = T if and
only if extw(r) ¥= A; and extw(PLUG(all, r, I)
= T if and only if extw{r) = X.
If rEP, then extw(PLUG(some, r, I») = T if and
only if extw(r) =F A, and ext.,(PLUG(all, T, I))
= T if and only if extjfr) = P.
(14.00) A global requirement: We require that ext., agree
with extw ' for any member of S U No U En U ·C

U PUB which is not the value of Ag(a) for any
assignment g and phrase a of (f). (For A and g, see
clause (16) below. This requirement isa purely
technical one; it is designed to ensure that worlds
not differ in inexpressible ways, and it is only
invoked here so as to simplify the results of chapter
4, section 3.. )
(15) The fifteenth member of M is a member Wo of G,
called the actualworld.
(16) The sixteenth member of M is an assignment function
A, defined on the primitive constant terms of (I), which
satisfies:
(i) If lX is a primitive n-place nuclear predicate constant,
then A(a) E N n •
(ii) If a is a primitive n-place extranuclear predicate con-
stant, then A(a) E En-
(iii) If (3 is a name, then A(f3) E S.
(iv) If fi is a propositional attitude constant, then A«(3) E
B.
It is also required that the assignment A determine an extended
assignment of propositions of M to the closed wffs of (f); this is
defined indirectly as follows. First, we define an assignment g
of values to the variables of (f) to be a function defined on those
variables such that:
(v) g(xn ) E OB.
(vi) g(qj) E N n •
(vii) g(Qj) E En.
(viii) g(sn) E P.
(ix) g(Xn) E X.
Then we define an extended assignment to all the terms and wffs
of f!) in terms of A and g. We will call this extended assignment
Ag, and it is defined recursively as follows (again the numberings
correspond to certain of the syntactic rules (1)-(27):
(16.1-16.7) Ag agrees with A wherever A is defined, and
with g wherever g is defined.
(16.17) Ag(a-r k ) = PLUG(Ag(a), Ag(-r), k).
FURTHER DEVELOPMENTS OF THE LANGUAGE e 133
(16.18) Ag(w(a)) = W(Ag(a)).

(16.19) Ag([Axkl ~ · · Xkn~]) = the member of En which
>
maps any given member (xl' . . . ,xn of OBn to
Ag' (ifJ), where g' isjust like g except that g' (Xki) = Xi
for each i between I and n. (Note: it is a require-
ment on M that this function be present in En;
i.e., this clause requires it.)
(16.20) Ag«1X)<jJ) = PLUG(iota, Ag([Axc;h]), r) and
Ag«mx)<jJ) = PLUG(mein, Ag([lxqS]), I).
( I 0.2 I) Ag(B n{<P}) == Ag(B n) (Ag (if;) )·
(16.23) If a is a one-place nuclear predicate term, then
Ag(a't"l) = PLUG(Ag(a), Ag('!), I); and if a is a
one-place extranuclear predicate term, then
Ag(az-1) = Ag(a)(Ag(t")).
(16.24) If f3 is a name or definite description and a is a one-
place predicate term, then Ag(J9a) = Ag({3) (Ag(a)).
(16.25) Ag(a E Xn) = PLUG(epsi, Ag(a), Ag(Xn))·
(16.26) Ag(a = (3) = PLUG(iden, Ag(a), Ag({3)).
(16.27) Ag( ~<fi) = PLUG(neg, Ag(ifJ), I).
Ag(<j> & ifJ) = PLUG(and, Ag(ifJ), Ag(cj;)).
Similarly for the other connectives.
Ag(D~) = PLVG(necc, Ag(ifJ), I), and similarly for
04>.
Ag«3xn)<fi) = PLUG (some, Ag([Axo~]), I), and
similarly for (xn)rp.
4g«3Q,j)~) = PLUG(some, r, I), where r is that
member of En such that for any r* e En' r(r*) =
Ag'(~), where g' isjust like g except that g'(Q,j) =
r*.
(Note: this clause also imposes a requirement on M:
that the appropriate r be present in En.)
Similarly for Ag( (Q,j)qS).
Similarly for the other sorts of quantifiers: (3qf)<jJ,
(3Xj)tjJ, and (3sj )<jJ .
( 17) Finally, we require that X be so related to the other
elements of M that every instance of the set abstraction
axiom ABSTR be true in M with respect to every
assignmentg (see below for the truth definition). This
completes the definition of a modal structure.
Given the above characterization of a modal structure we

can define "<jJ is true in world w of modal structure M with re...
spect to assignment s" as ((extw(Ag(~)) = T", and "tjJ is true in
M with respect to s" as "<jJ is true in the actual world Wo of M
with respect to g. U Then we naturally define:
<p is valid == df </J is true in
every modal structure with re-
spect to every assignment g.
and:
r
ifJ is a valid consequence of the set of wffs, =df for any
modal structure M and assignment g, if every wff of is r
true in M with respect to g, so is t/J.
I believe that all the theorems of this system are valid wffs, and
that any wff that is derivable from a set of wffs is a valid conse-
quence of that set. I also suspect that the theorems constitute
all of the valid wffs, and that any valid consequence of a set of
wffs is derivable from that set. But I haven't proved any of this.
One natural alteration of the semantics would be to require
that in a modal structure En be the set of all functions from OBn
to P, and that X be the set of all subsets of N v and that S be the
set of all functions from NI U El to P, and similarly for En' N n;
C, P, and B. Then the "structure" part (clauses (1)-(13)) could
be defined independently, as constituting the basic ontology of
objects and meanings; clause (14) could be an additional clause
imposed on the first (13), and it would constitute a view about
what possible worlds there are, and what they are like. Then
clauses (15) and (16) would be independent of one another; (15)
would say which world was the actual one, and (I6) would be a
purely linguistic stipulation concerning how to interpret our
language. The requirements on the "structure" that appear in
various subclauses of (16) would be redundant, and so would
clause (17). The additional strength of the new requirements on
the early clauses would, however, make the theory not completely
axiomatizable.
ApPENDIX: A Semanticsfor a Fragment ofEnglish
Throughout chapters 4 and 5 I have made various remarks about

FURTHER DEVELOPMENTS OF THE LANGUAGE () 135
how the system (!) is related to our native language. Many ofthose
remarks relied on a tradition of translation between English and
the first-order predicate calculus, but some of them (e.g., the
suggested treatment of definite descriptions, of identity state-
ments, and of de re beliefstatements) were somewhat unorthodox.
The point of this appendix is to give a systematic way to relate (f)
to English. This is not just an exercise; in. chapter 2 I suggested
that much of the motivation for believing in nonexistent objects
stems from our acceptance of certain sentences which seem to
require for their truth that there be nonexistent objects, and it is
incumbent on me to provide evidence that (f) really does provide
the resources for a semantical treatment of those sentences.
The techniques here are primarily due to Richard Monta..
gue, and the system proposed resembles that of Montague [PTQ].
The system addresses itself only to a fragment of English, and the
fragment is a very restricted one. It contains only present-tense
indicative sentences, it does not contain any ordinary adverbs,
nor any prepositions (except for the 'in' of location, which is
given an ad hoc treatment), nor any nonpredicative adjectives
such as 'genuine' or 'apparent); it does not contain any plurals,
and it lacks a host of other kinds of words and grammatical con-
structions as well. Still, it is fairly rich; for example, it contains
the sentence:
Every JDaD who dates a certain clever WODlaD who is
written about by Agatha Christie wonders whether
she loves him..
And it predicts that that sentence is ambiguous with respect to
whether it was "the man" or "the woman" who was written
about by Agatha Christie, and with respect to whether the 'she'
refers back to "the woman" or to "Agatha Christie."
The semantics js provided indirectly. I will show how to
associate sentences of the English fragment with certain sentences
of (!); then we say that a sentence expresses a given proposition if
and only if the associated sentence of (!) does. In general, a given
English sentence will be associated with more than one sentence
of (!); in interesting cases, this is because the sentence is ambig-
uous (see below).
The technique goes roughly as follows. First, we will have

lists of English words which are assigned to certain syntactic
categories of English, and which are also given "translations"
into certain formal symbols (which are typically part of the
vocabulary of (f)). For example, we will have the following, where
'X*' represents 'the translation of X':
Holrnes E DP ("denoting phrase") Hobnes* = HLl
its E Pro ("pronoun") its* = Xs
exists E VP ("verb phrase") exists* = E!
(The 'Ll' in the translation of Holm.es is a placeholder, not a
symbol of (f).)
Second, we will have various syntactic rules which will allow
us to generate English sentences. Specifically, the rules are used
to construct analysis trees; the top node of such a tree is the Eng-
lish sentence generated by the tree. An example of an analysis
tree is the following:
Holznes exists
HOI~ ~
"
exists
/
its exists
(As you see, proper names "get into" sentences by replacing
pronouns; there are various reasons for this, one of which is that
the position in the tree where the name replaces the pronoun
tells you its "scope.") Corresponding to each syntactic rule for
building up trees will be a "semantic" rule which tells you
how to translate the top node of the tree, given that you know
how to translate the lower nodes. The translations associated
with the tree given above are displayed to the right of the nodes
here:
Hohnes exists H[AxSE!xsJ
HotL HLl ~xists E!xs
./
Its Xs " "eXIsts
. Et·
FURTHER DEVELOPMENTS OF THE LANGUAGE {f} 137
Some sentences occur as the top nodes of different analysis trees;

typically in such a case, the sentence will receive different transla-
tions (one per tree). If these translations express different prop-
ositions, the sentence is said to be ambiguous. (Typically, a
sentence will receive lots of translations that differ only with
respect to the choice of bound variables; this does not make the
sentence ambiguous.) Examples will be given below.
The syntactic categories of English are listed here, along
with some examples of their members, This is to familiarize the
readerwith the categories.
Adj ("Adjective") round~ ilDpossible

Nom ("Nominal"; "Common noun or noun-like phrase")
WOJDan, woman wholD SODle lDan sees, man
in the doorway
Dp· ('.'Denoting phrase"} Agatha, the detective,
every dog that is bald
VP ("Verb phrase") exists, sees it4
Quant ("Quantifier") each, the, a certain
Advs (" Sentence adverb") necessarily
Vs ("Verb with sentential object") believes that,
wonders whether
Pro ("Pronoun") j~, it1,..it2' ~ ..
TV ("Transitive verb'") occupies, dreams about
S ("Sentence"· or "Formula") HollDes exists, it14
occupies it3
Some of the examples above are syntactically simple, and

some are complex. The simple ones. are assigned to their cat-
egories one by one (or by a simple general rule}; the complex
ones are assigned to their categories derivatively by means of the
rules that we use to- construct analysis trees. The .assignment of the
simple words to their categories is given by the following lists:
Adj: round, square, blue, tall> .Greek, clever, real

(E)-, fictiona1(E), m.ythieaI(E), complete (E) ,
incornplete(E), golden, winged, possible(E),-
·iIDpossible(E)y bald; exisnng(E), happy
Nom:. god, detective, city,· philosopher, square,
mountain, spy, president, pres~dent of the

U.S., king of France, DIaD, woman, unicorn,
centaur, horse, objecteE), doorway, -thing,
-body ,
DP: Any proper name of English
VP: exists(E), runs, walks, talks
Quant: each, every, a(n), a certain, the, them' some
Advs: necessarily, possibly
Vs: believes that, wonders whether, says that
Pro ; ito, itl' it2, . · .
TV: occupies, worships(E), dreaDls about(E),
thinks about(E), W-cks, is more famous
than (E) , is in, sees, shuts, loves, knows,
writes about(E)
s: There are no syntactically simple sentences in this
fragment 4
If a is a word or phrase of English, I will use a* to indicate its

translation. Sometimes a will have lots of translations; I will
still use a* when context makes it clear which translation is
meant. Most of the words above are to be given translations into
(f) subject to the following constraints:
(1) If zr E Nom or a E Adj or a E VP, then a* is to be a

one-place nuclear predicate of (9, unless an "E" appears
after a in the list, in which case a* is to be a one-place
extranuclear predicate of (!).
(2) If ex E TV, then a* is to be a two-place predicate of (f),
nuclear or extranuclear, depending on whether or not
an "E" occurs after it in the list.
(3) If er is a proper name of English, a* is to be a proper
name of (f) followed by LI.
(4) If a E Vs, then a* is to be a propositional attitude
verb of(!)..
(5) necessarily* = 0, and possibly* = <>.
(6) itn * = Xn •
(Although certain of the adjectives listed may have nonpredica...
tive meanings or uses, such uses are not realized in the present
FURTHER DEVELOPMENTS OF THE LANGUAGE (J) 139
fragment; they are always to be construed here predicatively.)

There is to be one further metaconstraint: when we go to apply
the system being described, we must use the "right" interpreta-
tion of (!) discussed in chapter 5, section I; the lists above es-
tablish the "correlation" between words of English and symbols
of (f) that was referred to in that section.
Now we need some rules for constructing analysis trees.
In each case we state the syntactic rule, followed by the rule that
tens which translation or translations are to be associated with
the output of the rule (based on translations associated with the
inputs), followed by one or more examples of uses of the rule.
The rules fall into several classes. The first class of rules lets us
construct certain primitive sentence forms and VP forms that are
similar to formulas and predicates of @ that contain unbound
variables (the subscripted pronouns of English mimic the vari-
ables of (9). Later rules will turn these into full-fledged sentences
of English.
In stating the rules, I will use the following convention:
if a is a one-place predicate term of (f) and if 7:' is a variable of (!),
then a-r° is to be (i) ea if a is not an abstract, or (ii) the result of
applying abstract elimination to a-r if a is an abstract.
Rulesfor BasicSentence and VP Forms
(RI) Pro-VP.: Ifitn E Pro and {J E VP, and if {3 is not of the

form ritn[acc], then itn[nom]~ E S.
(SI) (itn[nom]/J)* = {1*x~ .
EXAMPLE:
(Pro-VP) its[nom] exists E!xs
it~ ~ts El
This example illustrates a use of the rule in constructing an an-

alysis tree, where the translations of the nodes are displayed to
the right. The "nom" in parentheses is there to keep track of the
case of the pronoun (in this case, the case is "nominative.") The
restriction on the form of {J is to prohibit formation of sentence

forms in which both subject and object pronouns "refer" to the
same thing as a matter oflanguage (forms like it310ves it3' which
would translate into something like PXaX3). In English such forms
are always realized linguistically as reflexioes (it3 loves itself),
and I am avoiding reflexives in this fragment because they are
somewhat complicated (at least).
(R2) TV-Pro: If {3 E TV and itn E Pro, then -13 itn[acc] e V.P.
(82) (f3itn [ace]) * = [Ax([x,B*]xn ) ] , or
[Ax(X[~*Xn])]' or
[Ax(Xfj*Xn ) [,
The different choices of translations represent the different mean-
ings that result from supposing that the first place of the relation
in question is plugged tip first, or the second place, or if both are
being asserted. Actually, the reading in which the subject place
is plugged up first (the first one listed above) is probably not
realized in normal English usage, and should probably be
omitted. One might even hold that the second reading should be
omitted as well; I'm not sure about this ...
EXAMPLE:
(Pro-VP) it4[nom] sees it7[acc]
.>
it(x, -. (TV-Pro) sees it-r[acc]
/~ it
sees S 7 X7
(The reason that the top string is generated in two steps instead
of one is that, in, natural extensions of the fragment, we would
want to have the intermediate form (the output of TV-Pro)
available for modification by an adverb; this would permit the
reading of Henry shut every door quickly, according to which
each shutting was quick; this needs to be distinguished from the
most natural reading of Henry quickly shut every door, ac-
cording to which it was the whole series of shuttings that was
quickly accomplished.)
FURTHER DEVELOPMENTS OF THE LANGUAGE(f) 141
The next rule is for the is of identity:

(RS) Pro-is-Pro: If ito' itm E Pro and if n ¥= "m, then
itn[nom] is itm[acc] E S.
(S3) (itn[nom] is itm[acc])* = X n = Xm •
In pedantic speech, the second 'ace' should be changed to 'nom'.
EXAMPLE:
(R4) is-a-Nom: If a E Nom, then is a(n) er E VP.

(54) (is a(n) a) * = a*.
EXAMPLE:
(is-a (n)-Nom) is a wOlDan w
\
wom.an W
(R5) is-Adj: If a E Adj, then is a E VP.
(85) (is a)* = a*.
EXAJ.'\tfPLE:
(is-Adj) is round R
I
round R
The next rule is in a class by itself; it is the rule which in-
troduces denoting phrases into sentences. That is) this is the only
rule which permits the introduction of proper names and of
quantificational phrases into sentences. It is actually an infinite
set ofruIes, indexed by 'n', This is also one of two rules (the other
being the relative clause former) which introduces pronouns that
"refer back" to antecedents.
(R6) Quantn : If a E DP and tP E S and itn occurs in fjJ, then
t/J (a. n) e S, where t/J (a, n) is the result of replacing the first
occurrence of itn in r/J by a and each succeeding OCCur-
rence of itn by the singular pronoun which agrees with er
in gender, and agrees with the occurrence of itn being
replaced in case. (It is understood that when an occur-
rence of itn is replaced, its case marking is erased.)
Restrictions ". This rule may not be applied under any of the
following circumstances if a = every {j, or a = ea. {J,
or a = a{n)J3 and {3 is.short:
(i) The first occurrence ofitn in rp is in a relative clause.
(ii) a certain already occurs in {l.
(iii) the first occurrence ofitn immediately follows is.
(86) (<!'(a,n»)* = the result of replacing .4 in cr· by [AX n P*].
EXAMPLE:
(Quantj) Cicero is TuUy C[Ax3 T(Axs(X3 = xs)]]
/~
Cicero Cd (Quantg)
Tully T,d (Pro-is-Pro)

-.
it3[nom] is Tully T[Axs(X3 = xs)]
it 3 [noml is itg[acc]
.
It3
/ X3
.I
Its Xs
The rule Quantn cannot fully be appreciated until we have

enough other rules to provide interesting inputs for it. For ex-
ample, we cannot yet use it to produce any pronouns which refer
back to antecedents because we cannot yet generate any S's
which contain more than one occurrence of a pronoun with the
same subscript. And we can only use it to put proper nouns into
sentences because we haven't yet shown how to produce phrases
like every dog. But examples will be available shortly. Anticipat-
ing other constructions somewhat, we can explain the three
restrictions on the rule. Restriction (i) is to prevent the genera-
FURTHER DEYELOPMENTS OF THE LANGUAGE{f) 143
tion of a sentence like some man who dates every woman is

happy in such a manner that the semantics makes it mean that
every woman is such that some man who dates her is happy. It
does this by prohibiting the structure:
sOlDe man who dates every woman is happy
every woman
/ ~
some man who dates ita is happy
Restriction (ii) is to prevent every, .. " from taking wider
scope than a certain.. And restriction (iii) is to prevent the gen-
eration of a sentence like Fido is every dog (which may make
perfectly good sense, but isn't good English).
The next rule lets us make complex denoting phrases.
.a
(R7) DP: If et E Quant and e Nom, then a{j E DP.
(87) (the 13)* = ('x) (7Jx & f3*XO)L1.
(each {J)* = (every 13)* = (x)(7Jx & ~*xo :::l JxO) .
(a(n) ~)* = (a certain j3)* = (soDle ~)* =
(3x) (1Jx & (j*xo & L1xO).
Proviso: In each of the above, either x is not in the argu-
ment place of an extranuclear predicate in {3*x o
and 7J is a category word, or else x is in such a
place and "1)x &" should be omitted.
(them") * = (mx)(t3*X O ) 'il, where (l3*xO)' comes from f3*x o
by replacing every part of {3* X O of the form
lXX, where a is extranuclear, by w(a)x.
EXAMPLE:
(DP)
every WODlan w
The "category words" referred to above are exists *, fie-
tional*, D1ythica1* (probably the list should be longer), and
object*. Probably the mechanical choice of whether or not to
insert a category word in DP is not really adequate to the flexi..

bility of English. The convoluted clause after them is a (probably
vain) attempt to make the "Meinongian" the be useful, The
variable x is used for illustration; other variables will·sometimes
have to be used if 13* or the wff eventually inserted into L1 Con-
tains x bound..
The next rule is just about the most fruitful rule in the gram-
mar; it is the rule for making relative clauses, and it is also the
original source for adjectives occurring in attributive position
(preceding the nouns that they modify):
(R8) RCn : If a E Nom and ifJ E S, then:
(i) a such that ifJ(a,n) E Nom, where if/(a,n) is the result
of replacing each occurrence of itn in t/J by the singular
pronoun whose gender is the gender of a and whose
case is the case of the occurrence ofitn being replaced.
(ii) a rP' (atn) E Nom, and a that cjJ' (atn) E Nom, and a
who(D1) p' (atn) E Nom, and a which if/ (aJn) E Nom,
where ifJ' (a,n) comes from <P(a.,n) by deleting the first
pronoun that replaced itn in <jJ.
Restrictions on (ii):
(I) The rule cannot be applied vacuously; that is, at

least one pronoun must be deleted.
(2) The deleted pronoun must not be in a relative
clause within cjJ (a; n) ..
(3) who(m) may be used only if the deleted pronoun
is. a personal pronoun; the case of who(JD) de-
pends on the case of the pronoun deleted.
(4) which may be used only if the deleted pronoun
is it.
(5) The first application (where nothing comes be-
tween a and if/ «(%.0») may be used only if ifJ begins
with a DP. (Note: pronouns are not DP's.)
(S8) Regardless of the form of the output of (R8), the transla..

tion is always to be [AxQ(a*x~ & p*)].
FURTHER DEVELOPMENTS OF THE LANGUAGE o 145
EXAMPLEs:
every -body P
(I am beginning to omit case markings on pronouns for simplic-

ity.) .
(RCs)
(x) (E!x & Px:)xgLx}
The rules that we have so far can interact in reasonably intricate

and interesting ways. Here is a. partial analysis tree as an ex-
ample ;
(Quan4) Every m.aD who dates a certaia woman. that everybody knows
everybody (A)
(H)
The reader might want to check that the semantic rules provide
the following as one translation of the sentence (the letters refer
to the constituents of the analysis tree) :
(A) xZKxg.
(B) (x)(E!x & Px :::> xKxg).
(C) [Ax3(Wx3 & (x)(E!x & Px :::) XKX3))].
(D) (3y) (E!y & UJ7"& (x) (E!x & Px ::J xKy) & ~yO).
(Here y was selected in order to avoid a clash of bound variables.)

(E) .XSDX4.
(F) [Ax6(Mx6 & xflJx4)].
(G) (z)(E!z & Mz & zDx4 ::) L1z0).
(H) X5Lx4"
(1) (z) (E!z & M« & zDx4 ::::> Z;LX4).
and finally:
(J) (3y)(Ery & Wy & (x)(E!x & Px ::> xKy) &
(z)(E!z & Mz & z-Dy :::> zLy)).
Once we· have relative clauses we can play with them:
(Rg) WD: If a who is in {3 E Nom, or if a which is in {J E
Nom, then a in {J E Nom..
(RIO) WDP: If a who is ~ E Nom, or if a which is {3 E Nom,
then, if ~ E Adj, then {3 a E Nom (provided that a #- r-
thing or r-body)"
(89) and (810): There is no change of translation when (Rg)
or (RIO) is applied..
EXAMPLES:!
(WD) man in the doorway

I
ntan who is iD the doorway
(WDP) happy man

/
:man who is happy
(WDP) happy man who dates Agatha
(RC 497)
/
man who dates Agatha who is happy
~
JUan who dates Agatha
~
it is happy
497
~~
I. Guided by these examples, the reader should now be in a position to
generate A certain fictional detective is ttlore fa~ous than every real
Incidentally) there is no way in this fragment to have a relative

clause modify a proper noun (so that who is happy above can-
not modify Agatha). This is because propernames can only be
modified by nonrestrictive relative clauses, and the only relative
clauses that are dealt with in this fragment are restrictive ones .
(RI I) Adv-S: If a E Advs and ifJ E S, then a, tjJ E S.
(SI I) (rx,(p) * = a*tjJ* .
EXAMPLE:
(Adv..S) Necessarily, ever~ exists
<>:
Necessarily 0
<,
everything exists
The translation of this sentence depends crucially on the choice
of the category word in the translation of everything. If exi.~ts*
is chosen, then the resulting translation is:
o (x) (E!x & Tx ~ E!x),
but if object* is chosen, it comes out saying:
O(x)(Ox & Tx =' E!x)
a sentence I can only agree with if thing means something fairly
special.
(RI2) Prop-Att: If a E Vs and ifJ E S, then aif> E VP.2
(812) (a<fr)* = a*{~*}.
detecelve. The generation will presume that is.. Dlore famous than is to be
treated exactly like a transitive verb; this is probably incorrect (see chapter
6, section 4), but it is perhaps an approximation of the truth. A similar treat-
ment of knows Dlore about chem.ical analysis than would also let us
generate sentences like Every good Dlodern chem.ist knows more about
chemit=al analysis than Sherlock Hobnes. We can straightforwardly
generate Pegasus is the m.ythological winged horse; to get Pegasus is
the winged horse of Greek m.ythology we would have to add some way to
generate of Greek lDythology..
2. I don't think that this is the best way to generate propositional attitude
constructions. r have followed Montague [PTQ] here for simplicity. However,
amalgamating the that to the believes as we have done in the vocabulary
makes it very difficult to handle other constructions in which the that isabsent:
for example, Mary believes everything that John believes. For a discus-
sion of what I take to be a better treatment, see Parsons [77&0£].
EXAMPLES:
(Pro-VP) it3 believes that somebody is a spy
belie~mebOdY
its- - - - ; : : : : is a spy
believe~ J somebody a spy
~
(Quantg) its believes that somebody Is a spy
~
somebody (Pro- VP)
~-
it
3 believes that its is a spy
(Prop-Att)
\
believes that its is a spy
believes~ its
~
ii spy
The first of these examples exhibits the analysis tree for the S
in question that yields the de dicta reading; its translation will
be xgB{(3x) (E!x & Px & Sx)}. The second example exhibits the
analysis tree for the very same S which yields the de re reading;
its translation will be (3x)(E!x & Px & X3B{SX}).
(R13) Neg: If ifJ E Sand rjJ isn't already negative, then the nega-
tive form of ifJ E S (provided that a certain doesn't occur
in 4».
(8 13) (negative form of <fi)* = ~ifJ*.
EXAMPLES: (Neg) The king of France doesn't exist
(Quantj)
I
The king of France exists
------------
The king of France ~. it4 exists
~ /">.
(Qpantj) The king of France doesn't exist
The king~(Neg~Sn't exist
~ I
it4 exists
~
The translation of the former will be ..-v (7X) (.~!x & Kx)
[AX4E !X4], and that of the latter C,x)(E!x & KX)[Ax4 ~ E!x4].
Clearly, if there exists no king of France, then the former will be
true and the latter false.
(RI4) Pass: If a E TV and if a isn't a passive or a comparative

or is in, then is a en by E TV, where a en is the past
participle form of a.
(814) (is a en by) * = [Axlx2(X2a*xl)].
EXAMPLE:
(Quantj) Gladstone is seen by Holmes
Gladstone
.> <.(Quant4) it3 is seen by Hohnes
HO~ ~ it3 is by it,
it3~ is seeL
(Pass) is seen by
-,by it,
I
sees
Finally, one last rule:

FURTHER DEVELOPMENTS OF THE LANGUAGE f!) 151
(RIS) Non: If a is a simple Adj or simple common noun} then

so is DOD a, unless a already started .with a prefix (some-
times nOD should be spelled an or im.).
(815) (non a)* = [).(N)x ,...., a*x], where uN" is chosen or not
depending on whether er is nuclear or extranuclear.
EXAMPLE:
(DP) thel:& Dozagoldell DlOuntaiD

? it. \ _agoldeD
its (is-AC:ij) is Iloagoldeu
(Non)
I
Jlongoldea
(WDP) lloDgolden lIlountai:n
r I
Whi~OngOldeD
golden
Dlouutaia
III
Applications and Discussion
6
Summary of the Formal Development

and More on Nuclear and Extranuclear
The purpose of this chapter is to summarize the formal de-

velopment of the theory from part 11 and, given the formal
development, to discuss in greater detail the distinction between
nuclear and extranuclear properties.
I. Summary -ofChapter 4
Chapter 4 was devoted to a basic development of the theory that

was sketched in part I. The technique used was to develop a
formal system, called (!), within which to make explicit various
theoretical assumptions about objects. The system (f) has a dis-
tinctive syntax, semantics, and axiomatic development. For-
tunately, most of it bears a strong resemblance to the second-
order predicate calculus with identity ('second-order' here means
that predicates as well as individual terms are quantifiable).
There are three main ways-all discussed in part I-in which (9
differs from the predicate calculus..
First, there is a distinction made between nuclear and ex-
tranuclear properties and relations, and certain assumptions are
made about how they are interrelated. In particular, for every
extranuclear predicate P, there is a nuclear predicate wP which
is the "watered-down" version of P that was discussed in chapter
2; among real objects, P and wP are required to be coextensive.
Second, it is explicitly assumed that variables may range
over all objects, not just those that exist . There is an extranuclear
155
156 APPLICA TIONS AND DISCUSSION
predicate of existence, written El, and there are axioms which

entail that there are objects that don't exist: (3x) E!x.
f"'J
Third, the system incorporates a special treatment of rela-

tions. In chapter 3 it was suggested that where nonexistent
objects are concerned, a relational statement that is superficially
of the form xRy may be ambiguous between x[Ry], which says
that x has the property got by plugging up the second place of
R by y, and [xR]y, which says that y has the property' got by
plugging upthe first place of R by x, These two readings are dis-
tinguished within (f).
I will now give a more detailed sketch of (f), beginning with
its symbolism and a simultaneous commentary on the intended
semantics. The notation used here differs in minor ways from the
symbolism of chapter 4, which was developed with an eye on
logical generality and on having a system easy to prove metathe-
orems about. The present notation should be a bit handier to
use.
The symbolism of {f} contains, first of all, variables which
range over (all) objects: x,y, z, Xl' X2, X3, • • • • It also contains
variables and constants for nuclear properties: pr, pr, ... , and
nuclear relations: Rr, R'l, .... (Although the official sym-
bolism allows for n-place relations, for any n, I will only discuss
two-place relations in this chapter.) The symbolism also contains
variables and constants for extranuclear properties: Pf, Pf,. . . ,
and extranuclear relations: Rf, Rf, .... One of the one-place
extranuclear predicates is singled out for the special role ofstand-
ing for existence; it is written EL The identity sign is also -present,
and it may go between any two terms of the same kind (e.g., we
may have Xl = Y4' R~ = Rf1); such a wff is true if and only if the
two terms stand for the same thing. In addition to identity for...
mulas, atomic formulas may be formed in two other ways. First,
anyone-place predicate, nuclear or extranuclear, may be com-
bined with an object tenn-for example, PjlX'b OrPfx19- A formula
of this form is true if and only if the object has the property in
question. Second, there are relational statements. If R is a two-
place relation term, nuclear or extranuclear, then both x[Ry] and
[xR]y are atomic wffs. The former is true if the object denoted by
x has the property got by plugging up the second place of the-
SUMMART OF THE FORMAL DEVELOPMENT 157
relation denoted by R with the object denoted by y, and the latter

is true if the object denoted by y has the property got by plugging
up the first place of the relation denoted by R with the object
denoted by x. This explanation presupposes that [Ry] and [xR]
themselves behave just like one-place predicates which denote
properties, and that is exactly the way they are treated. So wffs
of this relational form are really just special cases of the Px form
(except that [Ry] has its argument written in front of it instead of
after it). These complex predicates may also flank indentities, so
we may write pN = [RNy ] , for example, . w in
. hich is true just case
each side denotes the same property.
We sometimes want to say that a relation R holds between
x and y regardless of the order of plugging up. This is written as
the simple form xRy; this is defined notation, and its definition is
simply:
xRy =df x[Ry] & [xR]y.
We have one more way of making complex predicates. As
noted above, if Q is any extranuclear predicate, then wQ is a
nuclear predicate of the same number of places as Q. wQ may
appear in formulas in any place that any other nuclear predicate
can.
Last, we can make complex wffs by combining simpler ones
with connectives' and quantifiers. This is just like the predicate
calculus, with the understanding that the range of the quantifiers
that bind object variables includes all objects, notjust the existing
ones.
The symbolism may be made clearer at this point by a few
examples..
(3x)E!x & (3x) ~ E!x
says that some objects exist, and some don't.
(3pN) (x) (E!x :::> PNx)
says that there is a nuclear property which all real objects have.
(PN)(3x) ( -c Eix & PNx)
says that every nuclear property is had by some nonexistent.
(3pE) (x) ( ""'E!x ::> PEx)

says that there is an extranuclear property which all unreal
objects have.
(RN)(3x)(3y)(x[RNy] & ~ [xRN]y)
says that for any nuclear relation there is a pair of objects such
that the first has the property got by plugging up the relation (in
its second place) by the second, but the second does not have the
property got by plugging up the relation (in its first place) by
the first.
says that for any nuclear property there is a coextensive extra-

nuclear property..
Incidentally, all these sample sentences except the first are
theorems of (f). (The first isn't a theorem of (!) because its first
conjunct isn't; (9 does not assume that it is a necessary truth that
any objects exist. But the second conjunct is a theorem.) The
theorems are whatever wffs can be proved from the axioms and
rules, and these will be illustrated now. They come in four
groups..
I. First, there are some "ordinary" rules and axioms for the
predicate calculus with identity.. These include the rule of modus
ponens, universal specification, substitutivity of identicals, etc.
This first set of rules does not contain any abstraction principles;
these are included in the second set.
11. Second, there are axioms which make assumptions about
what properties and relations there are. For extranuclear prop-
erties we have an "ordinary" abstraction schema; it says that,
for any wff if> containing one free variable x, there is an extranu-
clear property which is had by all and only those objects which
satisfy ifJ:
(3pE) (x)(PEx == rfi).
There are similar axioms for extranuclear relations as well as
properties, and ifJ may also contain free variables other than x.
There is not a full-blooded abstraction axiom for nuclear
properties, but a weaker one is derivable as a theorem from the
SUMMARr OF THE FORMAL DEVELOPMENT 159
"watering.. down" axiom. This axiom relates the watered-down

version of an extranuclear property with. the property itself; it
says that they are coextensive on the real objects:
This axiom, together with the abstraction axiom for extranuclear

properties, yields the following weak abstraction principle for
nuclear properties:
(3PN)(X)(E!x ::::> (PNX == ifJ)).
Ill. Third, there are axioms that deal specifically with
objects . These are the axiomatic versions of principles (I) and (2)
of chapter I.. The first is the axiom of the identity of nuclear
indiscernihIes :
(X)(y)«PN)(PNX == PNy) :::> x = y).
This states simply that if x andy have the same nuclear proper-
ties, then they are the same object. The second axiom guarantees
that, for any set of nuclear properties that is expressible in the
language, there is an object that has exactly those nuclear prop-
erties. For any wff ifJ we have: .'-.-
This axiom is called the axiom for objects, and I will refer to it as
axiom OBJ. This is the axiom that lets us prove, for example}
that there is such an object as the gold mountain; just substitute
(PN = GN V pN = MN) for ifJ in OB], getting:
(3x)(PN)(PNx ':= (PN = GN V pN = MN)).
The round square is obtained similarly.
IV. Finally, there are some axioms that govern the behavior
of relations. First, there is an axiom which says that, for extranu-
clear relations, the order of plugging up is always irrelevant:
(x)(y)(RE)(x[REy ] == [xRE]y).
Second, there is an axiom schema for nuclear relations which, in
its two-place form, is equivalent to two claims. The first claim is:
(RN) (x) (y)(E!x & E!y ~ (x[RNy ] == [xRN]y)).
So, for existing objects, the order of plugging up is always ir.. .

relevant. But this is not always true when one or more of the
objects is unreal; in fact, for mixed cases we have the second
claim:
(RN) (x)(y) (E!x & ~E!y::> (~x[RNy] & ~[yRN]x)).
Consequently, a nuclear relation may hold between existents

and existents, and between nonexistents and nonexistents, but
never between existents and nonexistents (recall that a relation
is said simply to "hold" between x andy just in case it holds no
matter which end is plugged up first).
In its precise formulation the system of axioms can be shown
to be consistent, and when the semantics is formulated set-
theoretically, a completeness theorem is provable. The reader
is referred to chapter 4 for details and discussion..
2. Comparison with Orthodox Systems
The assumption about objects made in axiom OB] depends

essentially on the nuclear/extranuclear distinction.. For if this
distinction is eliminated from the syntax (by erasing all super':'
scripts), the system becomes inconsistent. (The abstraction axiom
for extranuclear properties already allows us to prove that there
is an extranuclear property--e.g.. , nonself-identity-e-which no
object has . With the nuclear/extranuclear distinction eliminated,
axiom OBJ would tell us that some object has this property.)
On the other hand, the nuclear/extranuclear distinction
together with axiom OB] are what gives the system its import..
For if'these are both dropped, and if we add the assumption that
at least one object exists, then the system of axioms becomes
equivalent to a version of the second-order predicate calculus
(to which has been added a single extension-preserving predicate
operator w). The order of plugging up relations becomes prov..
ably irrelevant in all cases, and it becomes a theorem that every-
thing exists: f- (x)E!x.
As it stands, the system (!) says. a great deal (consistently)
about objects that transcends the orthodox view, but it also
"contains" the "neutral" part of the orthodox metaphysics

within it. Suppose we say that a neutral formula is any closed wff
 is neutral just in case every existen-
tial quantifier (3Xk) of <p occurs in ifJ only in contexts of the form
(3Xk) (E!Xk & <jJ), and every universal quantifier (Xk) of 4J occurs
in <p only in contexts of the form (xk)(E!xk :J ifJ). Intuitively,
neutral wffs are those that explicitly refuse to discuss nonexistents.
Because their quantifiers are restricted to existents, the
order of plugging up a relation within a neutral wff is provably
irrelevant, and so we may ignore this idiosyncrasy and compare
neutral wffs directly to the wffs of a version of a two-sorted
second-order predicate calculus (to which a single extension...
preserving predicate operator w has been added). This predicate
calculus will be peculiar in that it will have two styles of predi-
cates, (the ones with superscript E and the ones with superscript
N), so it will be a two-sorted theory. But within neutral wffs
there will be no, difference in the logical behavior of these two
sorts of predicates, so the peculiarity has no special import. That
is, when all talk is explicitly limited to what exists, the nuclear/
extranuclear distinction has no logical significance; this is true
both within the ordinary predicate calculus and within (9. It
now turns out that the neutral theorems of (!) correspond exactly
to the neutral theorems of this predicate calculus.. (For this fact
to obtain, it makes no difference whether or not the predicate
calculus in question contains the special axiom (x)E!x.. )
So any neutral truth of the orthodox conceptual scheme
can be preserved within (f) just by making explicit that it is ex-
istents that are being discussed.
Other comparisons with orthodox systems can be made;
the reader is referred here to chapter 4, sections 5-7.
3· Summary of Chapter 5
III chapter 5 the system (!) was expanded so as to include modal i-

ties, abstracts, verbs of propositional attitude, names, and definite
descriptions. These additions will be discussed briefly here. An
162 APPLICATIONS AND DISCUSSION
appendix to chapter 5 contains a formal procedure for translat-

ing a certain fragment of English into this expanded version of
system {f}; that procedure will not be discussed here.
The modal operators 0 ('necessarily') and 0 ('possibly')
can be added to {f} and given a semantics similar to a possible
worlds treatment, The result is a system of quantified 85 in which
the Barcan formulas hold; that is, 0 commutes with the uni-
versal quantifier and 0 with the existential quantifier. The
standard objection to the Barcan formulas are that they lead from
a certain truth, 'necessarily, everything exists', to a certain
falsehood, 'everything necessarily exists', or from 0 (x)E!x to
(x)DE!x~ But this objection depends on the orthodox assumption
that (necessarily) everything exists, an assumption that is re...
jected within o. In the system (9, both O(x)E!x and (x)DE!x
are provably false.
With the addition of modal operators, it is natural to place
necessity signs in front of all the axioms discussed previously.
I t is also natural to adopt certain de re variants of some of the
axioms, and this is also done. For example, we now have not
only:
O{3pE)(x)(PEx == </J),
but also:
For example, not only is it true in every world that there is a

property which (in that world) applies to all and only purple
unicorns (of that world), but there is a property which does this
uniformly in all worlds.
The addition of abstracts is an enormous convenience.
Given a formula rP, we write [AxsbJ for "the property of being
<jJ," and we add the axiom of abstract elimination (also called
the axiom of abstract introduction):
D(Y)([Axq)(X)]Y == <jJ(y».
For example, if pN represents being purple, and UN being a
unicorn, then [AX(PNx & UNx)] will represent the property ·of
being a purple unicorn. This is a property that, in every world,
is possessed by all and only those objects that are both purple
and a unicorn; that follows from the axiom of abstract elimina-
tion:
O(y) ((Ax(PNx & UNx)]y == (PNy & UNy)).
Such abstracts always denote extranuclear properties, and we
are often interested instead in nuclear ones. So the symbol AN
is introduced for the nuclear abstraction operator, and it is de-
fined as:
This, together with previous axioms, yields a weak axiom of

elimination / introduction of nuclear abstracts:
O(y)(Ety :;:) ([ANXqS(X)]Y == tPy))·
Abstracts are convenient when discussing t'complex" nuclear
properties. For example, the (nuclear) negation of a nuclear
property P may be defined as [ANX ,...., Px], the nuclear property
of not being P. This is not a property possessed by all and only
those objects which lack P; as was pointed out in chapter 1,
there is no such nuclear property (though there is an extranu..
clear one: [AX ,...., Px]). But it is a property that is possessed by all
and only those real objects that lack P.
Verbs of propositional attitude, such as 'believes', are in-
corporated into (J) as symbols which precede wffs to form one...
place nuclear predicates. For example, 'believes' (or 'believes
that') combines with 'some objects exist' to form 'believes that
some objects exist'; and this combines with subjects to form
sentences just as a simple predicate like 'runs' does. Within f!)
these verbs are represented by the symbols B l , B z, B3 , • . • ;
they combine with wffs which, for reasons of punctuation, are
enclosed in braces. So, for example, the de dicta reading of )
believes that someone is a spy' would be written:
yB{ (3x) (Px & Sx)},
where P represents 'is a person' and S represents 'is a spy'. The
de re reading of that sentence (the reading which might be par-
aphrased :, believes of someone that they are a spy') is written:
(3x)(Px &yB{Sx}).
Except for the law of substitutivity of identity, no logical
principles at all are adopted for the behavior of items within the
braces. For example, although '» is a villager' and 'x is a vil-
lager who does not shave all and only those villagers who do not
shave themselves' can be proved to be necessarily equivalent, it
is not assumed that these phrases are automatically interchange-
able within propositional attitude contexts. That is, it is ·not a
theorem that y believes x is a villager if and only ify believes x
is a villager who does not shave all and only those villagers who
do not shave themselves'.
Thelaws of identity do hold even where propositional at-
titude verbs are concerned, so that from P = Q and xB{Py} we
can infer xB{Qy}. Although it's hard to find interesting versions
of that inference, a contrapositive version of it is often useful:
fromxB{Py} and ~xB{Qy} we infer P :F Q. For example, from
'x believes y has a heart' and ~ x doesn't believe y has a kidney',
one may infer that having a heart ::1= having a kidney.
Propositional attitude constructions are often thought to
provide contexts within which singular terms may violate the
law of substitutivity of identicals. For example, from 'Pegasus is
the winged horse of Greek mythology' and '» believes that the
winged horse of Greek mythology is winged', we cannot (on the
de dicta reading, anyway) infer that '» believes that Pegasus is
winged'. In fact, the treatment of singular terms within (9 avoids
this problem. That \ViII be explained below, after a sketch of how
singular terms are treated.
First, definite descriptions. They are written in the normal
way; that is, if you have a formula rP, then you can put an 7x
in front of it, ('x)ljJ, and you read it 'the thing such that if/, or
words to this effect. For example, you read:
(1X)(Wx & Hx)
as 'the thing such that it's winged and it's a horse', or just 'the
winged horse'. And the semantical account is that (7x)rfi refers to
the unique object that satisfies t/J, if there is one, and otherwise
('x)ifJ just doesn't refer at all.
SUMMARY OF THE FORMAL DEVELOPMENT 165
Definite descriptions are used to make sentences in a manner

that is just slightly unorthodox: they go ui front of the predicates
they combine with (there .is no logical significance to this). In
other words, any definite description may be placed in front of
anyone-place predicate to form a wff. 'So if we want to write,
say, 'The man in the doorway is clever', we write something like
this:
(JX) (MNx & ]Nx)CN.
I say "something like this," for although what has been written
is perfectly well formed, it probably doesn't accurately symbolize
most normal uses of that English sentence. For if you recall how
varied our ontology is, you will realize that there are lots of men
in the doorway (this was pointed out in chapter I), but most of
them are just not relevant to what I am saying. Generally, I'll
be referring to (or at least attempting to refer to) a real man. And
so the best way to symbolize most natural uses of the sentence is:
That is, 'The existing man in the doorway is clever'; I don't say
'existing', but the context makes it clear that this is what I mean.
When are such sentences true or false? The account pro-
posed in chapter 5 was that a sentence of the form (?x)if>P is true
when (7x)rjJ refers .to an object which has P, and false in all other
cases. (The reader is referred to chapter 5, section 4, for a dis-
cussion of the assignment of falsehood in cases in which the def-
inite description is nonreferentiaI.)
Names are like definite descriptions except that they have no
logically significant parts.. They are symbolized in (f) by the sym-
. bols A, B, C, . . . ; they go in sentences in the same places as
definite descriptions (i.e., preceding one-place predicates); they
either refer or not; and a sentence of the form AP is true when A
refers to an object which has P and false otherwise.
The major remaining question concerning singular terms
has to do with how to symbolize English sentences which use
English singular terms by means of sentences of {f} that use the
termsjust discussed. This is dealt with in the appendix, and I will
give only two illustrations here. The first has to 'do with the de
re/de dicta ambiguity of a sentence such as 'Agatha believes the

man in the doorway is a spy'. Its de dicto reading is represented
as:
AB{(1x)(Mx & Ix & E!x) [JySy]},
and its de re reading is represented as:
(1x)(Mx & Ix & E!x) [Ay(AB{Sy})].
The former attributes to Agatha the belief that the man in the
doorway is a spy, whereas the latter attributes to the man in the
doorway the property of being believed by Agatha to be a spy.
The identity sentence 'Pegasus is the winged horse of Greek
mythology' gets symbolized:
P[).y('x)(Wx & Hx & Mx)[k(y = z)])].
This does not have the form, in (!), of an "atomic" identity sen-
tence, and it is this which makes the laws of identity innocuous
where belief contexts are concerned.. The available laws do, how-
ever, let you go from the above identity plus the fact that the
winged horse of Greek mythology is winged:
(7X)(Wx & Hx & Mx) W,
to the conclusion that Pegasus is winged:
pw.
4- More on Nuclear and Extranuclear
Within f!) there is a syntactic difference between nuclear and

extranuclear predicates-they look different. But if we begin with
an English predicate, how do we tell if it is nuclear or extranu-
clear? In this section I will suggest that although it is sometimes
very difficult to tell, this will typically be so only when it is not
very important,
Although extranuclear properties or predicates behave quite
differently as a class from nuclear ones, sometimes individual
members of each class mimic one another quite closely. Let
Part of this section appeared in different form in Parsons [N&EQM&L].
SUMMART OF THE FORMAL DEVELOPMEJVT 167
me say that two predicates P and Q are essentially equivalent if

D(x)(Px := Qx) is true, and, analogously, that two properties P
and Q. are essentially equivalent if necessarily they are true of the
same objects. Then it may sometimes happen that a given Eng-
lish predicate has two equally good and essentially equivalent
representations within (!). If this is so, and if one of the represen-
tations within (p is nuclear and the other is extranuclear, then it
may be impossible to classify the English predicate as either.
Of course, it doesn't automatically happen that essentially
equivalent predicates are equally good representations of a
given English predicate, for essentially equivalent predicates
may not be equivalent within propositionaI attitude contexts (i.e.,
they may not be interchangeable within such contexts without
alteration of truth value). But the truth values ofsuch contexts
are notoriously hard to evaluate, so this may give us very little
help in deciding between two proposed symbolizations. On the
other hand, if no help is forthcoming even here, then we seem to
have a case where it is not important how to classify the original
English predicate. It is not important in the sense that. there are
two equally good ways to classify it, equally good in the sense
that neither choice makes a discernible difference in the truth
values of the sentences that result.
Does this happen often? Well, yes. For according to the
abstraction axiom for extranuclear properties, every nuclear
property is essentially equivalent to some extranuclear one. (The
following is derivable from the abstraction axiom: (PN) (3pE) 0
(x)(PEx = PNx), where pN x occupies the if> position..) So unless
propositional attitude contexts come to the rescue, the problem
will occur in the case of every nuclear predicate. My policy so far
has been to classify a predicate as nuclear whenever it can con-
sistently be classified that way. (If a property is essentially equiv-
alent to some nuclear property, then it is essentially equivalent
to exactly one nuclear propertyjl so if a predicate can be con-
sistently classified as nuclear, there is always a unique choice of
r , For suppose that P is essentially equivalent to both Q,N and RN, that
is, we have O(x)(Px == Q.Nx) and also O(x)(Px == RNx ). It then follows that
(x) (ft'lx == RNx), and so Q.N = RN by the extensionality of nuclear properties
(see discussion of axiom OB] in chapter 4, section 2).
which nuclear property it stands for.) But other options are

available. For example, having used nuclear properties in for-
mulating the metaphysical theory, we could then totally ignore
them when we discuss language, by holding that all English
predicates stand for extranuclear properties. Or we might hold
that in the cases in which a nuclear property is indistinguishable
from an extranuclear property, that is so because they are iden-
tical. This would make the nuclear properties a subset of the
extranuclear ones. In presenting the theory, I have talked as if
nuclear and extranuclear properties were disjoint, but in fact
there is nothing in the formal development of the theory which
requires this; indeed this can't even be formulated within (f) as
it stands.
All this suggests that we focus our attention on a slightly
different distinction. Although all nuclear properties are es-
sentially equivalent to extranuclear ones, the reverse is not true;
some extranuclear properties are not essentially equivalent to any
nuclear ones. Let me say that P is essentially extranuclear if P is not
essentially equivalent to any nuclear property. Then we can tell if
a property is essentially extranuclear or not, for essentially extra-
nuclear properties do not behave like nuclear ones.
How does their behavior differ? There is probably no good
general answer to that. But a test that works for many cases (for
all the cases discussed in this book, I think) is the one from chap...
ter I : if P satisfies the following, then P is extranuclear: for some
set X of nuclear properties not containing P, it is possible that
every object that has every property in X has P. (A supplemen-
tary test is to replace 'has' by 'lacks' before (P'.) This test presup-
poses that we already know that some properties are nuclear; in
practice, a few paradigm cases of nuclear properties suffice.
The only cases that bother me somewhat are comparatives.
Intuitively, they seem to be just like ordinary nuclear relations,
at least at first sight. But they turn out instead to be extranuclear.
Or at least they seem to. Here is the evidence. If 'taller than'
were a nuclear predicate, then I could not be taller than Hercule
Poirot. For Hercule Poirot does not exist, and so I would not have
the property being-taller-than-Poirot, And since I do not occur
in any of the Agatha Christie novels, Poirot would not have the
SUMMARTOF THE FORMAL DEVELOPMENT 169
property being-such-that-Parsons-is-taIIer-than-him. So on both

counts, Poirot and I would not be related by the taller-than re-
lation-if'taller than' were nuclear, anyway. But Poirot is very
short, and I am at least of average tallness, so doesn't this estab-
lish that I am taller than him?2
Maybe not. It seems to me that I can truly be compared with
certain fictional characters. I'm taller than Poirot, less clever than
Holmes, more agile than Nero Wolfe, and so on. But perhaps
these are only loose ways of speaking, or appearances to be dis-
pelled. I'm not sure. But if not, then I do stand in these relations,
and then (unless the theory in question is quite wrong about rela-
tions) these comparatives must be extranuclear.
Why shouldn't they be extranuclear? I have only an intuition
to go on here. But maybe the intuition can be explained even if
comparatives are basically extranuclear. Let us first look more
closely at comparatives.
There is a lot of evidence that comparatives are much more
complex than they might seem at first sight. The evidence is
both syntactic and semantic.. Syntactically, there is a tendency
to use the nominative case for the "object" of the relation: 'she
is taller than I' rather than 'she is taller than me' (the former is
even held to be the correct usage). This suggests that such con-
structions are short for the more complex "she is taller than I
am'. And the latter construction suggests others; we can say
things like 'the box is longer than the table is wide'. If 'longer
than' is a simple relational expression it's hard to see how to make
sense out of such locutions. Semantically, there are examples
such as 'I thought your yacht was longer than it is' (see Russell
[OD]), which tends to provoke paraphrases like 'Your yacht is a
certain length, and I thought its length exceeded that'; but this
is crazy if 'longer than' is a simple relational expression. For
reasons like these,. both philosophers and linguists have proposed
that comparatives are the surface forms of fairly complex "deep-
structures." If this is 50, one would expect that the attempt to
classify comparatives into one of two classes of relational expres-
sions would run into trouble.
2. Comparisons with fictional objects are discussed in Howell [FO :
HAHA].
170 APPLICATIONS AJYD DISCUSSION
One sample kind of account of comparatives goes something

like this: '» is taller than y) means 'x has a certain quantity u of
tallness, andy has a certain quantity v of tallness, and u exceeds
u' (see Prior .{PP&F] VIII. I I). (It's clear, then, how to handle
the box/table case, at least in an ad hoc way.) Suppose that some...
thing like this is the correct account, and that comparative
sentences can be represented within (!) as some such complex
formulas. Then there will be some derivative "comparative"
relations: there will be the extranuclear relation denoted by
[AXY(X is taller than y)], and its watered-down nuclear version,
denoted by [A,Nxy(x is taller than y)] . Perhaps these are the choices
that come to mind when we discuss comparative relations.. Per..
haps I can stand in the former relation to Hercule Poirot (that
would depend on the details of the analysis), and perhaps it is
the latter that gives rise to the intuition that 'taller than' is nu..
.clear.
5. Modifying the Distinction
On various occasions people have proposed to me modifications

of the nuclear/extranuclear distinction. These proposed modifica-
tions have typically had one of three motivations. First, there is
a general discomfort with the distinction, and a desire to do with-
out it if at all possible (while still having a theory of nonexistent
objects). Second, the theory of fictional objects proposed here
discriminates between nuclear and extranuclear properties in a
way that some people find inappro.priate: Sherlock Holmes, for
example, is the object correlated with the set of nuclear properties
that he has according to the novels, Some people would rather
see objects correlated with arbitrary sets of properties, so that
Sherlock Holmes could then be the object correlated with the set
of nuclear and extranuclear properties that he has according to
the novels. Third, the presence of extranuclear predicates in
certain definite descriptions prevents them from referring to
objects which satisfy the description; for example, 'the gold
mountain that exists' cannot refer to a gold mountain that exists,
because there are none. But perhaps if there were objects cor-
related with arbitrary sets of properties, there would be an ap-
propriate referent for the definite description. And then perhaps

we could accomplish the goal, attributed by Russell to Meinong,
of having all definite descriptions refer.
I suspect that the first and third of these goals. cannot be
accomplished at all, and that the second cannot be accomplished
in a satisfying manner.
First, we need to get clear about what is being proposed. We
are to suppose a one-one correlation between objects and ar-
bitrary sets of properties. But what is the correlation? It can no
longer be that x is correlated 'with X if and only if X is the set of
all and only those properties which x has. For this would rein-
troduce Russell's objections. For example, correlated with {golden-
ness, mountainhood, existence} would be an object which is golden
and a mountain and exists.. But that can't be; for no gold moun..
tains exist. The law of noncontradiction would also come into
play. For if there is any property which has a genuine contrary,
the theory would be inconsistent. Suppose that P has a contrary,
Q.-that is, no object has both P and Q.. But the object correlated
with {Pness, Qness} has both P and Q., which contradicts the
assumption that P and Q. are contraries.
The obvious thing to do here is to make a distinction be-
tween a property's being in the set correlated with an object, and
the object's having the property. In the former case, let us say
that the object includes the property. Presumably, real objects
have exactly the properties they include, but for others there will
be a divergence. For example, let g be the object correlated with
{goldenness, mountainhood, existence}. Then g includes exist-
ence without having it, and has incompleteness without including
it. When applied to language, this theory has the consequence
that predication is ambiguous: Fa can mean either that a has
Fness or that a includes Fness. A slightly more complex theory that
resembles this one has been proposed by Castaneda ([TSW]), in
which predication is at least three-ways ambiguous,
To what extent can any such theory accomplish the goals
mentioned above? First, what about the nuclear/extranuclear
distinction? This can be dealt with in various ways. At the level
of "data,' I would like to think that the distinction is present in
the following form: excluding impossible and necessary proper-
ties, the essentially extranuclear properties are those for which a

distinction between having and including isforced on the theory.
For example, we are forced to distinguish having from including
for existence, incompleteness, etc., but we are not similarly forced
to make this distinction for goldenness. At the level of theory, the
distinction may appear in different forms. For example, in Cas-
taneda's theory, one is to symbolize normal uses of statements
which use nuclear predicates differently from parallel statements
that use extranuclear predicates. And in Routley [EM]] one
may use the rule of abstract elimination on abstracts containing
nuclear predicates, but not, in general, on abstracts containing
extranuclear predicates. Other theories are of course possible.
My suspicion is that any natural theory of this sort will preserve
the distinction in some guise.
Second, what about fictional objects? Well, we can easily
treat nuclear and extranuclear properties uniformly with respect
to what is included in fictional objects, but the treatment will
have to be nonunifonn with respect to what properties a fictional
object has. For extranuclear properties, there can be no neat
correlation between having and including; Sherlock Holmes,
for example, will include existence without having it, and have
incompleteness without including it. But for nuclear properties,
no such divergence seems appropriate. Perhaps fictional objects
have all the nuclear properties that they include, as on the pres..
ent theory, or perhaps they have none.as on the variant of van
Inwagen's theory discussed in chapter 7. But there seem to be
no grounds for supporting a third alternative.
Third, the attempt to make all definite descriptions refer to
objects which satisfy the description cannot work if any means
at all is available to eliminate ambiguity, As we have seen, a
predication like 'x is golden' may be ambiguous between "x has
goldenness' and 'x includes goldenness', but if the distinction
can be made in language, then definite descriptions can be
formulated which make Russellian objections good. For example,
'the thing which has goldenness and has mountainhood and has
existence' cannot refer to an object which has goldenness, moun-
tainhood, and existence.
One other proposed modification is worth mentioning.. The
idea here is to take the theory I have developed as the initial

stage ofa kind of hierarchy. Just as I have "expanded" the stock
of objects beyond the existing ones by invoking a correlation be-
tween objects and sets ofnuclear properties, maybe there ought to
be a second stage in which the extranuclear properties of the
first stage are treated as the nuclear ones were at the first stage.
So at the second stage, there will be a further expansion of the
stock of objects, guided by a new correlation between objects
and sets of properties regarded as extranuclear in stage one.
This will result in a new set of properties that behave "extra-
nuclearly" from the point of view of stage two. These would
then be treated as nuclear in a new expansion of objects at stage
three, and so on.
There are several ways to develop such a theory. As far as
I can see, none of them will accomplish the goals described at
the beginning of this section, for reasons like those given pre-
viously. The original nuclear/extranuclear distinction is not
dispensed with in such a theory; it is still with us, multiplied
infinitely. As for fictional objects, it is not clear how best to
formulate a theory of them which takes into account the various
stages of objects and properties, but there will still be some sense
in which fictional objects do not exist, and presumably it will
be in this very sense that stories will say (implicitly or explicitly)
that they do exist. So it is doubtful that all nuclear and extranu-
clear predications can be treated on a par where fictional objects
are concerned. Predication, of course, will be ambiguous,. as
in the simpler sort of theory discussed above; a property which
is treated extranuclearly at stage n will be treated nuclearly
at stage n + I, and these two treatments will not in general
be equivalent.f So, as above, if there is any means within the
3. For example, consider the property of being complete at stage n, that

is, the property C which applies extranuclearly to all and only those objects
which are formed by stage n and which have, for each property p which is
treated nuclearly at or before stage n, either p or its nuclear negation. Prop-
erty C is not treated nuclearly until stage n + I, so no object formed by stage
n will have C nuclearly. Thus no object at all will have C both nuclearly and
extranuclearly, though some will have it nuclearIy and some extranuclearly
(and some in neither manner).
language to resolve ambiguity, it will be possible to formulate

unsatisfiable descriptions, and the third goal will not be met
either.
Of course, such a theory may be worth developing for rea-
sons other than these.
7
Fictional Objects, Dream Objects, and Others
In chapter 3. I gave a sketch of a theory of fictional objects,

Briefly, it is this: We distinguish between immigrant objects and
objects native to a story. Then if the tP of story s is native to s,
the has in (according to) s. Sections 1-3 of the pres-
ent chapter deal with this theory in somewhat more detail, sec-
tions 4-7 discuss possible objections to the theory, and section 8
discusses alternative theories. The remaining sections briefly
discuss objects of legend and myth, dream objects, objects in
pictures, and others.
I. ~cln the Story ~ .. "
The theory of fictional objects given here depends on our making

sense of locutions of the form 'In the story, <j>' as well as their de re
variants, '(3x) In the story, Fx'.. In this section I will attempt a
general outline of how these judgments originate. I'll limit myself
here explicitly to realistic narrative fiction.
Whether or not something is true in a story ought to accord
with what a normal attentive reader understands to be true in the
story. We can model the origins of such understanding as follows:
as the reader reads, an account is constructed by extrapolation
from the sentences being read, The account is modified and ex-
panded during the reading, and the final result may be called the
maximal account. Then what is true in the story is just whatever the
maximal account explicitly says, and nothing else. The trick is to
175
176 APPLICA TIONS AND DISCUSSIOJY
- get an understanding of the extrapolation process that produces

the maximal account.
Basically, the idea is that as the reader reads the story
from start to finish a partial account is gradually developed as
follows:
(a) Typically, as a new sentence is read, that sentence is
added to the account.
(b ) Typically, lots of other sentences are simultaneously
added as well.
(c) Often, sentences are removed from the account.
Each sort of modification requires so~e comment:
(a) Typically, we add the sentences we read to the account;
this amounts to accepting them as descriptive of what is happen-
ing in the story. But this is not always done. Occasionally, an
author departs from story telling to make general observations
about human affairs, cosmic forces, etc.; and while these may
influence (b) and (c),'we do not automatically add them to the
"plot" of the story. Sometimes the story is told through a narrator
whom we take to be a character of the story-sometimes an
ordinary character, as is Dr. Watson in the Conan Doyle novels,
and sometimes a strangely omniscient observer of the plot.
In such a case we may sometimes judge a sentence (which is
not in quotes) to be a lie or error on the part of the narrator. And
we may then add to the account 'the .narrator claims that </>'
rather than '4/, and we may even add '~ifJ' as well. And there
are also ~ases which we judge to be mere misprints. In short, the
presence of a sentence in a novel has a primafacie claim to author-
ity regarding the plot, but this authority is subject to the evidence
that may count against any authority.
(b) Certainly, the vast majority of the sentences in a maxi-
mal account of a story will be sentences that do not occur in print,
but ones that have been inferred from those that do. Any reader
who failed to extrapolate far beyond the printed word would be
quite unable to understand the story.. This is typically not a con-
scious process; we just instinctively assume all sorts of things on
the basis of the given information, just as we do in perception. We
read "Commander Roderick Blaine looked frantically around
FICTIONAL OBJECTS, DREAM OBJECTS, AND OTHERS 177
the bridge ...... ,"1 and we immediately assume, without con...

scious reasoning, that Blaine is named 'Blaine', that he is a com-
mander, that he is male, that his eyes are open, that he is located
in a space more than one cubic inch in volume, that he is up...
set, . . . . Any of these inferences may be revised in the light of
further information (e.g.. , if we read that his eyelids had been
replaced with some transparent material, and were now fastened
shut), but we make them anyway, and if they remain unchal-
lenged, they remain stored-in the account.
There is now a growing body of literature concerning what
principles are operative here.s The first principle that comes to
mind, and perhaps the least satisfactory, is that we extrapolate
to exactly those sentences that follow logically from those that· are
explicitly stated in the text. In addition to the problems dis-
cussed under (a) above, this proposal is incorrect because it (i)
adds unwanted sentences, and (ii) fails to add crucial ones. I'll
comment briefly on what is wrong, because the failings are
informative.
First, the proposal adds all sorts of irrelevant material to
the story. In particular, it adds all logically true sentences with
any subject matter whatever, including all of those (true) theo-
rems that mathematicians have tried unsuccessfully for centuries
to prove. They aren't part of our understanding of the story.. Still,
this isn't too serious an objection; we could possibly live with it .
But logical entailment lets in other consequences that are too
remote to be part of a normal reader's understanding. An ex-
ample where this is vivid would be a story that was inconsistent,
but where the inconsistency would never be noticed. For exam-
ple, imagine a 1,000 page novel which contains the following
claims widely scattered throughout its pages:
( I ) None of the Hatfields ever gave any McCoy anything.
(2) Samantha doubles as a baker and a mortician.
(3) There are at least seven Hatfields.
I. A quote, chosen somewhat at random, from Niven and Pournelle
[MGE], p. I.
2. See Blocker [TAFE], Heintz [RIF] , Howel1 [FO :HAHA], Lewis
[TIF], Martin and Schotch [MFN], Parsons [RW], Pollard fF&S), Routley
[SSFD], vVolterstorff [WWA], and Woods [LF].
(4) Herman never gave Molly anything.

(5) No mortician ever gave anybody anything..
(6) There are nine bakers,
(7) One of the bakers once gave Molly a gold watch which
had a greyish-colored face.
(8) Herman isn't a Hatfield.
(9) Herman is a baker (but not a mortician), as are all of
the Hatfields. .
(10) Molly is a McCoy.
(I I) Samantha isn't a Hatfield,
These sentences are inconsistent, and we can exploit this fact to
infer from them things like:
(12) Molly isn't a McCoy.
(13) Molly didn't get a gold watch from a baker..
and so on. But these need not be part of our understanding of the
story, even though they follow logically from things that are.t
We might try cutting the logical inferences down to those
that are obvious. But this won't meet the opposite problem: our
extrapolations take us far beyond what is logically implied by the
text (assuming it to be consistent), even when logical implication
is broadly construed. We bring a great deal of understanding of
the world with us to the text, and we utilize this understanding to
expand on what is explicitly stated. But what is the guiding prin-
ciple here? Is it that we add known empirical generalizations to
what is stated in the text, and make obvious inferences from the
combination? Certainly riot, for often a text contradicts known
fact. If we read '14,000 unicorns invaded the town' we don't add
the known fact that no unicorns exist, and start deriving the then
all too obvious contradictions. Nor do we salvage consistency by
3. I think that this will create difficulties for Wolterstorff [WWA], in

that these undesired sentences will be members of lots of "strands." Some
versions of relevance logic will also have difficulties here, if applied to this
phenomenon (e.g., system EQ of Anderson [CTE&EQ]). Relevance logic
typically restricts itself to what is relevant, but the present problem has to do
with avoiding conclusions that are remote, not ones that are logically irrelevant..
More restricted systems might be appropriate here, but they will encounter
the next problem discussed in the text.
FICTIONAL OBJECTS, DREAM OBJECTS, AND 0 THERS 179
concluding that the unicorn invaders were all unreal, for the task
is to decide what is true in the story, and in the story the uni-
corns are real. Maybe we add a known fact when (and only
when) it's consistent with the text? No. If we read that Agatha
bought a horse with wings, we don't immediately conclude that
she bought a horse that can't fly, just because we know that no
real horse can fly, and the story doesn't literally contradict that.
Maybe the story implicitly contradicts it? But the goal of the
enterprise is to say what the story implicitly says; there's no point
to giving principles for doing this which use what the principles
are designed to produce.
We can't even keep this problem "pure" by focusing on the
text alone, in isolation from the literary tradition in which it is
produced. Consider a story whose hero is named 'Ed King'; he is
a Mafia big shot with some unspecified scandal in his past who
has married a wealthy and powerful woman much older than
himself. Late in the novel he discovers something (which is not
revealed to the reader) which causes him to abandon his position
and become an alcoholic, and the novel ends with him nearly
blind and dying of liver damage.
Now if this were written in a certain literary tradition, facts
extraneous to the text would influence its interpretation. The
name 'Ed King' would be seen as a play on 'Oedipus Rex', and
other similarities to the Oedipus myth would clinch it. The
reader would know that King had unknowingly murdered his
own father and married his mother, and that this is what he had
discovered and what had led to his downfall. On the other hand,
the same text could have been produced in a different situation,
one in which we would judge the parallels' with the Oedipus
myth to be coincidental and irrelevant to the proper interpreta-
tion of the plot.
(c) The fact that material already stored in the account can
be removed on the basis of new information (as illustrated above)
is one of the biggest impediments to the formulation of principles
of inference to shed light on textual interpretation. In fact, our
actual inferences seem to exploit some principle of total evidence:
whether a sentence is to be included or not depends on what else
is included. Judging what is true in a story is very much like
judging what is true according to a given person, based on what

they say, together with what we know about their situation and
about the world. We can gain insight into the ,process by seeing
how various formal and mechanical principles fail to do justice
to the process; I think this is their major source of interest here.
There is another problem here of a different sort. I have been
speaking as if accounts consisted of sentences. But this must be
wrong. For one thing, the process of amassing an account is a
mental one, and we don't do this by storing up English sentences
in the mind. It isn't so much that we don't think in English
(though perhaps something could be made of this), but rather
that English sentences aren't the right sorts of information
bearers. For they are mostly ambiguous, both lexically and struc-
turally, and in the theory I am discussing the account is not ambig-
uous. Perhaps ambiguity (and, e.g., cross-reference of pronouns)
could be resolved by means of elaborate paraphrase, but such
paraphrase typically adds content in ways that would go beyond
the extrapolations justified by the text .
Maybe, then, we need an account to consist of propositions?
This is better, but it, too, has its difficulties. For example, we read
a sentence containing the name 'Sherlock Holmes', and we want
to store the proposition it expresses. What does the name contrib-
ute to the proposition? There are two main theories here: the
name contributes its reference, or it contributes its sense. But how
can we mentally store up a proposition containing Holmes, when
we havejust started the novel and don't yet know who he is? And
how can we know the sense of the name if, again, we've just
encountered it for the first time? The problem here isn't that
Holmes doesn't exist, for the same problems arise in the case of
existing objects as well. The problem has more to do with the
mental storage .of information. The foundations of the theory are
obscure here; I don't know how to make them clearer, so I'll
just go on in the hope that the.y are in fact secure enough.
I am also going to suppose that as we build our account, we
amass a stock of characters. Normally, the use of a name or definite
description is sufficient to establish a character. But this is not
necessary; characters are often introduced by means of indefinite
description. A fairly sure sign that this has been done is if there is
FICTIONAL OBJEGTS:1 DREAM OBJECTS, AND 0 THERS 181
a later back-reference to the description by means of a singular

pronoun. So if we read 'A man entered the room. He ~ . .', we
very likely have a character.
The stock of characters determines which de re judgments
may be made with respect to the operator 'in the story' ~ For ex-
ample, Sherlock Holmes is a character of the Conan Doyle nov..
els, and we can make the de re judgment 'Sherlock Holmes is such
that in the story he lived at 22IB Baker Street'. But even if we
know that a description is uniquely satisfied in a story, that does
not give rise to derejudgments if the description does not pick out
a character of the story. Suppose we read that on a certain
morning Holmes had one soft..boiIed egg for breakfast. Then we
may extrapolate that exactly one chicken laid the egg that
Holmes ate on that morning. But the chicken is not a character
of the story. And so, although we might include in the account
the de dicto:
In the story there was a unique chicken that laid the egg
that Holmes ate.
we would not include:
The chicken in the story that laid the egg that Holmes ate is
such that in the story it laid an egg.
(What about the egg? I would make the egg a character of the
story. Characters need not be limited to persons.s)
4~ The theory I have been sketching can be made a little more precise
as follows: we take as our input an ideal normal reader plus a text (and perhaps
a context). The output is a maximal account plus a stock of character terms.
Then, if S is a closed sentence, the sentence:
In the story, S.
is true if and only if S is a member of the maximal account. And if ~ is an
open sentence containing one free variable x, the formula:
In the story, ifJ.
is satisfied by an object 0 jf and only if there is some character term A such
that the sentence A[Axsb] is a member of the maximal account, and 0 is prop-
erly related to A. The object 0 will be properly related to A just in case for
any nuclear property p, 0 has p if and only if there is a predicate P such that
P stands for p and AP is a member of the maximal account.
A maximal account resembles somewhat a description of a

possible world, but with at least two differences: it is typically
highly incomplete, and it is occasionally impossible. The incom-
pleteness is obvious. There is no clue in the novels about whether
or not Holmes has a mole on his back, and no reason to extrap-
olate to either side of the issue. So neither 'Holmes has a mole on
his back) nor its negation will appear in the maximal account. It
might be thought, then, that a maximal account will describe a
class of possible worlds: those consistent with the account. But too
often such a class will be empty, because the account is impos-
sible. This can happen in various ways. One way is this: we read
'As a youth, Quagmire amazed the mathematicians of the day by
showing how to trisect the angle with ruler and compass'. There
may be no reason not to add this to the account, and we will even
extrapolate various things from it-for example, 'Quagmire per-
formed a mathematical feat previously thought to be impossible'.
This example depends on mathematics for the impossibility
in question, but logical impossibilities appear as well. If an author
contradicts himself, then we can sometimes avoid inconsistency in
the account by rejecting some statement as a misprint or an error.
But that is just not plausible in other cases. Suppose that in the
Hatfield-McCoy story alluded to above, each of the eleven cited
sentences is integral to the plot where it OCCUrs6 Then rejecting
any of them would be unfaithful to the story. 5
26 What Are They Like?
Given the account sketched above, we can make some generaliza-

tions concerning fictional objects.
I have been assuming that when we compile our stock of
characters we occasionally include real or unreal objects that are
already familiar from other contexts. For example, in reading the
Conan Doyle novels, we include London-the real London-
among the stock of characters. This will be what I called in chap-
ter 3 an immigrant object. Immigrants need not be real; I take it
5. This will probably create difficulties for Lewis's theory as presented
in [T/F], and for Woods's intended application in [LF] (see Parsons [RW]
note 14).
that in The Seven Percent Solution the character who is there called
'Sherlock Holmes' is an immigrant object, one that is native to
the Conan Doyle novels. The Seven Percent Solution is a fictionalized
account about Sherlock Holmes (and others) in much the same
way that Burr is a fictionalized account about Aaron Burr (and
others). These immigrant objects have de re judgments made
about them of the form 'In the story, x is such and such'; some-
times the 'x is such and such'- is true ("outside" the story) and
sometimes not. In The Seven Percent Solution Holmes was a detec-
tive with delusions about Mori arty ; actually he was a detective
who didn't have delusions about Moriarty,
As I said in chapter 3, I think that someone could con-
sistently maintain that there are no immigrant objects, but only
surrogates thereof However, on the account sketched above of
stockpiling characters, we must, in the controversial cases, either
stockpile an immigrant or a surrogate, but not both. We either
interpret The Seven Percent Solution as being about Sherlock
Holmes, or as being about a detective who resembles him in
some ways, but we do not do both simultaneously. There is only
one character there named 'Sherlock Holmes', not two.
Characters native to (created in) a story have all and only
those nuclear properties attributed to them in the story. This
means that such objects will typically be highly incomplete, and
will occasionally be impossible. Two sorts of incompleteness may
be distinguished here = radical and nonradical incompleteness
(this point is due to Howell [FO:HAHA]). Nonradical incom-
pleteness is the notion of incompleteness that I defined earlier; an
object is incomplete in this sense if, for some nuclear property p,
it lacks both p and the (nuclear) negation of PA Objects created in
ordinary fiction are always incomplete in this sense, because the
account specified in an ordinary story is never completely specific.
The novels do not specify, even implicitly, whether or not Holmes
has a mole on his back, so he lacks the nuclear property of having
a mole on his back and he lacks the nuclear negation of this prop-
erty as well. An object is radically incomplete if it has a property of
the form: not-having-p-and-not-having-the-negation-of-p (in the
notation of chapter 5, if it has [ANX( '" px & '" px)]). An object
which has such a property is impossible, for that is a property
which no real object could have. Holmes is not radically incom..

plete, For example, he does not have the nuclear analogue of
neither-having-a-mole..nor-not-having-a-mole, for the novels do
not specify that he neither has nor lacks a mole; they simply fail
to specify whether he does or he doesn't. Some fictional objects
(e.g., of some sorts of fantasy) may be radically incomplete, but
Holmes is not..
Fictional objects which are not radically incomplete may
still be impossible. If we read at one point that Watson's old
war wound is in his leg, and we read elsewhere that it is in his
arm, then Watson may turn out to be an impossible object. That
depends on how the story goes. vVe might discount one of the
two statements as a slip by the author. But we might not. In
particular, if each statement is integral to the plot where it oc-
curs, we might add both to the account, especially if they are so
widely separated that we don't notice the incompatibility.. And
we may very well fill in 'the wound is in his leg and not elsewhere',
together with 'Watson's arm is not located where his leg is', etc.,
until we have actually got an inconsistency in the account. And
this may eventually lead us to attribute to him a set of nuclear
properties such that no real object could have every property in
the set; if so, Watson will be impossible. .
I am assuming that in the account we typically do not dis-
tinguish the watered..down version of an extranuclear property
from the property itself, in the sense that we automatically add
the one to the account if we add the other, unless there is some
reason not to. This principle holds, of course, only for objects
which exist in the story, and is only clearcut where the story is
fairly realistic-so that we are led to make the same kinds of
presumptions about the objects in the account that we would
normally make about objects that we believed to exist.
With one exception, a character native to a piece of fiction
will never turn out to be a real person, even if the story happens
to truly describe them, in the sense that if it were interpreted as
being an account of their life it would contain only truths. This
is because a character created in a piece of fiction is typically
incomplete, whereas real people are complete. This seems to be
the accepted view in the literature. The one exception, on the
FICTIONAL OBJECTSJ DREAM OBJECTS, AND OTHERS 185
theory given here, is this: if a piece of fiction accidentally turned

out to be a complete and completely accurate account of the
entire (past, present and future) history of the universe, then all
characters created therein would be identical with real objects.
Some will think that this shows the theory to be incorrect. My
own feeling is that the hypothesis that a work of fiction could
accidentally coincide with the entire history of the universe is so
bizarre that it undercuts any intuitions we might have developed
about these matters on the basis of normal cases.
The theory I have given determines certain properties of
fictional objects and leaves certain others open. It determines all
of their nuclear properties and some of their extranuclear prop-
erties. The nuclear property determination is relatively straight-
forward, except that some care must be taken in dealing with
complex nuclear predications, Here is an illustration of that.
Following Lewis [TIF] , let us suppose that the real building
located at 22IB Baker Street is in fact a brewery. Then, since
Holmes has the property of having lived at 22IB Baker Street,
can we infer that he lived in a brewery? In one sense we can, but
not in the sense that is naturally called to mind by these words.
Suppose that we use T for 'is a building located at 22 I B Baker
Street', B for 'is a brewery located at 22IB Baker Street', L for
'lived at', and h for 'Holmes', Then the following are true:
[ANX(3y )(E ry & 1Y & x[Ly])]h
and:
(3y)(Ery & Ty & h[Ly]).
The first says that Holmes has the nuclear property of having.
lived-in-a-real-building-Iocated-a t -22 I B.. Baker.. Street; clearly, he
has this since it is attributed to him in the story. The second says
that there is a real building located at 22 IB Baker Street, and
Holmes has the nuclear property of having-lived-in-it. In saying
this, I am assuming that the building in question is an immigrant
object, and the claim in question follows from the simpler one:
h[Lt],
in which t denotes the building in question, together with the
factual assumption that t is real and is located at 22IB Baker

Street. Now we add the hypothesis that the building in question
is in fact a brewery; this justifies :
(y).(Ety =' (Ty == By)),
which lets us then infer:
(3y)(E!y & By & h[Ly]);
that is, there is a real building located at 22 rB Baker Street,

which is a brewery, and Holmes has the nuclear property of
having-lived-at-it. This is the sense in which 'Holmes lived in a
brewery' is true. (If we had not assumed that the building in
question were an" immigrant object, we could not have inferred
even this.) But we cannot infer that Holmes has the nuclear
property of having-lived-in-a-brewery; that is, we cannot infer:
[ANX(3y)(E~ & By & x[Ly])]h.
In fact, the theory says that this is not true, for the property in
question is not attributed to Holmes in the novel, even implicitly..
The theory is neutral about many extranuclear "predica-
tions:
(I) Intentional properties: Was Sherlock Holmes thought
about by Meinong? I don't know, because I don't know what
Meinong thought about, The theory leaves it open..
(2) Essences: The theory has as a consequence that Holmes
doesn't exist (because he's incomplete); but might Holmes have
existed? In other words, does Holmes have the de re extranuclear
property of possible existence ([AxOE!x])? The account is silent
about this, and that's fine with me, since I'm not at all sure what
the right answer should be . (We know that Holmes has the
watered-down version of this property since he has possible ex..
istence in the story-as a trivial consequence of his having ex-
istence in the story, But since he isn't real, we can't presume any
equivalence between his actual possession of watered-down
possible existence and possession of full-fledged possible exist-
ence.) The theory is also neutral about whether he is necessarily
a detective, necessarily incomplete, etc.
(3) Counterfactual properties: Some people think that fictional
objects can be distinguished by the possession of certain counter-

factual properties. For example, they think that Holmes has:
[Ax (If Conan Doyle had written slightly differently, then x
would have been slightly less clever)]
(so that Holmes is somehow ontologically dependent on his
creator), instead of:
[;tx (If Conan Doyle had written slightly differently, then
Doyle would not have written about x at all, but instead
would have written about some other object)].
I find it impossible to make up my mind about these, but I don't
think that this indicates anything especially peculiar about non-
existence or fictionality. Instead, it has something to do with
Holmes being an "artifact" created by Doyle, Compare these
with the following properties appropriate to a vase:
[ilx (If the potter had moved his/her "hand slightly differ-
ently, x would have been slightly less round)],
versus:
f,<x (If the potter had moved his/her hand slightly dif-
ferently, the potter would not have made x at all, but
instead would have made a different vase)].
I can't choose between these either. The theory is neutral on
such issues.
3. Digression Concerning the Author
So many people have written making the point that an author

does not assert when writing fiction that I feel it prudent to insist
that I have not been discussing the speech acts (or writing acts)
of authors. I do not know how to give a positive account of the
illocutionary, or even the locutionary, aspects of utterances by
an author. Clearly, the meanings of an author's words cannot
differ too much from their meanings in nonfictional settings;
otherwise, we would not be able to understand the story. But this
leaves a lot of issues open. For example, I do not know whether,
when Doyle wrote the words 'Sherlock Holmes', he was refer-

ring to the same object that I refer to when I use the same words,
or even whether he was referring at all (see Davidson [RRS]).
I have said that, in a popular sense, an author creates char-
acters, but this too is hard to analyze. It does not mean, for
example, that the author brings those characters into existence,
for they do not exist. Nor does he or she make them objects, for
they were objects before they appeared in stories. We might say,
I suppose} that the author makes them fictional objects," and that
they were not fictional objects before the creative act. We might
even say that the author bestows on them fictional existence (i.e.
[Ax(3y)(y is a real work of fiction & iny, x exists)]), as long as this
is not confused with ordinary existence.
It is a consequence of my view that if two authors were to
write exactly the same story, they would thereby have created
the same characters. And this makes it theoretically possible for
an author to create a character that has already been created.
Some people have objected to this aspect of the theory. But my
feeling is that this is no more objectionable than is the view that
if two authors had independently typed out the same words,
working within roughly the same literary tradition, they would
have thereby written the same story. Theoretically, an author
.could write a novel that has already been written; this wouldn't
automatically make the second author any the less creative.
4. Unclarities in the Theory
There are at least two sources of unclarity in the theory being

discussed: unclarity in the selection of the relevant body of
literature and unclarity in deciding what is true according to
that body of literature, once it has been selected. The first is often
quite unproblematic. For example, Robinson Crusoe was created
in Defoe's novel by the same name; the relevant story includes
what is true according to that novel and does not include any-
thing due to Crusoe's appearances in other stories by later
authors. Sherlock Holmes is a more complicated case, for here
6. This suggestion" is from David Kaplan.

there is a whole series of novels and short stories (all by Conan

Doyle) to take into account, and they don't always cohere very
well with one another. But really the situation here is no dif-
ferent from that of a single novel which consists of a whole series
of chapters, or a single short story that consists of a whole series
of paragraphs. Usually a series of novels coheres less well than
the chapters of a single novel, but .there is no difference in prin-
ciple. The major problems will occur when it isn't clear whether
a given story is to he included in the creating corpus for a char-
acter or not. But if it is really unclear whether or not to include
it, we will probably be dealing with a situation in which it is
not clear, pretheoretically, which fictional character we are
dealing with. The theory could be made artificially clear here, but
it would be an imposed clarity that would be untrue to the sub-
ject matter.
Occasionally, we will not want to include a whole corpus
even when it is readily identifiable. This would happen if, for
example, we wanted to compare "the Sherlock Holmes of The
Hound oJ the Baskeroilles" with "the Sherlock Holmes of The
Valley of Fear." Here I think we would compare an object native
to the former novel with one native to the latter; each of them
would resemble the Holmes of the whole series in many ways,
but neither would be identical with him. This sort of linguistic
context can also affect the native/immigrant question. For
example, the phrase 'the Sherlock Holmes of The Seven Percent
Solution' most naturally refers to the surrogate Sherlock Holmes
of that story, not to the immigrant from the Doyle novels.
Even if it is clear which story to use, there may still be prob-
lems of interpretation. If you and I interpret a story differently
(and this will be the normal situation), we will possess different
accounts of what is true in the story, and if we use your account
to determine a character, we will get a different object than if
we use mine. There seem to be two options here: (i) we are both
equally right, and we are talking about different characters, or
(ii) the character we are talking about is that character deter-
mined by the right account (which might be neither of ours).
But (i) seems to be untrue to the way we talk about fictional
objects, and (ii) makes a dubious assumption about literary
interpretation. This bothers me. But it would bother me more

if it were a problem peculiar to the present theory. I suspect
instead that it is a fairly ubiquitous one. For the.same problem
arises for anyone who wishes to talk about sets of physical objects.
Suppose that you and I are discussing "the set of living things."
Then, recalling that there is a lot of controversy about what
things are alive (e.g., viruses), which set are we discussing? If
you and I have different opinions about what living things there
are, are we (i) discussing different sets, or (ii) both discussing the
"right" set? The example of the set of living things is notoriously
difficult, but the difficulty is not atypical. Most notions that we
use to pick out sets are notions that allow of borderline cases>
and in the case of sets, as with fictional objects (according to the
present theory), decisions about the borderlines influence deci-
sions about the identities of the objects in question.
This kind of problem is not restricted to set theory. The
literature on meaning change in science suggests that it may be
no easier to determine the reference of a scientific term than it is
to determine the reference of a fictional name in the present
theory. Perhaps some useful results in the philosophy of science
could be exploited here.?
These reflections don't improve the theory much, but they
do put some of its difficulties in perspective.
5- Indiscernibles : Crowds and Siblings
The theory of fictional objects being discussed is an attempt to

apply the ontology developed in parts I and 11 to fictional objects
as pretheoretically conceived.. Maybe it won't work. There are
in fact a host of issues that suggest there are more fictional objects
than are provided for in the ontology. The next three sections
can be seen as dealing with three ways in which the theory might
not be providing enough fictional objects.
The concern of this section is with the following type of
situation: a story may "say something about" a group of things
without saying anything at all about the members of the group,
7. See Field [TC&lR] and K. Pyne Parsons [CMC].

FICTIONAL OBJECTS;, DREAM OBJECTS, AND 0 THERS 191
or at least without saying enough for us to distinguish one mem-

ber from another. In such a case the theory does not identify the
members of the group as fictional objects. I think this is correct,
but some have thought otherwise.
An example which illustrates the phenomenon and which
simultaneously casts the theory in a favorable light is the crowd
example from chapter 3: if a story says "a crowd gathered" but
says nothing at all individually about the members of the crowd,
then there is no way to get the theory to yield any particular
member of the crowd as a fictional object~ I think that this in
fact corresponds to our practice in discussing fictional objects.
At least, it is a view that is automatically immune to straight-
forward counterexample. For a counterexample would have to
consist in specifying a particular fictional object (a particular
crowd member) which the theory fails to accommodate. But this
can't be done, because in the case mentioned no one has the
resources to refer to a particular crowd member. Phrases like
'the first member' fail because they do not specify first in what
respect; they presuppose an ordering of the crowd members
that is not given in the story.. And phrases-like 'the crowd mem-
ber that I now have in mind' fail because no one can have a
particular object in mind if he knows of no respect at all in which
that object differs from certain others.
Other examples look initially less clear-cut, but I think they
come to the same thing. Suppose we are reading a story which
mentions a pair of twins; the story says a few things about them,
but it in no way distinguishes them from one another. In reading
the story, you naturally expect to learn of something to distin-
guish them, and you expect each- of them to emerge as a character
in its own right-but it never happens. Do 'we here have two
fictional objects, indistinguishable within the theory F' I think
not. This is just like the crowd case (except that here we have
had some expectations frustrated). The pair of twins is a fictional
object created in the story, but neither member of the pair is.
Things are somewhat confusing because we use the phrase 'the
twins' to refer to the pair, and we say things like 'The twins were
B. An example due to David Kaplan.

created by the author'; the use of the plural here suggests that we
mean something like 'Each of the twins was created by the au-
thor'. But we don't ; the phrase 'the twins' has its group reading..
Things can be made more complicated. Suppose we have a
play whose script lists two characters named 'Rosencrantz' and
'Guildenstern', but which specifies that they always act and talk
in unison (and no other character treats one differently than the
other). Then perhaps the theory does not distinguish them. Yet
it seems that we can truly say that Smith portrayed Rosencrantz,
but he did not portray Guildenstern, so they must be distinct
objects.f The example deals with plays, not stories, and this can
raise special problems, but I think it is a fair example in the
present context. The solution depends on how the use of the
names in the script is to be interpreted. Normally, the names in
a script count as part of the story: they presuppose that the
characters (according to the story) bore those names. But bearing
a given name within a given society is a perfectly ordinary
anthropological property of an individual, (It is this anthropo..
logical relation ofa name to its bearer that I think is the focus of
the "causal" or "historical" theory of names.) So the script does
provide a theoretical means of distinguishing the characters:
one has the nuclear property of being named 'Rosencrantz" and
the other instead has the nuclear property of being named
'Guildenstern' .10
The example may be pushed another way. Suppose that it
is somehow made clear that the names used in a story are not the
.narnes borne by the characters (according to the story). For
example, the. author states: "the names have been changed to
protect the innocent." This sort of example divides into two
sorts of cases:
Case I : The statement cited is made in the course of the story
in such a manner that, in the story, the statement is being made
by the narrator N, who is then himself a character of the story.
9. This example is due to Saul Kripke.

10.I am told that Alvin Plantinga has named everything in the universe
"Charlie." I think that this can perhaps be done, if 'naming" is understood in
its extranuclear sense. But not in the nuclear sense. If I apply for a driver's
license and list my name as 'Charlie', no appeal to Plantinga's activity can
make the entry correct.
FICTIONAL OBJECTS, DREAM OBJECTS~ AND OTHERS 193
Then in the story one character has the property of being called
'Rosencrantz' by N, and either this property or its watered-down
version will distinguish it from the other object within the theory.
(If this answer seems bizarre and concocted, recall that it is a
response to a bizarre and concocted objection.)
Case 2: Maybe the statement is clearly made outside the
story. But then in the story the characters have the names used to
refer to them there, and this is just like the original case. The
author has merely stated, outside the story, that the names used
for characters in the story are not the same as the names actually
borne by the real people on whom the characters are based.
Suppose that we return to the play case, then, and suppose
that the script makes clear that in the play both characters have
the same name. It would then be very peculiar to list them under
the names 'Rosencrantz' and 'Guildenstern', but something like
this happens when a play lists. roles as 'first crowd member',
'second crowd member', etc. Usually this is done only when the
different crowd members do different things in the play, but sup-
pose that in this particular play that happens not to be the case.
Then I think we have returned to an example just like the original
crowd example: the purported charactors of the original play
taken in isolation are not characters at all. The impression that
there must be distinct characters here comes, I think, from the
fact that in most performances of the play (though not all) there
would be distinct characters, portrayed by different actors. But
the characters of a performance of a play are typically distinct
from the characters of the original script (this must be true
since different performances often have different numbers of
characters) .
One might attempt to exploit the logic of the situation to
force a multiplicity of objects in some of the cases we have been
discussing. Suppose that in the twins case the story contains their
mother as a character. Call her m. Then in the story m has the
nuclear property of having two distinct children; so according to
the theory she has that property. That is, she has:
[A NX(3y )(3z)(y =F Z & x hasy & x has z)].
Then it would seem that there are two distinct children, each of
which m has:
(3y)(3z)(y :f:. z.& mhasy & mhasz).

No, this does not follow. The latter statement is true in the story,
but we cannot in general infer that whatever is true in the story
is actually true. We can export certain nuclear predications, so
that we can infer that m has the nuclear property listed above,
but the latter statement does not follow from this. This was the
sort of mistake discussed near the end of section 2 ..
6. Relations to Unreal Objects
According to the story, Holmes talked to Watson. So, according

to the story, Holmes has the property of having. talked to Watson,
and Watson has the property of having been talked to by Holmes,
Then, by the theory, each of them has those properties:
h[Tw] & [hT]w.
Neither Holmes nor Watson exists, and neither is identical to the

other, so:
h :f; W & ",E!h & ~E!w & h[Tw] & [hT]w,
which entails:
(3x)(3y)(3RN)(X #y & ",E!x & ~Ery & x[Ry] & [xR]y).
Recall that x[Ry] & [xR]y is the theoretical analysis of 'x stands
in relation R toy', so the above may be summed up as:
Some distinct nonexistent objects stand in nuclear relations to one
another. This is interesting because it is not a theorem of the basic
theory, and because it has consequences for our view of what
nonexistent objects can be like. It is an extension of the basic
theory, got by combining that theory with its application to
fictional objects.
One consequence of this version of the theory is that it is not
plausible to maintain a certain kind of constructioist view of non-
existent objects. Here is the view I have in mind. Suppose that we
start with an initial stock of objects, the existent ones, EX, and an
initial stock of nuclear properties Po, which does not include any
relational properties got by plugging up nuclear relations with
nonexistent objects. In other words, we start with materials to-
FICTIONAL OBJECTS, DREAM OBJECTS;, AND 0 THERS 195
tally acceptable to the orthodoxy. Then we "make" some non-

existent objects by making arbitrary sets of nuclear properties.
We can either suppose that the objects so constructed arethe sets
produced, or that producing such a set somehow produces an
object corresponding to it. Call the set of such objects, including
those in EX, OB l - Now, using the objects in OB l , we can extend
our original set of properties Po to a new set PI by adding those
relational properties that are "made" by plugging up nuclear
relations with arbitrary members of OB I . Then we again make
objects out of sets of"members of PI' yielding a new set of objects,
OB z. Some members of OB z are already in OB b because some
members of PI were already in Po, but some are new. We then
continue the process, getting P2' OB g, Ps, .. ~ . Maybe we even
extend the process into the transfinite, in an attempt to ensure
that we have enough objects for our needs. (If we assume that
every time we plug up a nuclear relation with a new object we
get a new property, then the process never terminates. The result
is a theory reminiscent of Zermelo-Fraenkel set theory. We
never get, for example, a "universal" object, an object which has
all nuclear properties; so if we need such an object, it will be im...
possible to go far enough to meet our needs.)
This particular theory is not a version of the basic theory; it
is incompatible with axiom OB] (which, e.g., yields a universal
object). But if we liked it enough, we might be willing to modify
the basic theory in its direction. One trouble with it would be
that it is incompatible with the theory of fictional objects under
discussion. Here is why. On the constructivist theory, for every
nonexistent object there is an earliest stage at which it is con-
structed. Then suppose that Holmes is first made at stage a.
Since Holmes has the property of having. talked to Watson, that
property musthave been made before stage (X. So Watson must
have been made even earlier; that is, Watson must have been
constructed before Holmes. But a similar argument shows that
Holmes must have been constructed before Watson, and both of
these cannot be true.
Fictional objects created in the same story are on a par;
constructively, they must be made simultaneously, if made at all.
But then they cannot be made out of things (relational properties)
which are made out of their compatriots. So fictional objects

provide an impediment for a joint constructivist view of non-
existent objects and relational properties.
There is, however, a quasi-constructivist view which mqy be
compatible both with the basic theory and its application to
fictional objects. The idea is to maintain a constructivist view of
objects, while ignoring, insofar as possible, relational properties
got by plugging up nuclear relations with unreal objects. We
cannot deny that there are such properties without significant
alteration of the basic theory, but we can "trivialize" them to get
much the same effect. We are already assuming that it is impos-
sible for any real object to have such a property. Suppose, then,
that we identify all such properties with some particular impos-
sible nuclear property (some nuclear property such that. it's
impossible for any real object to have it). We pick some such
property, call it Po, and we add the axiom:
(X)(RN)("'E!x::J ([xRN] = Po & [RNx] = Po)).
In terms of the constructivist view sketched earlier, this amounts
to giving up the assumption that we can initially distinguish re..
lational properties got by plugging up nuclear relations with non-
existent objects from other properties; in particular, we assume
that all such properties are already in Po, since they are all identi-
cal to Po, which is in Po- Then OBI turns out to contain all objects,
since PI contains nothing that is not already in Po, and so OBz
contributes no new objects; OB I = OB 2 = OB3 = . . . and
Po = PI = P2 = .... 11
Is this view at all viable? There are at least two sorts of
problems with it. The first has to do with the central issue of
sections 5-7: if so many relational properties are identical with one
another, we may not have enough nuclear properties to distin-
guish different fictional objects. For example, suppose that I
rewrite the Conan Doyle novels by interchanging certain charac-
ters in certain episodes; say that Holmes occasionally discovers
that Watson, instead of Moriarty, has committed the crime. My
Holmes would be a different fictional object than Doyle's, but
I I. This quasi-constructivist view guided much of my early work on this
theory; I am now quite doubtful that it is the best approach.
FICTIOJVAL OBJECTS:J DREAM OBJECTS, AND OTHERS 197
mightn't the theory identify them? For example, we cannot now

distinguish my Holmes from Doyle's by claiming that mine has
"attributed a crime to Watson" and Doyle's does not, for Doyle's
has "attributed a crime to Moriarty," and that is the same nuclear
property as "attributed a crime to Watson." But, of course, my
Holmes and Doyle's will differ in other respects. Mine will have
"attributed a crime to someone he lived with," which Doyle's
will not, and we have not proposed identifying these properties
with one another. (See Parsons [MAFO] for more discussion of
this point.)
Perhaps the view in question is immune to the first objection,
but there is another. Let us use w for Watson, m for Moriarty,
and L for lived with. Then we seem to have the following truths
(where wand m are understood to occur de re):
(a) According to the story, Holmes has [Lw].
(b) according to the story, Holmes has [Lm].
IV
(c) [Lw] = [Lm].

But according to the logic employed in the formal theory, the
property identity in (c) permits intersubstitutivity in (a) and (b),
so (a)-Cc) are inconsistent. And this is one of the reasons that I
have for not liking the quasi-constructivist approach; it seems to
me that the logic is right in not permitting (c) to be true In the
face of (a) and (b). However, this is a fairly abstract issue on
which there is plenty of room for disagreement.P so others may
not regard this as seriously undermining the approach.
7. Extranuclear Predications; Stones within Stories
Typically, stories attribute to their characters extranuclear

properties as well as nuclear ones. The theory in question says
that these characters have the nuclear properties so attributed,
but does not say this for the extranuclear ones. In many cases
this is essential. According to the story, Holmes exists; but we
would not want to attribute existence to him on that account,
12. One might, for example, argue that it is not true that according to the
story Holmes has [Lw]; it is only true that he has [lNxW[A)(xLy)]], where Wis
the proper name 'Watson'; recall the discussion of names in chapter 5, section 4.
for he does not exist. There are cases, of course, in which a fic-
tional object does have some of the extranuclear properties
attributed to it in the story. Holmes has the extranuclear property
of having been thought about by criminals, but that is because
they have read the novels, not because it says so there; had
Doyle not published his work, Holmes would have lacked this
property..
These facts lead to some possible difficulties with the theory;
these difficulties form the subject matter of this section.
To begin, suppose that a character in a story is given only
extranuclear properties in that story. Then the theory may say
either that there is no such character, or that the character is
the null object, both of which sound suspicious. Are there such
characters? Suppose that this is the story:
Story: "Jay exists. The end."
No, this won't do) for two reasons. First) the story is too meager
to be plausible; pretheoretically, it is unclear that this is a story,
or that it has a character. And second, if we were to insist on the
example, then the theory in fact would identify the hero of the
story as a certain nonnull object; it would be an object which
has at least watered-down existence, and the property of being
named "Jay." Let us try avoiding such properties by omitting
the name and changing existence to nonexistence:
Story: "An object doesn't exist. The end."
But now we pretty clearly have a "story" without a main char-
acter, so the theory does not apply incorrectly here. The moral
of these examples seems to be that if we want to make trouble
for the theory, we're going to have to produce stories which have
fairly detailed 'and well-delineated characters; extremely brief
and abstract examples just aren't convincing. And since real
objects have the watered-down versions of all the extranuclear
properties that they have, we will probably have to have recourse
to characters which, according to the stories in which they occur,
are not real. AIl this is. quite difficult to do in a convincing
manner (the reader is invited to try), but there is a closely re-
lated strategy which seems more promising. Try to falsify the
FICTIONAL OBJECTS3 DREAM OBJECTS, AND 0 THERS 199
theory by producing examples of pairs of fictional objects, objects

which we would distinguish pretheoretically, but which are
alike in all of their nuclear properties. This allows us to attribute
to each object a large number of nuclear properties, which makes
the example realistic, without automatically defeating the strat-
egy designed to produce a counterexample,
Here is an illustrative example :13
Story: "On my way home from work I encountered a thug
who attempted to hold me up. I said to him, 'In a moment a
policeman will come around the corner and capture you 4 '
A moment later a policeman came around the corner

and captured him. The end."
The difficulty is supposed to be this: that there are two police-
men in the story, the "made-up" one PM) and the "real" one Pr-
But they differ only with respect to extranuclear properties, such
as existence, having been mentioned by the narrator, etc.
There are various weaknesses in the objection. First, we
might maintain that in the story the policeman who came around
the corner was the policeman described to the thug, so that
there would in fact only be one policeman in the story. But this
isn't very convincing; I don't think that it is the right way to
interpret the given story, and if it were, we could probably alter
the story in such a manner as to avoid this interpretation. Second,
and more interesting, one might maintain that PM is not a char-
acter of the story (so again there is only one policeman in the
story). The narrator said, "In a moment a policeman will 4 •• ,"
but this just reports what he said, it does not introduce a new
character into the story, I think that this is the right way to inter-
pret this story. -However, this way out could be avoided by
embellishing the example somewhat. For example, the narrator
could have told the thug a fairly lengthy story about a policeman,
and the story could then have "come true." So let me suppose
that the story has been sufficiently embellished so that it is clear
that according to the story there are two policeman, that PM #
Pr. This supposition is needed if the story is to provide a counter..
example.
13. This example originated with Robert M. Adams.
But now the theory does distinguish the two policemen.

We know that according to the story, PM "# Pr. So according to
the story:
We also know that according to the story Pr exists, so, in the

story, Pr also has the watered-down version of the above property;
it has:
But then this is a nuclear property that Pr has in the story, and it
is a property that there is no reason at all to attribute to PM in
the story. So the two policemen do have different nuclear prop-
erties in the story, and the theory does distinguish them from one
another.
One might, of course, attack the above reasoning, for there
are limits to the inferences that one can make within stories (see
section I above). But normal inferences can be made, so long as
they are not unduly complex, and so long as they contribute to,
rather than detract from, the surface coherence of the story. Our
inferring that Pr has [jtNx(PM =1= x)] is a case of such a normal
inference.
This sort of reasoning will suffice to avoid most realistic
counterexamples based on characters which are distinguished
within a story "only" by the fact that one is real and the other is
not. But there is one more type of possible counterexample (the
last that I'll discuss) which is not so easy to avoid. The idea is
to have two stories, each of which contains subsidiary stories (or
dreams) or legends) about objects which, according to the main
story, do not exist. Then if the subsidiary stories are sufficiently
similar, perhaps the theory will be unable to distinguish char-
acters of the subsidiary stories when they should be so distin-
guished. But before discussing this sort of objection, we need to
see how the theory applies to stories within stories.
Suppose that a story occurs within another story, such as the
famous case of the play-within-a-play in Hamlet. How does the
theory apply to the characters of the subsidiary story ? Well, it
identifies such a character as the object which has all and only
those nuclear properties attributed to it in the story-that is, in

the main story. These properties are determined by the reader
in two different ways: (I) often the main story tells us some things
about the plot of the subsidiary story, and we can judge for
ourselves some of the nuclear properties possessed by the char..
acters ef the subsidiary story-roughly by imagining ourselves to
be characters of the main story who are reading or"hearing the
subsidiary story; and (2) sometimes we discover things about the
characters of the subsidiary story by things that the characters
of the main story (or possibly the author) say about them.
Occasionally, we get the entire subsidiary story verbatim, but,
typically, we learn a great deal about it by means (2). For exam..
pIe, in Hamlet we discover from Hamlet, not from the reported
acting out of the play-within-the-play, that Gonzago's murder
took place in a garden, and that the motive was economic.
Characters of stories-within-stories, then, will be relatively
ordinary fictional objects, with ordinary nuclear properties. They
will typically be ('isolated". from the characters of the main
stories in that they will not be related to them by ordinary nu-
clear relations. Nor will they stand in watered-down versions of
typical extranuclear relations to them. For example, in Hamlet
Gonzago has the extranuclear property of being thought about
by Hamlet, but, being unreal (in Hamlet), he does not inherit the
watered-down version of this property.
Now what would a counterexample to the theory (of the
sort under discussion) be like? Well, it would consist of two
stories, each containing subsidiary stories, such that:
(I) the subsidiary stories are given to us, by one means or
another, with enough detail so that we would judge
them to give rise to genuine characters, and
(2) it is clear, pretheoretically, that the stories contain
different characters, but
(3) the theory identifies them as the same characters.
The power of the theory here is that it will only identify the
characters as the same if the same nuclear things are true of them
in the two stories, and that makes it difficult to satisfy (I) and
(2)~ In fact, I have been unable to construct a clear and compel-
202 APPLICATIOJVS AND DISCUSSION
ling counterexample along these lines. However, this could be

because of a feebleness of motivation; perhaps others will have
better luck.
8. Alternative Theories
Most theories of fiction take the orthodox position regarding fic-

tional objects: there aren't any. Many writers have commented
on the implausibility of this view, but hardly any have offered a
theory that is developed in any detail. (Plantinga [XX] proposes
a possible worlds account, but he proposes it only as an example
of how not to develop such a theory.) One exception to this is an
article by Peter van Inwagen ([COF]). Here is a liberal para-
phrase of that theory:
(a) fictional objects are theoretical entities of literary criti-
cism;
(b) fictional objects exist;
(c) typically, a fictional object does not have the properties
attributed to it in stories; and
(cl) for any property p which is ascribed to a fictional object x
in a story s, x has the property of having-p-ascribed..to-
it-in-s.
The notion of "ascribed-to" is a technical one for van Inwagen;
in terms of my discussion above, we can equate 'p is ascribed to
x in s' roughly with 'in s, x has p' (see [COF] for discussion). Van
Inwagen does not discuss the native/immigrant distinction in
detail, and the theory could take many different forms, depend..
ing on how this is treated. I'll try to avoid this multiplicity of
versions by avoiding the issue; I'll limit my discussion to those
fictional objects which are native to a story and which do not
appear as immigrants in any other stories.
Suppose that we temporarily ignore (a) and (b). Then we
are left with the outline of a theory of fictional objects which can
be formulated within the ontology of parts I and 11 of this book.
Instead of:
Sherlock Holmes = the object which has exactly those nu..
clear properties attributed to Holmes in the Conan Doyle
novels,
FICTIONAL OBJECTS, DREAM OBJECTS~ AJVD OTHERS 203
we have:
Sherlock Holmes ~ the object which has exactly those
nuclear properties of the form "having p in s," where p is a
(nuclear or extranuclear) property attributed to Holrnes in
the Conan Doyle novels, and s is the account determined by
those novels .
Let me call the former theory 'Mod I' and the latter 'Mod 2'.
Then Mod 2 has at least one possible theoretical advantage over
Mod I : it seems to bypass at least some of the worries about ex-
tranuclear predications within stories that were raised in section
7. For if one character has P attributed to it in s, and another
character has P' attributed to it in s' (where P and P' are dis-
tinct), then the former (but not the latter) has [ANX(X has P in
s)], and the latter (but not the former) has [ANx(x has pi in s')],
and so the theory automatically distinguishes them. (Actually it
does so only if [;{Nx(x has Pin s)] and [ANX(X has P' in Sf)} are
different nuclear properties. This could fail to happen, for it is
not in general true that distinct extranuclear properties have dis-
tinct watered-down versions.H But this could probably be main-
14. We can prove that for some P and Q, P #- Q and wP = wQ, by the
following adaptation of Cantor's theorem.
By axiom OB] there is some object x o, such that:
(1) (P)(PX o == (P)(P = wP ~ (y)(Py => ,.... py))).
By axiom AB(E) there is some Qsuch that:

(2) (y)(Qy == y = X Q) ..
Substituting wQ. for p in (I) yields:
(3) wQxo == (P){wQ = wP ::) (y)(Py :::> ~wQy)).
Now we can show that "'W wQxo• For assume otherwise. Then by (3) we
have:
(4) (P) (wQ = wP ::;) (y) (Py =' "'WwQy)).
Substituting Q for P in (4) and using the law of identity yields:

(5) (y)(Qy ::> "'wQy).
And so:
(6) Qxo ::J I'VwQxo•
204 APPLICATIONS AND DISCUSSION·
tained for typical cases of extranuclear predications in stories.)

With regard to other problems discussed above, Mod 2 seems to
be pretty much on a par with Mod I, provided that "immigrant
objects are treated appropriately.
Does Mod 2 have any additional difficulties? All that I have
discovered amount to niceties. For example, consider the s-n-
tence:
The fictional detective who lived at 22lB Baker' Street
lived with a doctor.
To make this come out true on Mod 2, we need somewhere in
the semantical account to replace 'is a detective', 'lived at 22IB
Baker Street', etc., by 'is a detective in a story', 'lived at 22IB
Baker Street in a story', etc.-or something like this. A little bit
messy, but not too bad (see discussion of a similar point in [COP]
P: 3°5). Notice that these paraphrases do not eliminate commit-
ment to fictional objects, so they are not the sought-after para-
phrases of chapter 2.
Let us return now to claims (a) and (b). In (a), van Inwagen
says that fictional objects are theoretical entities of literary criti-
cism. Well, the intent of both Mods I and 2 is that the fictional
objects described therein should form part of the subject matter of
literary criticism, so in this respect all theories may agree. More
may be meant by calling them "theoretical entities" than this,
But this contradicts the conjunction of (2) and the assumption that
wQxo- So we have shown :
(7) ~wQ.xo·
This, together with (3), yields:
(8) (3P)(wQ= wP& ~(y)(Py::> --wQy»).
That is, for some P, we have:
(9) wQ = wP and (3y)(Py & wQ.y).
It suffices then to show that P :f:. Q, and this is easy, For assuming
P = Qin (9) yields:
(10) (3y)(Qy & wQy),
which contradicts the conjunction of (2) and (7).
but to pursue this topic in any detail here would take me far
beyond my present task (but see chapter 8, section 6).
In (b), van Inwagen says that fictional objects exist. This is a
little tricky to evaluate, since he also insists that 'there exists'
means the same as 'there is', a view which I do not accept. From
the perspective of the present work there are two ways to inter-
pret this claim. First, we may take his 'there is/exists' to mean
just 'there is'; on this interpretation his theory may be consistent
with Mod 2. (It lacks any detailed ontological development, but
so does Mod 2 if taken in isolation from the rest of the book.. )
Second, we may take his 'there is/exists' to mean 'there exists'.
Then there are two issues to face. First, a superficial difficulty:
Sherlock Holmes doesn't exist, but van Inwagen says he does.
However, van Inwagen can explain this: when we say that
Holmes doesn't exist, we don't say literally what we mean. We
mean something like 'Holmes is not a man' ('There exists no man
who is Holmes') or "Holmes does not have the properties com-
monly attributed to him' ([COF] p. 308). I don't think this is
right, but I have no argument against it that doesn't beg the
question. This is a twist on the paraphrase approach to non-
existent objects: instead of paraphrasing away the apparent ref-
erences to the alleged nonexistent objects, we paraphrase away
the allegation that they don't exist. (However, van Inwagen
explicitly limits this technique to fictional objects.)
If we assume with van Inwagen that fictional objects exist,
Mod 2 cannot be the same as his theory, since according to Mod
2 fictional objects are incomplete, and so they do not exist. Van
Inwagen's objects must be complete, and so they must have lots
of nuclear properties that they do not get by means of attribution
in a story. To say more about the details of his theory here would
be to go beyond his statement of it; let me just suppose that it has
been filled in in some appropriate manner. Then we could com-
pare this enterprise with Mods I and 2 as follows: both recognize
the need for fictional objects, including objects which are said not
to exist, and both add them to our ontology, One approach is to
bloat the realm of existence and the other the realm of nonexis-
tence. As theories offictional objects alone, it is hard to assess ei...
ther of them against the other. Conceivably, the only important

difference between them might turn out to be a fruitless debate
about whether such objects really exist.
9. Fantasy, Myth~ and Legend
Although I have explicitly limited myself to realistic narrative

fiction, departures from this style of story do not typically lead to
new problems; they only enhance old ones. In fantasy writing,
for example, it may be much more difficult to be clear about what
is to go into the account of the story. Sometimes fantasy can be so
coherently done that this problem is no worse than in typical
realistic writing; Tolkien's Lord of the Rings is an example of this.
In this work there are lots of things left undetermined, and some
of them are left undetermined because it is fantasy-s-for example,
the extent to which magic can influence natural events, and how
and by whom it can bepracticed-c-but this is no different in
principle than the underdetermination of the furnishings in
Holmes's sitting room.
.Other examples can be harder. Suppose a story specifies that
in the year 4000, civilization is about to be destroyed by a disease
that is caused by a bacterium that was produced by mutation in
an experiment in the year 2000. A time traveler is sent back into
the past to interfere with the experiment, and does so successfully.
Civilization is saved.
But wait a minute. What did happen (according to the
story) in 4000? Was a time traveler sent back into the past? Why?
Because civilization was about to be wiped out? Not according to
the end of the story; there we are told that the mutation was
prevented in 2000, and so no disease threatened the civilization of
4000. I suggest that this is simply another case of an impossible
story. Its main character will have the watered-down version of
having traveled back into the past, and will also have the (nu-
clear) negation of this property as welL It's more poignant, per-
haps, but no different in principle than realistic fiction.
The hardest issues in fiction arise when authors intentionally
depart from "traditional" forms, say by mixing themselves in
with their characters. This can make the theory quite difficult to
FICTIONAL OBJECTS, DREAM OBJECTS~ AND 0 THERS 207
apply in given cases, but again the difficulty seems appropriate to

the intent ofthe work.
Myths are "traditional stories serving to explain some phe-
nomenon, custom, etc." They make vivid at least two problems.
First, in a myth there will typically be an unclarity regarding
what happens in the story, because there will be no definitive
text; instead there will be a long tradition of tellings of stories
which will diverge from one another in various ways. Here we
probably need to appeal to what is most common and most im . .
portant to the various tellings. As above, I think that unclarity in
the identity ofmythical objects within the theory sketched meshes
with a real unclarity in our pretheoretic conceptions of such
I
objects.
Second, there is often an early time at which the myth is
generally believed to be a true account. At such a stage in history,
apparent references to the participants of the recounted happen-
ings should be treated as genuine failures of reference to real
objects rather than as successful references to unreal ones. When
the situation is in flux, so that some people believe the story and
some don't, there may be unresolvable difficulties regarding
which line to take..
Legends raise a slightly different problem. Suppose that we
know or suspect that a given legend is based on the activities of a
real person, but that we assume that most of the feats recounted
in the legend were not in fact performed by that person. When we
use the name from the legend (e.g.. , 'Noah' or 'Homer'), do we
refer to the real person on whom the legend is based, or to a non-
existent "fictional" object who is native to the story? (This prob-
lem has been discussed recently within the context of the causal
theory of names, where it is usually formulated as the issue of
whether or not a name from a legend is "empty.") There is
probably no simple general answer to this question.
10. Objects in Dreams
At a pretheoreticallevel, no one is puzzled to hear someone say,

"The unicorn I dreamed about last night resembled my depart-
ment chairman." But unicorns don't exist, and so dreams ap...
parently provide us with more examples of objects that don't

exist.
I propose that dream objects be treated, theoretically, al...
most exactly the same as fictional objects. In particular, we may
employ the schema:
The ifJ of dream d = the object which has exactly those
nuclear properties which the rp has in d.
There are many parallels between stories and dreams which make
this plausible. We use the locutions 'in the dream' and 'according
to the dream) in much the same way that we use 'in the story' and
(according to the story'. The immigrant/native distinction also
seems to be relevant. I wake up and report: "1 had a dream about
my dog last night; she was tearing up the living room. Here both
J}
my dog and the living room are immigrant objects. But if I

report; "Last night I had a dream about a big black dog that was
tearing up a living room," then I am probably describing a native
of the dream. As with fiction, the distinction is not always un-
problematic. Suppose that I dream about a purple dragon, and
my psychiatrist tells me that it is really about my first-grade
teacher, It may be hard to tell whether this is correct. At one end
of the spectrum are those cases in which I respond by exclaiming;
"My God, you're right; I didn't realize it!"; then I would say,
ceteris paribus, that the dragon was my first-grade teacher. At the
opposite end of the spectrum are those cases in which the psychia-
trist was making a shot in the dark, and there's no evidence at all
in favor of the proposaL The problem cases come in between. But
none of this should obscure the fact that the distinction is usually
clear.
A major difference between dreams and fiction is this: in
fiction (at least in unillustrated fiction) the account of what
happens in the story comes to us primarily in propositional
form-we construct the account from" a series of sentences-
whereas in dreams the account of what happens in the dream
comes to us by gestalt; it must be constructed by interpreting a
continuum of global impressions, images, feelings, etc. In this
respect, dreams are a kind of thought that does not consist of
inner speech. It is not clear to me whether it's ever possible for
FICTIONAL OBJECTS, DREAM OBJECTSJ AND OTHERS 209
an account originating in a story to be identical to an account

originating in a dream.
The principles of interpretation differ also. In the case of
fiction, literary tradition is relevant, whereas this is not the case
with dreams. (At least, I don't think that there are traditions in
dreaming; there may be traditions in opinion concerning the
proper methods of interpreting dreams, but this is another mat..
ter.) And dreams, of course, present a special problem with re ..
spect to availability ofdata. In a novel, the story is in a sense com-
pletely: present to the reader, whereas dreams seem to proceed in
part at a subconscious level; their availability to the conscious-
ness of the dreamer is a problematic issue. But this is a problem of
evidence, not one of ontology.
An object may be native to a series of dreams, as when some-
one is plagued by a recurring nightmare; this is like a fictional
object which is native to a series of novels. The problems of de-,
termining when this happens don't seem to be different in prin-
ciple.
In the case of dreams the incompleteness and occasional im-
possibility of their objects is vivid. It is typically very difficult to
describe a dream, even to oneself, without imposing details on the
characters and happenings in a way that distorts the dream. This
is because dream objects tend to be highly incomplete and, more
important, they tend to be indeterminate with respect to those
properties in terms of which we normally "identify" things. I
dream about a person with a face, but with no particular face, a
person who is either-male-or-female but whose sex is completely
undetermined. In conceptualizing dream objects, we instinctively
try to treat them as existing objects; we expect them to be deter-
minate in every detail, and we try to fill in the details as we are
accustomed to do with existing objects. But it doesn't work, and
this is one of the things that is so distinctive about them.
Dream objects provide some of the nicest examples of im-
possible objects. I dream that I am in a room; the ceiling is above
me but you get to it by going down those stairs in the corner. The
tiles on the floor keep passing through one another, yet with
absolutely. no change in their relative geometrical relationships.
A horse shows up; it is also a person, yet clearly not a person.
210 APPLICA TIONS ANI) DISCUSSION
Fortunately, all this is compatible with the theory under dis-

cussion.
I I.. Visual Objects
In addition to fiction, the visual arts provide another source of

nonexistent objects. One of the clearest cases is that of realistic
illusionistic painting. The unicorn in the picture (if native to the
picture) is a unicorn, and has the other nuclear properties attrib-
uted to it there. (For an attempt to explicate what is true ac-
cording to a picture, see HowelI [LSPR].) The native/immigrant
distinction appears here also. The main figures in portraits are
always immigrants, no matter how untrue to life, and immigrants
appear in other paintings as well. For example, in Rubens's
painting The Judgment of Paris, Paris is an immigrant from the
Greek legend, but the women in Chardin's Back from the Market
is native to that painting. Paintings, like dreams, operate by
gestalt; except for their titles, their content is not propositionally
presented.. Butunlike dreams, a knowledge of the tradition within
which they are painted is relevant to their interpretation; this
tradition influences the question of immigrancy, and also more
mundane issues, such as the way in which three-dimensional
volumes are represented two-dimensionally.
Comic strips with accompanying text (e.. g., Prince Valiant)
form a medium midway between pure fiction and painting; their
content is conveyed by an interaction between prepositional and
gestalt information.
It is traditional to make a distinction betweenfarm and con-
tent that applies to most of visual art. It is mainly the content of a
work of art which provides nonexistent objects, for it is the con-
tent which determines what is true in the work (in the sense of
'true .in' used above; this probably has very Iittle to do with
"artistic truth"). In those areas of art, such as modern abstract
painting, in which there is an attempt to depict form without
separate content, it is doubtful that nonexistent objects are rel-
evant.
Realistic sculpture provides another source of nonexistent
objects. Statues provide both native and immigrant objects in
FICTIOJVAL OBJECTS, DREAM OBJECTS, AND 0 THERS 211
much the same way that paintings do. Rodin's The Thinker de-
fines a native object, whereas Michelangelo's Bacchus yields an
immigrant one.
Objects native to works of visual arts are always incomplete,
and, as with Escher's prints, occasionally impossible as well.
Plays are peculiar in that we have both scripts (which are
often read as stories in their own right) and performances. The
information in a script is presented propositionally; that in a per-
formance (except for program notes) is presented by gestalt. It is
probably impossible for any performance to present exactly the
same information as is contained in a script, and that suggests
that we will have to distinguish the characters of a performance
from the characters of the script. This is initially surprising, but
in fact we do seem to talk in this way. For example, we compare
"Barrymore's Hamlet" with "Burton's Hamlet" as if we are
comparing two different things; they can't both be Hamlet, then,
and it is plausible to suggest that neither of them is. (This is not
to deny that they both play Hamlet, where this Hamlet is the one
from the script; this is like two different actors both playing
Lyndon Johnson, in McBird).
8
Traditional Issues from the Present

Perspective
The point of this chapter is to ask: what would certain traditional

issues look like if the present theory of objects were taken to be
correct? I will not be concerned here to defend the theory of
objects.
I" Ontological Arguments; Why Should Anyone Care Whether God

Exists?
The present system is so rich in objects that one might suspect

that God would be found among them. That depends on what is
meant by 'God'.
(i) If 'God' means, say, 'the main deity described in the
Bible', and if this is treated as an object native to the Bible
story, then there is such a god as God, though he will
definitely not exist (since he will be incomplete).
(ii) If 'God' means, say, 'the existing deity of the Bible',
then there mayor may not be such an object, and the
quoted definite description mayor may not refer..
In short, the system provides for lots of gods, with lots of inter-
esting characteristics, some worthy of veneration and some not.
But nothing in the system alone can be used a priori to show that
any deity exists. This is a relatively trivial logical point, for the
system allows of models in which nothing exists. It' won't help to
add words like 'perfect' or 'omnipotent' to a description in
hopes that some ingenious argument will yield an existent god.
212
TRADITIOJVAL ISSUES 213
For either the description will be purely nuclear in character,

and we will not be able to show that the objects which satisfy it
exist, or it will be partially extranuclear, and we will not be able
to show that any object satisfies it.
Descartes had a version of the ontological argument which
illustrates this point:
Whenever it pleases me to imagine a first and supreme
being, and as it were to bring down an idea of him from my
mind's treasury, it is necessary that I should attribute to
him all perfections, even though I neither then enumerate
them all nor attend to each one: and this necessity is plainly
sufficient that afterwards, when I notice that existence is a
perfection, I "rightly conclude that a first and supreme being
really exists. (Barnes [OA] p. 16)
Clearly, Descartes assumes that there is an object which satisfies
'x is perfect'. If perfection does not entail existence, then this
assumption might be justified, but then he cannot conclude with
necessity that existence is a perfection, and that such a being
exists; if perfection does entail existence, then he is not justified
in assuming that there is a perfect being to beginwith. The most
he can show is that there is no nonexistent perfect being, but
maybe that is because there is no perfect being at all.
Anselm's argument has always seemed more persuasive than
Descartes's because he first argues, in a seductive manner, that
there is a perfect being, and then argues that it exists. Most
reconstructions of Anselm's argument fail to do justice to the
persuasiveness of his presentation. Here is Barnes's translation:
Even the fool is bound to agree that there is at least in the
understanding something .than which nothing greater can
be imagined, because when he hears this he understands it,
and whatever is understood is in the understanding. And
certainly that than which a greater cannot be imagined
cannot be in the understanding alone. For if it is at least in
the understanding alone, it can be imagined to be in reality
too, which is greater. Therefore if that than which a greater
cannot be imagined is in the understanding alone, that very
thing than which a greater cannot be imagined is something
than which a greater can be imagined. But certainly this

cannot be. There exists, therefore, beyond doubt something
than which a greater cannot be imagined, both in the
understanding and in reality.. (Barnes [OA] p. 3)
I think that the persuasiveness of this argument hinges on a
double ambiguity regarding 'imagines'. This is an intentional
verb which takes two kinds of direct objects: prepositional ones,
as in 'Jones can imagine that we will soon land on Mars', or
'Jones can imagine us landing on Mars soon', and ordinary
denoting phrases, as in 'Jones can imagine a purple unicorn' .
And whichever it takes as object, it may do so in either a de re or
a de dicto sense (recall the discussion in chapter 2, section 4).1
I. Barnes [OA] pp. 88-89 actually offers a formalization of Anselm's

argument which encapsulates these ambiguities by using the same notation
(though with two different definitions) to symbolize both 'x imagines that P'
and "x imagines something F'J (with no notational distinction being made be...
tween the de dicta and de re readings). This turns out not to be important,
however, since other of Anselm's rules and assumptions do all the real work
in the proof. In fact, although Barnes does show how to derive 'God exists'
from five of Anselm's assumptions, it is possible using his Anselmian rules to
derive anything at all from no assumptions. Here is a sketch of the argument.
Firs t we gi ve two definitions:
(DJ) Ax =df Bx v -s B«.
(D2) a =df (1X) ",Ax.
(These definitions are patterned after ones given by Barnes.) Now let C be
any sentence at all not containing the name a. The proof, using Barnes's
notation and rules, is:
(1) "JC assumption
J1 (2) ;*yC =:> Ba V -e Ba tautology
(3) Ba v < Ba (I), (2) MPP
(4) Aa (3), Dx
(5) (x)Ax (4), DI
(6) Aa (5), UE
(7) A[ (Jx) '" Ax] (6), D2
A (8) ~ ~C (1), (7), RAAAns
A (9) c (8), DN
The work is all done by treating definite descriptions as terms subject to quanti...
fication, as in (6), and simultaneously using rule RAAAns ("reductio-ad-
absurdum-Anselm") in step (8); this latter rule essentially assumes the falsity
TRADITIONAL ISSUES 215
Anselm's argument begins by establishing that the fool "imagines

that than which nothing greater can be imagined" in its de dicta
sense (for the justification is merely that the fool understands the
words)" But then he begins referring back to the alleged referent
of the denoting phrase by means of singular pronouns, as if it had
been established that there is such an object imagined by the
fool (de re), a natural and reasonably subtle transition-but a
question-begging one. Then later, he implicitly switches to the
usage of 'imagines' in which it takes a propositional object;
he judges that such an object's being in reality is greater than
such an object's not being in reality.. But in the crucial phrase
'that than which nothing greater can be imagined' it is the thing
that is said to be greater, not some proposition or state of affairs
in which it figures. This makes the argument, as literally stated,
so patently invalid that virtually everyone who comments on
it rephrases the argument in some way, either by turning the
first use of 'imagine' into one which takes a that clause as its
object (as in Adams [LSAA]), or by turning the second use into
one which takes a denoting phrase as object (as in Barnes [OA]).
Anselm's statement of the argument gives no clear guidelines for
how either of these is to be done, and I suspect that one of the
reasons the argument has never been fully disposed of is the
thoroughgoing unclarity about just what the argument is (plus)
of course, the importance of its subject matter).
But however the argument is construed, I suspect that it
will either rely on fallacious reasoning, as in the first transition
from a de dicta to a de re reading of 'imagines', or it will employ
assumptions about the relationships between these readings
which are not viable. The only support it receives from the theory
of objects is the vindication of the coherence of reasoning about
a thing without prejudice as to whether or not that thing exists.
It has often been thought that the crucial fallacy in onto-
of any sen tence of the form D [(')x) Dx]. (In Barnes's formalization of Anselm's
'V
argument, these rules are employed in his steps (8) and (27).) The theory
of objects as I have formulated it would sanction rule RAAAns, but would not
allow the use of universal specification (Barnes's rule DE) for arbitrary definite
descriptions since the theory assumes that some definite descriptions do not
refer to anything.
216 APPLICA TIONS AND DISCUSSIOJV
logical arguments is treating existence as a property. There are

(at least) three main versions of this thesis.
(i) According to Frege, existence isn't a property of indi-
viduals, but rather a property of concepts ([FA] sec. 53); ex-
istence is not properly represented in language by a predicate
('exists') but rather by a quantifier ('there exists'). However,
there certainly is a property of individuals represented by the
complex predicate 'being something such that there exists such a
thing' ('[Ax(3y)(x = y)]'), and the arguments can all be re-
phrased (though perhaps clumsily) in terms of this predicate.
(ii) According to others (e.g., Kant), existence isn't a prop-
erty because it is trivial (it adds nothing) to say of something
that it exists. This view would appear to identify 'there is some-
thing such that . . . ' with 'there exists something such that
. . .'. If this identification were correct, and if we could es-
tablish that there is a god, then we would automatically have
established that a god exists. This would actually vindicate part
of the ontological argument. Unfortunately, no historical version
of the argument successfully establishes that there is a god. The
theory of objects does assert this, but it can't be combined with
the view in question to show that a god exists, for it rejects the
triviality of existence statements,
(iii) There seems to be an undercurrent in a lot of discussion
that existence, although a property, is not a property on a par
with others-it can't be treated like ordinary properties, especially
with regard to "defining" objects. This may be part of what lies
behind the issue of whether or not existence can ever be a part of
the "essence" of a thing. Sometimes existence is called a "tran-
scendental" property. This is, in fact, the view taken in the theory
of objects; existence is an extranuclear property, not an "ordi-
nary" nuclear one. We can "define" objects using nuclear
predicates-that is, using nuclear predicates we can specify an
object and be sure that there is such an object; there is no a
priori guarantee that we can do this using extranuclear predi-
cates, such as 'exists'.
Who cares? Why is it important whether or not God exists?
There are two questions to be distinguished here: first, why is it
important to people that there be a god, and second, if there is

one, why is it important that it exist? Presumably the answer to
the first question is that religious activity typically involves
worship and veneration, and that such activities are inappro-
priate without something to worship and venerate. Of course,
the phenomenological side of such activities may be present even
in the absence of an object, but such isolated mental states are
presumably just as undesirable to their agents as are beliefs that
are false.
But supposing there to be a god-for example, a fictional
one determined by a biblical myth-why would it be devastating
for it not to exist? In one sense the answer is clear: it's something
that people do care about. You can't always provide reasons for
attitudes, and this may be such a case. On the other hand, it
ought to be equally possible for someone to simply not care
whether God exists. Many people would find it inappropriate
to worship or venerate such a being, but I think that this is an
attitude of theirs which not everyone needs to share. As to its
feasibility, well, there are people who venerate Jesus without
believing him to be divine, and prima facie, divinity seems to be
just as important in a religious object as existence.
2. Epistemology: Might I Be a Nonexistent Object?
In the Meditations Descartes opposed reason to skepticism. Reason

was declared the winner, but later generations have almost
unanimously agreed that this was mainly due to the home-court
advantage, and that an unbiased referee would have declared
skepticism the victor. Reason should have lost in the first round,
when Descartes was overly impressed with the catchy "cogito",
The evidence was supposed to be, roughly, the phenomenologi-
cal side of doubt occurring. Somehow this got analyzed into an
activity with an agent, and the conclusion was then drawn that
the agent exists. It wasn't even fair to analyze the doubt into an
activity plus an agent.s But supposing that there is an agent, how
2. Cf. Nietzsche [BG&EJ pt. I sec. 6. I am not attempting to discuss the
cogito exactly as Descartes articulates it. As I see it, the relevant structure should
have been:
do we get the further conclusion that it exists? Descartes can

hardly presuppose the equivalence of being something with being
something existent, for the ground rules were established in col-
laboration with skepticism, and the skeptic doesn't need a theory
so elaborate as the theory of objects to demand a justification
for the step from being to existing (especially when the data are
so "mental" to begin with, and the conclusion is that the agent
is not just an object of thought),
As with the standard Cartesian approach ("How do I
know that I'm not just dreaming?"), there are homier applica-
tions of this issue. Even if I don't doubt that I am something,
how can I know, say, that I am not (merely) a (native) object
of a very detailed and cleverly designed story? What can I learn
about myself which would ensure my reality? I am human,
male, brunette, etc., but none of that helps. I see people, talk
to them, etc., but so did Sherlock Holmes.f
Instead of fearing that I am dreaming, and everything
around me is unreal, why shouldn't I be equally afraid that
someone else is dreaming, and I am unreal? This question was
faced by Alice in Through the Looking-Glass, chap. 4, and as
Tweedledum and Tweedledee mercilessly pointed out, her
efforts to gather evidence in favor of her reality met with no
success whatsoever. It seems to me that this is a philosophical
(I) Doubt is occurring.

(2) I (the doubter) am thinking.
(3) I am.
(4) I exist.
The criticism is for presuming (2) instead of (I), not for explicitly inferring (2)
from (1). My next criticism (below in the text) is for ignoring the difference
between (3) and (4), not for explicitly announcing that (4) follows from (3)·
(Note that (2) should not be confused with Descartes's later argument that
thinking is part of his essence.)
3. Hintikka [CES] suggests in this connection that the inference from
'Hamlet thinks' to 'Hamlet exists' is not a valid one. But his point is that the
former might be true even though the name 'Hamlet' fails to designate any-
thing. My point here is the different one, made by Meinong in (TO], that
there might be a thing that thinks without that thing existing. See Kenny
[DOA] for further discussion of this point within the setting of Descartes's
ontological argument.
problem that deserves to be treated seriously on a par with issues

like the reality of the external world and the existence of other
minds. (1 don't know how to solve it.) .
3- Leibniz's Monads and Possible Worlds
Certain aspects of the theory of objects look somewhat Leib-

nizian.. For example, all relational statements are analyzable into
conjunctions of property statements: '» bears R toy' is shorthand
for 'x has [Ry] and y has 1xR]'. These properties, of course,
"encode" the relations out of which they are made. This encod-
ing can be exploited so that certain objects "mirror" the pos-
sible worlds in which those objects "appear"; that is, whole
possible worlds can be encoded into the properties of single ob-
jects, if those objects are of certain special sorts, One sort I will
call monads; these include all existing objects, and there will be
others as well. Here is how it goes.
Suppose that we are working with a fully interpreted version
of (f), as developed in chapter 5. Sentences of (f) will have truth
values relative to various possible worlds. Call the actual world
'wo'. Then, following Castaneda [YSW] sec. I I, we can define:
x is a monad =df in Wo, x is both complete and possible.
Then if x is a monad, we define:
x appears in W =df there is an object which exists in wand
which has, in w, exactly those nuclear properties which x
has in WO.
Notice that to say that x appears in to does not mean that x itself
exists in tu, but rather that some "surrogate" for x exists there.
This surrogate might be x itself; this will happen when w is WO)
and might happen in other cases as well. But, in general, the
surrogate of an object in a different world may be a different
object.
Now various metatheorems may be proved about the se-
mantical structure of the modalized version of (f) that was deveI-
Portions of this section resemble material that appeared originally in
Parsons [N&EPM&L].
220 APPLICA TIOJVS AND DISCUSSION
oped in chapter 5, metatheorems which reveal some of its Leib-

nizian aspects. The key fact, on which all the rest are based, is:
Metatheorem 0: If object y exists in world tu, and if ifJ is a wff
that does not contain x free, then if> is true in w (with respect
to an assignment g of values to the variables) if and only ify
has, in to, the property that W[AX~] stands for in to (with
respect to assignment g).
This metatheorem tells us that facts about a world (of the
sort that are expressible in (9) are reflected in the nuclear prop-
erties that the objects that exist in that world have in that world.
The metatheorem depends crucially on the watering-down
principle (axiom WD), which might appropriately be renamed
the "principle of preestablished harmony"-objects that exist in
a world are, insofar as their nuclear properties are concerned, in
harmony with each other and the rest of the world. (The "pre-
established" part is that the system was designed so as to satisfy
this principle.) A proof of the metatheorern goes somewhat as
follows :4
4. This is a very informal sketch of a proof technique. It contains a certain

amount of use..mention confusion (e.g., it talks about an object's having a
predicate instead of a property) and it ignores the complications of assignments
to the variables. Other proof sketches below share some of this informality.
The official proofs all utilize the technical apparatus developed in chapter 5,
section 5. Since metatheorem 0 is so important for what follows I'll give a
somewhat more accurate description of its proof here.
Suppose thaty exists in wand that ~ does not contain x free. Recall that in
chapter 5, section 5, we defined a function Ag such that lOAg(a)' denotes what-
ever thing a stands for in the model with respect to the assignment J; to the
variables of (f). Wheny exists in to, :Y has the property r in w' is formulated as
yE extw(r)'. And '<jJ is true in w with respect to g' is formulated as 'ext",(Ag(fj:»)
= T'. A formal statement of the metatheorem then is: for any objecty, and
any wff <p not containing x free, and any assignment g, yE extw(Ag(w[hlftJ)}
if and only if extw(Ag(ifJ)) = T. The proof then goes as follows:
yE extw(Ag(w[AxcP])
iff yE extw(W(Ag([AxqSJ)) by condition (16.18)
iff yE extw(Ag«(hep))) by condition (14.18)
iff yE extw(h), where h is the function that maps any given object z to
Agz(ifJ)J where g% is just like g except that gz assigns z to x,
by condition (r6.19)
TRADITIOJYAL ISSUES 221
Assume thaty exists in w. Theny has w[).xpJ if and only if

y has [AxshJ (this is because y exists, and so by WD, y has the
watered-down version of a property if and ·only if it has the
property). But x is not free in <jJ, and so (by axiom AB(E) **)
objects have (Ax'q.S] if and only if <p is true. Putting these facts
together we have:
y has w[AXtPJ in w if and only if ifJ is true in to, which is what
we want.
We can now establish:
Metatheorem I: If x is a monad, x appears in exactly one
world.
Sketch of proof: First we show that x appears in at least one world.
Let X be the set of nuclear properties that x has in WO. Since x is
a monad, x is possible, and so there is an object z and world w
such that z exists in wand z has (in w) all the properties in x.
Since x is complete, e: cannot have any nuclear properties that
are not already in X without being impossible (in w), and so not
existing in w. Therefore, z exists in wand has (in w) exactly the
nuclear properties that x has in Wo, and so, by definition, x ap-
pears in tu,
Second, we show that x appears in at most one world. As-
sume not; that is, assume that there are objects y and e: and
worlds Wl and Wz such thaty exists in Wl and has, in Wh exactly
the nuclear properties in X, and z exists in Wz and has, in wz,
exactly the nuclear properties in X. Since wl :f= ~, there must be
a difference between Wl and W2 that affects the truth value of
some wff if> (which does not contain x free) with respect to some
assignment to the variables (see below). Suppose that ifJ is true
in Wl and false in W2 with respect to g. Then, by metatheorem 0,
y has, in Wl' the property denoted by W[AXgS] with respect to g,
iff extw(h(y)) = T by condition (14-3)

iff extw(Agy(~» = T by above characterization of h
iff extw(Ag (fj))) = T since <p does not contain x free
(Strictly, the last step involves the customary but complicated proof that the
truth value of a formula relative to an assignment to the variables is not affected
when the assignment is changed for a variable that is not free in the formula.)
and z lacks that property in W2, contrary to the assumption that

bothy and e: have (in those respective worlds) exactly the nuclear
properties in X-
The dirty work in spelling out this proof lies in showing that
if Wl ¥= W2, then there is some wff rp, not containing x free, and
some assignment g, such that cP is true in Wl with respect to g
and false in W2 with respect to g. You find such a ifJ and g as
follows. First, as worlds are constituted in chapter 5, section 5,
they cannot differ at all unless they differ with respect to:
(i) which objects exist in them, or
(ii) which objects have which nuclear properties in them, or
(iii) which extensions are possessed by which extranuclear
properties, propositions, etc.
In case (i), if u exists in Wl but not in W2, pick <p to be E!u, and
let g be any assignment that assigns u to u. In case (ii), if u has p
in Wl but not in wz, let ifJ be ql ul and let g assign p to ql and u to u,
Case (iii) has several subcases which are dealt with similarly.
It is now natural to define compossibility as follows:
x is compossible with y = df there is a
world in which x and y
both appear.
Because of metatheorem I it is now easy to prove:
Metatheorem 2: Compossibility is an equivalence relation on
the class of monads.
(See Mates [LPW].) Furthermore, compossibility is the "right')
equivalence relation. To show this we first define, for any monad
x:
Ixl =df the class of monads that are compossible with x,
and:
PW(lxl) =df the world in which any member of Ixl ap-
pears.
Then we have:
Metatheorem 3: PW is a one-one function from the set of
equivalence classes of monads (with respect to compossibil-
ity) onto the set of "nonernpty" worlds.
CA "nonempty" world is any world in which at least one object

exists.) So the monads fall into realms (equivalence classes of com-
possible monads), and the realms correspond one-one with the
nonempty possible worlds. All the monads in the same realm
appear in the same world-the world to which the realm cor-
responds-and each of the monads in the realm "represents" an
object that exists in that world (namely, the object which, in that
world, has exactly the nuclear properties that are had by the
representing monad in the actual world). And each such monad
mirrors the corresponding world, in the sense that every fact true
in that world is encoded in the nuclear properties possessed by the
monad in the actual world; in other words, we have:
Metatheorem 4: If y is a monad, then for any wff if> not con-
taining x free, and for any assignment g, pis true in the world
in which y appears if and only if y has W[AX~] in the actual
world with respect to g.
(This metatheorem speaks directly of wffs, but it can be applied
to "facts" as follows: If y is a monad, then, for any proposition
5, s is true in the world in which y appears if and only ify possesses
a certain uniquely determined nuclear property p in WOw The
nuclear property p in question is the property denoted by the
complex predicate w[;(xs] in Wo with respect to any assignment
g which assigns the proposition s to the propositional variable s.
In this application of metatheorem 4, the wff rp is chosen to be
the single variable s.)
Under certain circumstances, metatheorem 4 verifies a
certain "reduction" of the modal operator 0 to the modal predi-
cate 'is possible') which is a predicate of objects, not of sentences
or propositions.. The special circumstances are that every world
be nonempty.. In that case, the realms of monads correspond one-
one with all the possible worlds, and the following is true (with
respect to any assignment g) for any wff <p not containing x or
y free:
o <p == (3y) (y is possible & y is complete & w[Axq)]y).
(Recall that 'is a monad' is defined as 'is both complete and pos-
sible'.) This is not a reduction of the modal to the nonmodal, of
course, and in my articulation of the theory it is arguably not a
"reduction" at all, since the modal predicate 'is possible' was

defined in terms of the modal operator. The point is rather that,
if we had taken the predicate as primitive instead of the operator,
we could have reversed the order of definition-assuming that no
worlds are empty.
Actually, in one version of the theory we can effect a similar
reduction without making the assumption that all worlds are
nonempty. This is the version in which we assume that we are
working with a modal structure in which our set variables range
over all sets of nuclear properties.f Then we can use pure reflectors
to get a reduction of O. Let x be any object, and define:
x is a pure reflector = df for some world w, the following holds:
for any proposition s, s is true in w if and only if x has (in
wo) the nuclear property denoted by w[AXS] in Wo with re..
spect to any assignment g that assigns s to s.
Then the following holds for any wff <p which does not contain
x or y free (with respect to any assignment g) :
o</J == (3y)(y is a pure reflector & w[AxqSJy).
4. Essentialism
I have described (f) as quite neutral with regard to de re modali-

ties, and this means that it is also quite neutral with respect to
essentialism. It is clear how it is consistent with essentialism, even
of the strongest kind. For example, we can suppose that every
object has all its nuclear properties essentially; just take a modal
structure in which the extension of r is the same in every world,
for every nuclear relation r. This does not rule out a diversity of
possible worlds, for objects could still differ extranuclearly from
world to world; for example, different objects might exist in dif-
ferent worlds. (In such a model a monad would appear in a world
if and only if it existed in that world.) For the strongest kind of
essentialism, in which every object has every property essentially,
5. This is discussed briefly in chapter 5, section 5. Making this an apriori

condition on all modal structures would make the theory nonaxiomatizable;
see a discussion of a similar point in chapter 4, note 8, and in Henkin [CTY].
if at all, including the extranuclear ones, just take a modal struc-

ture in which there is only the actual world.
How consistent is the system with antiessentialism? Well,
clearly, any object will have certain properties essentially, say,
the property ofself-identity. This isn't ordinarily considered to be
a particularly bad kind of essentialism. But objects will also have
certain other extranuclear properties essentially, properties that
not all objects share. For example, in any modal structure the
following will be true for any object z :
(a) (3x)(3y)(Ox = z & ,...., DY = z).
Therefore, the following will also be true:
(b) (3x)(3y) (0 [AU(U. = z)]x & '" o [Au(U = z)]y),
from which, by existential generalization we get:
(c) (3P) (3x) (3y) (DPx & ~ DPy).
That is, there is a property P which some object has as an essence,
and which some other object does not.. In case z is Socrates, this
parallels Plantinga's argument (in [NN]) that Socrateity is an
essence of Socrates, an essence which is not shared by other
objects.
Now there is something peculiar about such properties. We
are inclined to view them as concocted. There seems to be a natu-
ral sense in which Socrateity is not a property at alL Perhaps this
is captured by the fact that Socrateity is not a nuclear property.
(The same holds for properties like being blue in world w; such a
property cannot be nuclear and hold in every world of exactly
those objects which are blue in world w-unless the same objects
are blue in all worlds.) With respect to nuclear properties, is it
possible to be completely antiessentialist? That depends on what
you mean by antiessentialism. For example, it will always be the
case that for any object x there is some nuclear property p which
x has in every world in which x exists. 6 If x is Socrates, the watered-
down version of Socrateity will be an example. And sometimes
'essentialism' is defined in these terms; it has to do with an ob-
ject's having a property in every world in which that object .ex-
6. I am indebted here to Thomas Ryckman.
226 APPLICA TIOJYS AJVD DISCUSSION
ists, But the clause 'in every world in which x exists' has a dif-
ferent import in the (orthodox) theories within which it is usually
invoked than it has here. For in those theories, existing in a world
is usually tantamount to falling within the range of the quanti..
fiers in that world, and in the present theory those conditions are
not equivalent. In short, there is another version of essentialism:
must there always be at least one object that has at least one
nuclear property necessarily? The answer is no, but this answer
may not be of any great significance (see below).
Let me call a modal structure an antiessentialist structure if,
in that structure, for any object x and any nuclear property p,
there is a world in which x does not have ft. Then it is relatively
easy to describe antiessentialist structures in which no object ex-
ists in any world." This may be regarded as not very interesting.
The difficulty with showing that there are more natural anties-
sentialist structures-ones in which lots of worlds (say, t\VO or
more) contain lots of existing objects (say, two or more) is that
such structures are so big and complicated that it is difficult to
verify that they satisfy the conditions for being modal structures;
this is due mainly to verifying that the watering-down and plug-
ging-up conditions hold. I haven't been able to show that there
are any antiessentialist structures of this sort, though my con-
fidence has been strengthened by my inability to show that there
are no such structures.
7. Such a structure may be constructed as follows: Let OB have the

same cardinality as &,(N 1 ) , that is, as the power set of N 1, and let X = &(N1 ) .
Let the set of possible worlds W be paired one-one with the set of all one-
one functions from OB on to f!lJ (JV1) If W E W, let J; be the function with
A
which w is paired, and let EXw = A. (For ext w , see below.) Let P = the
set of all functions from W into {T, F} . If wE Wand if SE P, define extw(s)
to be s(w). These conditions then force certain other conditions, and leave
certain others open. For example, if r s E n+ 1 with n ~ 1, then PLUG(r, x,
k) must be the unique r' E En such that for any Xl' • • . , X n € OB and WE W,
r'«x 1 , • • • , Xn»(W) = r«xl' ... , Xk' X, Xk+l' • . " X n ») (w).. It is
left open which atomic predicates stand for which properties and relations..
Since EXw = 11, many conditions on modal structures are vacuously satisfied;
this is what makes these structures easy to construct. It is easy to show that any
such structure is antiessentialist: let x E OB and pE JYl' and let w be the world
paired with the function that maps x to A. Then, in that world, x has no nu-
clear properties at all, and so, in particular, x does not have p.
5. Platonic Forms and Complex Properties
One way to mirror Plato's early theory of forms is to take the

forms to be the unitary objects (cc' Findlay [MTO·V] pp. 59,
164-66, Chisholm [HO]) and Parsons [PMS]); that is:
x is a form =dfX is an object with exactly one nuclear
property.
If this is done, then one problem in Plato scholarship is immedi-
ately solved: how can the form the-bed-itselfbe a bed in the same
sense in which beds existing in the physical realm are beds? In
the system f!) this is automatic.
On this reconstruction of Plato, the forms have a certain
kind of purity or perfection: the-bed-itself is purely a bed in that
it is· not compromised by having any other (nuclear) .properties
than bedness. (Plato's early theory did allow certain forms to
participate in certain forms, but he had a tendency in such cases
to identify the forms in these cases; e.g., wisdom is virtue.)
This is not a good representation of Plato's later theory, at
least in the Sophist. Here he explicitly allows some forms to par-
ticipate in others (see his discussion of the "blending" of forms at
25 Id-2s6e). This pretty clearly results from an extension of
formhood to paradigm instances of extranuclear properties; the
theory at this point becomes vaguer, and it's not clear how to
reconstruct it .
Are there "logically complex" forms or properties? For ex-
ample, are there negative properties? Throughout the ages, peo-
ple have vacillated between positive and negative answers to this
question, with yes being the current favorite. The system (9 ex-
plains this vacillation as follows. First, for any ordinary (nuclear)
property p, there is not an ordinary (nuclear) property q which is
had by all and only those objects which do not have p. So, in the
most straightforward sense, ordinary properties do not have
ordinary properties as "full-blooded" negations. But this is
disguised by two other facts. First, there is a nonordinary (extra-
nuclear) property which is had by all and only those objects
which do not have p. And second, there is an ordinary property q
which is had by all and only those real objects which do not have
228 APPLICATIONS AND" DISCUSSION
p. The nonrecognition of nonexistent objects together with the

nonrecognition of the nuclear/extranuclear distinction make it
appear, contrary to many intuitions, that ordinary properties
have full-blooded negations of the same kind as the originals.
Similar remarks apply also to conjunctive properties and the like'.
6. Theoretical Entities in Science
In this section I will discuss in a very tentative and speculative

way some applications of the theory of objects to theoretical enti-
ties in science. There does not seem to be one simple central issue
here; rather, the possible applications fragment along lines
that are somewhat familiar in the philosophy of science. I'll
discuss four sorts of application under three general headings; in
each case I'll center the discussion on a single example. The topics
and examples are:
I. Observable entities incorrectly postulated: Vulcan

2. "U nobservable" entities:
a. Correctly postulated: neutrinos
b. Incorrectly postulated: the ether
3. Ideal entities: point masses
Observable Entities Incorrectly Postulated

'Vulcan' was the name assigned, in advance, by the astronomer
Leverrier, to the planet that was supposed to lie inside the
orbit of Mercury. The gravitational influence of Vulcan was
supposed to account for some discrepancies between the observed
orbit of Mercury and its predicted orbit. It was a neat idea
(Neptune had been discovered by this means), but it didn't work;
Vulcan doesn't exist, and the explanation for the discrepancies
lies elsewhere (with relativity theory).
Does- 'Vulcan' then refer to a nonexistent planet, or does it
fail to refer to anything at all? A good case can be made for both
views. I'm inclined to think that the word is ambiguous, and that
each view is a correct description of one of its uses. When the
name was originally used, it was intended to refer to a real entity;
this was a mistake.f and, so used, it fails to refer. But now the
word is often used to refer to a character in 'a scientific legend.
I'll discuss the legend view first.
The Vulcan hypothesis has all the earmarks of a scientific
story. It has a setting: late nineteenth-century science. The
story line sanctions certain definite descriptions, some of which
do and some of which do not pick out characters. The description
'the planet that is chiefly responsible for the perturbations in the
orbit of Mercury) picks out a character; 'the chemical composi-
tion of Vulcan's crust' does not. Most of the characters of the
story are immigrants: the Sun, Mercury, the Earth, the other
planets. Vulcan itself is a native to the story. It has as nuclear
properties all such properties that are clearly predictable within
the science of the day from its being the planet that is chiefly
responsible for the perturbations in Mercury's orbit. Although
the story started as an intended description of existing things, it
is now retold under no such misapprehensions, often for amuse-
ment. And Vulcan is referred to in other contexts, where a mys-
terious astronomical figure is called for. This has all of the ear-
marks of a legend, with Vulcan its chief character. 9 And 'Vulcan'
refers to this character.
On the other hand, the scientists who searched for Vulean
used the name as we use the name 'Mercury', to refer to an exist-
ing object; such a use incorporates a "mistake» of the sort dis..
cussedin chapter 5, section 5, and so used the name fails to refer.
I base this judgment solely, however, on a linguistic intuition
of mine. Someone could maintain, for example, that 'Mercury'
and 'Vulcan' are' both used with "disjunctive reference condi-
tions"; they refer to real objects if there are any real objects
which fit the theories that we have about them, and, otherwise,
they refer, by default, to native characters of associated legends"
This seems wrong to me, but I feel some uncertainty on the issue)
and I have no argument to back up my view.
8. I don't mean here that the scientists themselves necessarily had any
false beliefs, for they might have been skeptical about the predictions they were
testing. It's rather that the word was used to refer to the existing object,.if any,
which would fulfill the predictions..
9. I am indebted here to Carl Coniglio.
A more radical idea would be to identify all objects with

characters in stories. Those characters native to stories that turn
out to be true (e.g., Mercury, you and I) would all end up being
characters of the big story, if there is one, into which the smaller
ones eventually evolve. If the big story "tells all," then such
objects would even turn out to be complete, and presumably
would exist; otherwise, perhaps no objects are complete, and
existence is a metaphysical ideal that many aspire to but none
reach..
"Unobservable:1:1 Entities
If Vulcan existed, we could fly up and look at it, perhaps land on
it, sink our teeth into it. Not so neutrinos or the ether; these are
classic "unobservable' entities. Maybe they aren't really un-
observable (maybe tests for their existence and properties count
as observations of them) ; I won't need to take sides on this issue..
First, neutrinos. I'll assume for purposes of discussion that
contemporary science is right in thinking that there are such
things. Now some people would hold that, although there are
neutrinos, they don't exist-because they aren't observable, or
aren't directly observable, or whatever. If this is correct, then
they constitute examples of nonexistent objects. But this view, in
isolation, runs the risk of being metaphysically boring. It's as if
we decided that automobiles don't exist, but that they have the
same properties they would have if they did exist. This idea
skews the extension of 'exists' while leaving everything else alone.
I suspect that many such views reduce to questions regarding
the usage of 'exists'.
An interesting variant is that neutrinos do not exist because
they are incomplete. Your typical neutrino neither attends
church regularly nor fails to do so, neither has hair nor is bald.'
This view, of course, depends rather heavily on what is meant
by 'incomplete'. This is not an easy issue, but I think we can make
a general judgment about neutrinos without settling it .
Suppose, first, that the aforementioned facts do show
neutrinos to be incomplete. Then ordinary objects will be in-
complete too, for they lack properties that are appropriate to
TRA.DITIONAL ISSUES 231
neutrinos. For "example, I do not have a spin of + t, nor do I

fail to have such a spin. But if this is the right way to view things,
if an object gets to be incomplete just because certain properties
cannot be appropriately assigned to things of its kind, then we
are looking at a situation in which the present version of the
theory of objects should not be applied. This issue was discussed
in chapter I, note 2: some people think that objects fall into
categories, with all objects being incomplete in a general sense of
'incomplete). For, given any object., there will always be some
property such that neither it nor its negation may meaningfully
be ascribed to objects in that category. But if so, this general
kind of incompleteness does not prevent an object from existing..
(We might then want to go on and discuss a relativized notion of
completeness: an object would be complete in this sense if, for
any property appropriate to things of its category, it either had
it or its negation. Being incomplete in this relativized sense would
perhaps suffice to guarantee nonexistence.) If this category view
is correct, the theory under discussion is an idealization, and its
application to the issue of the nonexistence of neutrinos (because
of incompleteness) would be an application to a situation that
was idealized away from. So it would then be inappropriate to
conclude that neutrinos do not exist for this reason. On the other
hand, the more complicated category view does not offer any
motivation to think that neutrinos do not exist, for that view
does not connect the general sense of incompleteness with non-
existence.
Perhaps there is no reason to think neutrinos incomplete
in the first place. Perhaps neutrinos do have the (nuclear) nega-
tion of going-to..c hurch-regularly, and this is a different property
than failing-to-go-to-church-regularly. (I think that this is the
right view.) If so, there is no particular reason to think that
they are incomplete, and so no reason to conclude, on grounds of
incompleteness, that they do not exist. On either account, neu-
trinos turn out to be poor candidates for the status of nonexistent
objects.
If I am correct in the foregoing, the present theory does not
make "unobservable" theoretical entities metaphysically special,
and this leaves the ether in the same boat as Vulcan. There is a
legendary ether, which fails to exist, and there is also a usage of
the word 'ether' in which the word simply fails to.refer,
Ideal Entities
Ideal entities differ from. all the examples considered previously
in that no one thinks they exist. For example, point masses are
used in physics as a "convenience"; their nonexistence was not
discovered, but was known prior to their invention. Are there
(nonexistent) point masses then? Or is discussion involving point
masses just shorthand for hypothetical talk about "limiting
cases"? When it is said that "Two point masses rebound with
such and such trajectories," does this mean that certain non-
existent objects actually do behave in such a manner) or that the
trajectories so described are the limits of what can happen to
existing objects as their sizes shrink and their behavior becomes
less and less influenced by factors not under discussion?
The trouble with making sense of the former view (that such
talk describes the actual behavior of certain nonexistent objects)
is in getting straight on specific applications. Here is one problem
that might have to be faced. One of the prevalent uses of point
masses in physics is one in which (a) the point masses are entities
that exactly obey the 'idealized physical Iaws under discussion
(e.g., Newtonian mechanics, ignoring electromagnetic effects),
and Cb)· the point masses are "associated" with certain actual
existing objects. For example, the earth is said to move around
the sun as if each body were a point mass located at the center of
gravity of the body in question. However, the paths of the center
of gravities of the earth and the sun do not exactly obey the laws
in question, so it is not clear how to effect the "association"
between the bodies and the point masses; there is either no way
to do. it, or many "inexact" ways, with no way to choose between
them. This suggests that in such applications the movements
of the actual bodies are being idealized as well. And this increases
the suspicion that in such applications it is the hypothetical
behavior of actual objects that is being discussed, rather than the
actual behavior of nonexistent objects. But these considerations,
I think, barely scratch the surface.
9
Global Issues
In this final chapter I want to discuss two possible problems with

the theory of objects as developed in this book. These are dis-
tinguished from other problems discussed earlier in that although
they are troublesome--even potentially devastating-they are
at least very similar to problems that are not peculiar to the
present account at alL These are problems of a sort that tend to
infect practically any very general metaphysical or metameta-
physical theory.
I have (somewhat artificially) classified these in terms of
cardinality issues r "Are there too many objects?" and "Are
there too few objects?"
I. Are There Too Many Objects i
In appendix B of The Principles of Mathematics, Bertrand Russell

presents a paradox about propositions. For any set X of proposi-
tions, let g(X) be the proposition that every member of X is
true. It seems reasonable that if X -1= ~ then the proposition
that every member of X is true is a different proposition from
the proposition that every member of r is true (for example,
someone could believe the former without believing the latter).
But then if X :f:. Y, then g(X) #= geT), and so g is a one-one func-
tion which maps the power set of the set of propositions into the
set of propositions itself. But this contradicts Cantor's theorem
that no one-one function can map the power set of any set into
that set itself..
One could take this to show that there was no such thing
233
as the set of all propositions (a solution that Russell called

"harsh and highly artificial" [POM] p. 528), or one could take it
to show that g is not one-one-that we must give up the naive
assumption that if X ¥= T, then the proposition that every mem-
ber of X is true is different from the proposition that every mem-
r
ber of is true. Whatever route is taken here, one of the as-
sumptions used to generate the paradox must be rejected-but
they are all plausible l-
This paradox has nothing to do directly with nonexistent
objects. (Nor is it formulable within "(,9, although it could be so
formulated if we expanded (9 to contain variables that range over
sets of propositions.) I am recalling it here only to put in per-
spective a closely related paradox that does have to do with
nonexistent objects. Here it is. Let R be some arbitrarily chosen
two-place nuclear relation and, for any object x, let g(x) be
[Rx]. Then g(x) is a nuclear property. Now it seems reasonable
to suppose that if x i: y then [Rx] is a different property than
[Ry]. For example, someone could believe that Holmes has the
former without believing that he has the latter. But then if x ¥=
y, then g(x) #= g(y), and so g is a one-one function from the set
of objects into the set of nuclear properties. But there are too many
objects for this to bepossible. For there is also a one-one function f
which correlates objects one.. one with sets of nuclear properties
(this is the correlation function discussed informally in chapter
I and formally in chapters 4 and 5). Then the composition of
the inverse of f with g is a one-one function which maps the
power set of the set of nuclear properties into the set of nuclear
properties, in violation of Cantor's theorem.
Roughly speaking, there are two horns to the dilemma:
either maintain that g is not, after all, a one-one function, or
give up the correlation of objects with sets of nuclear properties.
In its official formulation, the theory does the former. In partic-
ular, it is a theorem of the formal theory f!) that, for any nuclear
relation R, there are some objects x and y such that x =f:. y but
[Rx] = [Ry]. This means that the theory officially rejects the
I. Note that the paradox doesn't depend on the notion of truth. We could
have g map every set X to the proposition that someone believes every member
ofX.
GLOBAL ISSUES 235
"reasonable assumption" that if x =F y, then, always, [Rx] :F

[19']" Here is a sketch of the derivation:
Let R be any two-place nuclear relation. By axiom OB]
there. is an object x which has as its nuclear properties exactly
those properties p which satisfy (3y)(p = [Ry] & ~py).
Now we ask whether x has [Rx] . First, for purposes of a
reductio argument, assume not-that is: Assume "Jx[Rx]. Then,
by the "definition" of x, "J (3y) ([Rx] = [Ry] & "'y[Rx]). That
is, (y)([Rx] = [Ry] ;:)y[Rx]) . So, putting x for y, we can derive
x[Rx], contradicting the assumption. This shows that x[Rx].
But "now, since x[Rx] , by the "definition" of x we get:
(3y)([Rx] = [Ry] & ~y[R"x]). That is, for some j, [Rx] = [Ry]
and y[Rx] . "But x[Rx], so x =I: y, and so we have a casein which
i--
x =1= y but [Rx] = [Ry], which is whatwas to be proved.
There are at least two "nonlogical' assumptions in the deriva-

tion, One is axiom OB], which says essentially that every "de-
scribable" set of nuclear properties determines a unique object.
The other is the assumption, embedded in the notation, that
plugging up a nuclear relation with an object yields a nuclear
property. I'll discuss modifying these in a moment, but, first,
let us see if the theory can be -defended in its present form.
A persuasive defense would consist in making the theorem
plausible. I don't know how to do that, but perhaps I -can under-
cut the plausibility "of its negation. First, let me appeal to the
parallel of this example to Russell's example concerning prop-
ositions, and draw the moral that these are issues in which our
normal assumptions are just not mutually compatible. We have
to give up something that we believe, and so just because some-
thing seems right to us here, we cannot place too much weight
on it. Perhaps our conviction that if x =1= y, then [Rx] # [Ry],
for any x and y, is arrived at by hasty overgeneralization from
cases with which we are familiar to cases which have a quite
different nature (as when we overgeneralize abstraction prin-
ciples in set theory to arrive at paradox-causing sets).
Is there, in fact, any reason to maintain the above principle
in its complete generality? If x :j.:. y, then why must [Rx] be
different from [Ry], for any x andy? The only argument I know
of is the one alluded to earlier: someone could believe that

something has the one without believing that it has the other.
But is this so obvious? If [Rx] = [Ry], then the proposition that
Holmes, say, has the one is the same proposition that he has the
other, and- so no one could believe the one without believing the
other. But this begins to take us around in a circle. Why do we
think that someone could believe that Holmes has [Rx] without
believing that Holmes has [Ry], when the only thing we are sure
about is that x #:- y? I think it must be something like this: we
imagine someone having x in mind, and having y in mind, and
having R in mind) and then saying to himself, "Holmes bears
R to x, but not to y," or something- like this. But what if it is not
possible to"have bothx andy in mind? What if at least one of x andy is,
in- a very strong sense, not conceivable? It is usually taken for
granted that this cannot happen: that any object is conceivable.
But discussions of this issue (which are rare) usually presuppose
that it is real, existing objects that are being discussed---often
even spatiotemporal objects.. When we depart from the simple.
and familiar cases, doubt begins to creep in . Some have suggested
that there is even a bound on our ability to conceive of large
positive integers, and it is a common view that some real num-
bers (some of the transcendental ones) cannot be "picked out"
by the mind. Maybe the objects which violate our intuitions
about property identity are all inconceivable ones?
Well, which are the objects which give rise to the trouble-
some cases? That depends on what relation R we start with.. If R
is at all normal then we can be sure at least that for all existing
objects x and y, if x :f:; y, then [Rx] :# [Ry]. In fact, if R is a
nuclear relation such that there could be an existing object z
which bears R to x without bearing R to y, and if x andy both
exist, then we can prove that [Rx] #- [Ry].2 In short, the prob-
lematic cases will involve unreal objects. But which ones?
2. This is the most relevant case to consider because if there could not be
an existing object which bears R to x without bearing R to y, then it is harder
to be sure of the intuition that [Rx] :F [Ry]. The proof is sketched as follows.
We are given: .
(I) 0 (3z) (E!z & zRx & ,ov.z;Ry).
Bu t aRh is defined notation, so (I) is short for:
GLOBAL ISSUES 237
We can get a partial answer to this question by looking at

the proof given above, Starting with a relation R, we produce an
object X4 This object is an unreal one, and it is conceivable
(provided that R is), for we can say which object x is: it's the
object which has exactly those properties that satisfy (3y) (p =
[Ry] & ~py). It's the object that has as nuclear properties ex-
actly those nuclear properties that can be produced by plugging
up· R with an object, an object which does not itself have the
resulting property. Now this is certainly a strange object. It
probably wasn't the first to come to mind when the theory was
initially sketched in chapter I. But, if there is such an object, it
would be implausible to call it inconceivable, for we possess a
succinct way to uniquely pick it out..
What, then, abouty? Here the situation is entirely different..
For the proof gives us no means at all of picking out y. The proof
merely tells us that there is some object which does the trick. 'y'
is not a name of such an object. Perhaps, then, it is consistent to
maintain that for any object that is distinct from our given x, but
such that you get [Rx] by plugging up R with it, that object is
inconceivable. If so) the reasoning sketched above for maintain-
ing that x :1= y entails [Rx] #: [Ry] would be undercut. (This does
(2) o (3z) (E!z & e: [Rx] & [zR]x & '" (z[Ry] & [zR]y».
Now axiom PLUG(N) tells us that:

(3) O(z)(E!z =:l (z[Ry] ;::) [zR]y)).
By contraposition we then have:
(4) 0 (z) (E!z => (-- [zR]y ::> IV z[Ry])) ..
And then, adding a superfluous term to the embedded antecedent:
(5) O(z)(E!z =:l (",(z[Ry] & [zR] y) ::> .-vz[Ry]».
Using (5) within (2) yields:
(6) 0 (3z) (E!z & z[Rx] & [zR]x & -z[Ry]).
Now if [Rx] were identical with [Ry], we could substitute the former for the
latter in (6), getting:
(7) O'(3z)(E!z & z[Rx] & [zoR]x & """z[&]),
which contains a contradiction inside a possibility sign, and is easily disproved.
not mean that for such an object y, [Ry] is inconceivable, for

presumably [Rx] is conceivable, and [Ry] is [Rx].]ust because a
property results from plugging up a relation with an object, and
the property is conceivable, that doesn't mean that either the
relation or the object is conceivable.)
The view being sketched is this: if x and y are distinct ob..
jects, and ifboth are real, or both are fictional characters, or both
are characters in a dream) or both are in some way "familiar,"
then [Rx] ::f:. [Ry]. And this explains our intuition to the effect
that if x =F y, then [Rx] :P [Ry]. But there are cases, lots of them)
in which x =F y and [Rx] = [Ry]. In all these cases, at least one of
x and y are inconceivable. (Perhaps a similar line can be taken
with respect to Russell's problem about propositions as well.)
There are at least two sources of doubt regarding this de-
fense, The first is a logical one: is it even consistent to suppose
that if x =f:. y and [Rx] = [Ry], then at least one of x and y is
inconceivable? The second is philosophical: if it is consistent,
is it plausible? I don't know the answer to either of these. But
the first can be given a fairly clear formulation, so I'll leave it
as an exercise.
Exercise for the reader: Is there a version of the language (f)
(from chapter 4) and an interpretation for it in which:
(I) There are lots of existing objects.
(2) Every existing object has a name in the language.
(3) Lots of properties and relations have names.
(4) For any instance of axiom AB(E) of the form: (3 Q) (x)
(Qx == ifJ), in which <p contains no free variables except
x, there is a constant predicate P such that (x) (Px == </J)
is true in the interpretation.
(5) For any sentence of the form (3x)(y)(<jJ == x = y) which
is true in the interpretation, there is a name a such
that (y) (1J == a = y) is true in the interpretation.
(6) If a and {3 are object names of (fJ and if T is a two-place
nuclear predicate constant of @, then the following is
not true in the interpretation: a :f:. ~ & [raJ = (r~]·
The idea behind the exercise is that the things denoted by con-
stant terms of the language are the conceivable ones, and the
GLOBAL ISSUES 239
others are inconceivable. The conditions (I )-(6) ensure that all

existing objects are conceivable (this maybe overkill), and any
property or relation or object which is uniquely specifiable in
terms of conceivable objects is also conceivable.
If the defense above is unpalatable.s we might want to mod-
ify the theory instead. One way to do this would be to give 'up the
assumption that plugging up a nuclear relation with any object
yields a nuclear property. This would require a fundamental
overhaul of the theory,. and I will not explore it here.. The other
way to modify the theory is to weaken axiom OB]. Recall the
remark above about the strangeness of the object x produced in
the proof.. It's doubtful that x could be a native character of a
story. or a dream or a painting. Perhaps, then, the theory has
been formulated too strongly; perhaps it yields too many objects .
The danger in restricting axiom OBJ, of course, is that we
might then have a theory that provides too few objects for its
intended applications, so we must take care not to restrict it too
much.· Here is a proposal that may be viable. Suppose that we
replace axiom OBJ:
OB]: (3x)(p)(px == tjJ)
by a restricted version, <;>BJE:
OBJE: (3x)(P)(px == tfiE),
where <jJE is any wffwhose object quantifiers are restricted by the
predicate E!; in other words, every subformula of t/JE of the form
(x)<f; is actually of the form (x) (E!x ~ X),. and every subformula
of the form (3x)cjJ.is actually of the form (3x) (E!x & X).
3. One unpleasant consequence of the proposal is that it would probably
require a reassessment of the analysis I have given of de re belief. For suppose
that a believes de re of x that b has the property got by plugging up the second
place of R with x, but that this is a case in whichy is inconceivable and .[Rx) =
[Ry). Then it seems likely that a has no de re beliefs abouty at all, But if we adopt
the method of symbolizing de re belief that was adopted in chapter 5, section
3, then we symbolize the belief that a has by aB {b[Rx]} , and the belief that a
lacks by aB {b[Ryl} .. But if [Rx] = [Ry], the latter follows from the former by
substitutivity of identicals. The obvious way to save the proposal in the face
of this problem is to require something more for de re belief say, something
along the lines of Kaplan's proposals in [QI].
Axiom OBJE is designed to do much of the work of our

earlier OB] while impeding the derivation of the theorem dis-
cussed above. It easily allows all the simple applications that
OB] allowed; for example, the proof that "the gold mountain"
is an object is unaffected; it is:
(3x)(P)(px == P = g V P = m),
which is just as much an instance ofOBJE as ofOBJ~
The viability of this restricted theory depends on positive
answers to the following two questions, a logical one and a
philosophical one:
Exercises for the reader:
(1) Is the revised version of (f) consistent with (x)(y)(x #

y :::> [Rx] :F [Ry])?
(2) Does t~e amended version of @ support the important
applications of the theory as well as (f) does?
2. Are There Too Few Objects?
As developed, (J) is a type theory: what can be predicated of

entities of one type-s-objects, properties, propositions-cannot
meaningfully be predicated of entities of another type. So, for
example, we can say of the proposition s that Agatha wonders
(whether) s:
aW{s},
but It IS not even well formed to say that Agatha wonders
(whether) Sherlock Holmes:
aW{h}.
This case does not seem particularly troublesome, but others
give rise to greater uneasiness. Suppose I have a dream in which
there is a proposition which Agatha believes, and which (in the
dream) is chartreuse (i.e., the proposition is chartreuse) . Some-
how a chartreuse proposition seems like the sort of thing you can
dream about; yet it isn't allowed within @.
GLOBAL ISSUES 241
There are two problems here. First, it at least seems mean-

ingful to say that Agatha believes somethingwhich is chartreuse:
(3s)(aB{s} & Cs);
yet this is not well-formed in (f). And second, we might want to
hold that such an entity is a character in Agatha's dream, and
that there is something (something unreal, maybe impossible,
maybe even in some sense incoherent) which is dreamed about
by Agatha and is chartreuse. Crudely put, why admit round
squares and then stop short of chartreuse propositions?
The first problem is easily solved, probably in many different
ways. We could expand our symbolism so as to allow any type of
variable to occur in any position in a sentence, but that would
make the syntax hard to read. Instead, let me adopt a somewhat
rough approach that will do equally well. I will just relax the
well-formedness constraint on identities, so that terms of different
types may flank the same identity sign. Then we can write things
like 'Agatha is goldenness', symbolizing it as:
a = g;
or 'Bill is the proposition that every round square is round' :
b = (x)(Rx & Sx ::> Rx).
And then we can write meaningfully that there is something
chartreuse that Agatha believes, by writing:
(3x)(3s)(x = s & ~B{s} & ex).
The second problem is much harder to deal with. Here are
various options.
Option I: No cross-category identity statements are true.
All characters of dreams and stories are objects. Such characters
may, according to the dream, be propositions or properties,
etc., and they may (again according to the dream) have proper-
ties that are appropriate only to propositions or properties, but
they do not have these properties simpliciter. Such properties,
like extranuclear properties, should simply be ignored when
identifying the characters of dreams or stories (recall the discus..
sion of chapter 7, section 7. In the present case as well, watered...
down properties may serve to make the theory more plausible.

For example, if an object has "being a property" attributed to
it in a dream, then the dream attributes to it:
[lx(3P) (x = P)].
Then there will be a watered-down version of this property,
namely:
[).Nx(3P )(x = P)],
and this will normally also be attributed to the object, so the

object will have a property that might be written 'the property
of being a property',')
Option I is consistent with the official position taken in this
book. But some may find it implausible, and in any case others
are worth exploring. So, in the following discussion, I will sup-
pose, at least for the sake of exposition, that it is inadequate.
Option 2: All properties, relations, propositions, etc . , are in
fact objects, and can be treated as such in the applications of
the theory. In addition to the other axioms of (f) we also have:
(P) (3x) (x = P);
(s) (3x) (x = s);
etc.
Unfortunately this won't work, for it leads to inconsistency.
It could be carried out in part, for it would be consistent to
identify all one-place nuclear properties with objects. For example,
we could identify each property p with the object which has p
and no other nuclear properties (recall the discussion of Plato's
theory of forms in chapter 8). But in at least some other cases
the view is inconsistent, In particular, if extranuclear properties
are identified ,with objects, then, no matter which objects they
are identified with, the theory becomes inconsistent. The argu-
ment is a short one, patterned after Russell's objection to Frege's
theory of the Grundgesetze der Arithmetik, If all extranuclear prop-
erties are objects, then the extranuclear property of "not being
identical with any extranuclear property that you possess" is an
object; call it Xo. That is:
Xo =df [Ax-'" (3P) (x = P & Px)].
GLOBAL ISSUES 243
Now we ask whether Xo "has itself." Begin by assuming that it

does:
[AxI"V (3P) (x = P & Px)]xo.
Then, by abstract elimination:
~ (3P)(xo = P & Pxo).
That is, (P)(xo = P ::> '--Pxo). But plugging in Xo (i.e., [AXI"V
(3P)(x = P & Px)]) for Pyields "'[Ax'" (3P) (x = P&Px)]xo,
which contradicts our assumption.
So Xo doesn't have itself; that is:
'" [Ax-"" (3P)(x = P & Px)]xo.
But then, by abstract elimination and double negation:
(3P)(xo = P & PXo).
But the P in question must be Xo, that is, [Ax'" (3P) (x = P & Px)),
and so Xo does have itself after all, contradicting the above proof
that it doesn't.
So option 2 is not viable.
Option 3: We might try to make all objects, properties, prop-
ositions, etc., be species of a "neutral" sort of entity, and expand
the theory so that analogues of axiom OB] hold for this sort of
entity. This is actually a vague sort of suggestion, which could
probably be developed in many different ways. I think that al-
most any way would lead to the same sort of inconsistency that
infects option 2.
Option 4: 4 The idea here is that, when we have a dream in
which some character is "a property," we are really dreaming
about an object-an object which is not a property. Because,
when we dream that that object "is a property" we are not really
dreaming that it is a property, but rather dreaming something
else which it is easy to misreport as dreaming that the object is a
property. We are dreaming that the object has a certain prop-
4. For purposes of discussing this option, please ignore the cross-category
identity statements introduced above. To include them at this point would
complicate the discussion, (This does not mean that they should obviously be
prohibited from the theory.)
erty, a property which is appropriate to objects, and which is

therefore not the property of being a property. It is a different
property-let rne call it the property of being a property* .
This is reminiscent ofFrege's view in [C&O]. When we try
to attribute properties to properties in the same manner in which
we attribute properties to objects we fail, and instead attribute
(different) properties to objects. The properties that we attribute
are proxies for the properties that we appear to be attributing.
Thus when I dream of a "chartreuse property," I am really
dreaming of an object which has, in the dream, the properties of
being chartreuse and of being a property*-but not of being a
property, for this would make no sense.
But then there is the following objection: now that we un-
derstand the difference between being a property* and being a
property, we can have a dream in which we clearly do not confuse
one with the other, and in which we attribute the latter, not the
former, to an object-an object which is also, in the dream,
chartreuse.
The only viable rejoinder that I know of to this is to deny
that we understand 'being a property' as opposed to 'being a
propertys" at least to deny that we understand it in such a man-
ner that we can attribute it to an object. Perhaps dreaming (or
believing or surmising) that something chartreuse is a property
is like "wondering whether Sherlock Holmes'"; you can say the
words to yourself, and doing so may give rise to a host of images
and thoughts, but that doesn't mean that you are succeeding in.
believing a proposition that is expressed by those words. Perhaps
what happens is that we then have before our mind part of a
model of a theory of objects and properties, a model in which
certain objects play the role of objects, and certain others play
the role of properties. We are actually thinking about object
representatives and property representatives. (Again, this was
Frege's view in [C&O],. the difference being that Frege thought
there were enough objects to model the whole theory within the
realm of objects, whereas we now know that that isn't possible.)
There is a standard objection to any such view as this. It is
that if the attributions in question are indeed impossible (or if the
"wrong" ones keep getting substituted in thought and language
GLOBAL ISSUES 245
for the "right" ones), then the theory itself is undercut. For the
presentation of the theory of (!) used terms like 'is a property', and
employed them in such a way that it makes sense to attribute
them to objects. For example, it makes sense to say 'no object is a
property', and this is just as meaningless (according to the view
now being explored) as 'some object is a property'.
Actually I am not at all sure that my presentation of (9 does
presuppose that 'no object is a property) makes sense. Perhaps a
careful exposition of the theory could obey all the ground rules of
the theory itself. But suppose that this is not possible, Then the
theory could not be given an exposition which is meaningful. This
doesn't mean that the theory itself, (9, can't be meaningfully
stated; it's only the metatalk, the explanation of the .theory,
which is not meaningful. But this is a special sense of 'meaning-
ful', and it's not clear 'that this is so bad. Perhaps it is merely a
dogma that presentations of theories must be meaningful in order
to be worthwhile. Surely it is reasonable to take an instrumental-
ist view of the presentations of theories, if not of the theories them-
selves, and then the most" that can be expected of them is that
they lead us to true theories. Perhaps my presentation of (9 is like
that. The presentation of the theory is an act which has conse-
quences. One of the consequences is that people come to under-
stand a meaningful and true theory. If the presentation is not
itself meaningful and true, then they will not arrive at the theory
(from tbe presentation) by means of explicit definition and de-
ductions of logical consequences; but most of our learning does
not take place by those means anyway.
3. Some Concluding Remarks
Tve left a lot of loose ends, but in a way I think this is appropri-
ate. If I had tied them all up, the resulting theory would reflect
my own idiosyncrasies even more than the present one does, and
it would be more difficult for others to interact with it. My feeling
right now is that the things most worth exploring are issues in-
volving temporal properties (see the introduction, section 3, and
chapter I, note 10); the (possible) simplification of the notation
discussed in chapter 4, note 3; and the variants of the theory dis-
cussed in the present chapter-all together with a detailed and

sensitive consideration of applications of such theories. Others
may be better able to do some of these things than I am; they will
have a perspective on the work that I lack, and they may have a
better sense ofwhich issues are most important..
The orthodox view has been developed over several decades
now, and its present form is the result of a group effort, a process
that has been partly cooperative and partly adversariaL Cer-
tainly, if an interesting theory of nonexistent objects is viable, it,
too, will be developed by such a group process. The result is likely
to diverge considerably from what is presented here, and perhaps
in the end an improved orthodox tradition will carry the day. My
hope is that this book has enough interesting ideas in it to fuel
some good work on the subject. At the very least, I think it will be
harder now to simply take it for granted that there are no non-
existent objects.
Selected Bibliography
Adams, R,. M. [LSAA] "The Logical Structure of Anselm's

Arguments." Philosophical Review 80 (197 1) :28-54.
Anderson, A. R. [CTE&EQ] "Completeness Theorems for the
Systems E of Entailment and EQof Entailment with Quanti-
fication." Zeitschriftfur mathematische Logik und Grundlagen der
Mathematik 6 (1960) :201-18.
Barnes, J. [OA] The Ontological Argument. London: Macmillan,
197 2 .
Blocker, H. G.. [TAFE] "The Truth about Fictional Entities."
Philosophical Quarterly 24 (1974) :27-3 6..
Castaneda, H-N. [TSW] "Thinking and the Structure of the
World." Philosophia 4 (1974) :3-40 .
Chisholm, R. [P] Perceiving: A Philosophical Study. Ithaca: Cornell
University Press, 1957 .
. ... ---. [RBP] Realism and the Background of Phenomenology. New
York: Free Press, 1960.
--~-. [BB&NB] "Beyond Being and Nonbeing." In HaIler
[JSN] .
.... _-. [HO] "Homeless Objects." Reoue Intemationale de Philo-
sophie 104/5 (1973) :2°7-2 3.
--. [TIR] "Thought and Its Reference." American Philo-
sophical Quarterly 14 (1977) :167-72.
247
248 SELECTED BIBLIOGRAPHT
Church, A. [IML] Introduction to Mathematical Logic. Princeton:

Princeton University Press, 1956.
Crittenden, C~ (TANB] "Thinking about Non-being." Inquiry
16 (1973) :290-3 1 2 .
Davidson, D. [RRS] "The Role of Reference in Storytelling."
Unpublished draft.
Devine, P. [LOF] "The Logic of Fiction." Philosophical Studies
26 (1974) :389- 99.
Donnellan, K. [SN] "Speaking of Nothing." Philosophical Review
83 (1974) :3-3 1 • .
Field, H . [TC&IR] "Theory Change and the Indeterminacy

of Reference." Journal ofPhilosophy 70 (1973) :462-8 1.
Findlay, J. N. [MTOV] Meinong's Theory of Objects and Values..
Oxford: Oxford University Press, 1963.
Frege, G. [FA] Foundations of Arithmetic. New York: Philosophi-
cal Library, 1950.
- - . [C&O] "On Concept and Object." In P. Geach and
M. Black (eds.), Translations from the Philosophical Writings
ofGottlob Frege. Oxford: Blackwell, 1952.
Gregory, R. L. [E&B] Eye and Brain. New York: McGraw-Hill,
1966.
Grossman, R. [M] Meinong. London: Routledge & Kegan Paul,
1974·
Hailer, R. (ed.). (JSN] ]enseits ooti Sein und Nichtsein. Graz:
Akademische Druk- u. Verlagsanstalt, 1972.
Harmon, G. [HUP] "How to Use Propositions." American Philo-
Heintz, ]. [RIF] "Reference and Inference in Fiction.. " Un-
published draft.
Henkin, L. [CTT] "Completeness in the Theory of Types."
Journal of· Symbolic Logic 15 (1950) :81-9 I . Reprinted in
SELECTED BIBLIOGRAPHY 2'49
J-..
Hintikka (ed.), The Philosophy of Mathematics. Oxford:
Oxford University Press, 1969-
Hintikka, J. [CES] "CogitoErgoSum: Inference or Performance?"
Philosophical Review 71 (1962) :3-32.
Howell, R. [LSPR] "The Logical Structure of Pictorial Repre-
sentation." Theoria 40 (1974) :76-109.
- - . [FO:HAHA] "Fictional Objects: How They Are and
How They "Aren't." Poetics 8 (1979) :129-77.
Kaplan, D. [QI] "Quantifying In." In D. Davidson and
J. Hintikka (eds.), Words and Objections: Essays on the Work
of w. V. Quine. Dordrecht: Reidel, 1969. Reprinted in-
Linsky [R&M]..
- - . [B&.C&T&A] "Bob and Carol and Ted and Alice." In
J. Hintikka, J. Moravcsik, and P. Suppes (eds.), Approaches
to Natural Language. Dordrecht: Reidel, 1973.
- - - . [Dl "Demonstratives.' Unpublished draft.
Kenny, A.. [DOA] "Symposium: Descartes' Ontological Argu-
ment." In J. Margolis (ed.), Fact and Existence. Oxford:
Blackwell, 1968. "
Kripke, S. [SCML] "Semantical Considerations on" Modal

Logic." Acta Philosophica Fennica 16 (1963) : 83-94. Reprinted
in Linsky [R&M].
Kuhn, T. [SSR] The Structure of Scientific Revolution. Chicago:
University of Chicago Press, 1962.
Lambert, K. [B&BS] "Being and Being So." In HaIler [JSN].
- - - . [TO] "The Theory of Objects (A Critical Review of
J. N. Findlay's Meinong's Theory of Objects and Values)."
Inqui~ 16 (1973):220-30.
- - - . [10] "Impossible Objects." Inquiry 17 (1974) :303-14-
.-. [ODIO] "On 'The Durability of Impossible" Objects'."
Inquiry 19 (197 6 ) :25 1- 53.
.) and van Fraassen, B. [D&C] Derivation and Counter-

example. Encino, Calif.: Dickensen, 1972 .
Leonard, H. [EA&P] "Essences, Attributes and Predicates."
Proceedings and Addresses of the American Philosophical Associa-
tion 37 (19 64) :25-5 1 •
Lewis, C . I. [MOM] "The Modes of Meaning." Philosophy and
Phenomenological Research 4 (1944) :236-49·
Lewis, D. ITIF] "Truth in Fiction." American Philosophical Quar-
terly 15 (1978) :37-46. -
Linsky, L. [R&R] "Reference and Referents." In C. Caton
(ed.), Philosophy and Ordinary Language. Urbana: University
of Illinois Press, 1963.
- - - . [R] Referring, New York: Humanities Press, 1967.
- - . (ed.). [R&M] Reference and Modality. London: Oxford
University Press, 1971 .
-._-. [N&D] Names and Descriptions. Chicago: University of
Chicago Press, 1977~
Marcus, R . [Q&O] "Quantification and Ontology." Noiis ·6
(197 2 ) :24o-50~
Martin, R. M., and Schotch, p~ [MFN] "The Meaning of Fic-
tional Names." Philosophical Studies 26 (1974) :377-88.
Mates, B. [EL] Elementary Logic. New York: Oxford University
Press, 1965-
- - - . [LPW] "Leibniz on Possible Worlds." In H. Frankfurt
(ed.. ), Leibniz: New York: Doubleday, 1972.
Matheson, G. [SST] "The Semantics of Singular Terms."
Journal ofSymbolic Logic~27 (19 62) :439-66.
Meinong, A. [TO] "The Theory of Objects." In Chisholm

[RBP].
- - - . [USGSW] Uber die Stellung der Gegenstandstheorie im System
SELECTED BIBLIOGRAPHY 251
der Wissenschaften. Leipzig: Voigtlander, I g07. Reprinted in

[G], vol. V.
- - - . [EP] On Emotional Presentation. Evanston, Ill.: North-
western University Press, 1972.
- - . [G] Gesamptausgabe.. Graz: Akademische Druk- u. Ver-
lagsanstalt, 1973.
Montague, R. [FP] Formal Philosophy: Selected Papers of Richard
Montague. Edited by R. Thomason. New Haven and
London: Yale University Press, 1974.
- - . [PIL] "Pragmatics and Intensional Logic." In [FP].
- - - . [PTQ] "The Proper Treatment of Quantification in
Ordinary English." In [FP].
Nietzsche, F. [BG&E] Beyond Good and Evil. Translated in
W. Kaufmann (ed.), Basic Writings ofNietzsche. New York:
Random House, 1966.
Niven, L., and Pournelle, J. [MGE] The Mote in God's Eye. New
York: Pocket Books, 1975.
Parsons, K. Pyne. [CMC] "A Criterion for Meaning Change."
Philosophical Studies 28 (1975) :367-96.
Parsons, T. [PMS] "A Prolegomenon to Meinongian Seman-
tics." Journal of Philosophy 71 (1974) :561-80.
-_.. - . [MAFO] "A Meinongian Analysis of Fictional Objects."
Grazer Philosophische Studien I (1975) :73-86.
_ ..- . [N&EPM&L] "Nuclear and Extranuclear Properties,
Meinong and Leibniz." Noiis 12 (197 8) :137-51.
_._ ..- . [RW] "Review of J. Woods' The Logic of Fiction."
Synthese 39 (197 8) :155-64.
- - - . [TT&OL] "Type Theory and Ordinary Language."
Forthcoming (fall 1979) in a volume on Montague gram-
mar; edited by M.. Mithun and S. Davis, University of
Texas Press, Austin.
252 SELECTED BIBLIOGRAPHY
- - - . [RNO] "Referring to Nonexistent Objects.. " Theory and

Decision I-I (1979) :95- 1 10.
Plantinga, A. [NN] The Nature ofNecessity. Oxford: Oxford Uni-
versity Press, 1974.
Pollard, D. E. B. [F&S] "Fiction and Semantics.. " Ratio 15
(1973) :57-73·
Prior, A. [MQSS] "Modality and Quantification in 85." Journal
ofSymbolic Logic 21 (1956) :60-62.
- - - . [PP&F] Past, Present and Future. London: Oxford Uni-
versity Press, 1967.
Quine, W. V.. o. [OWTI] "On ·What "There Is.') In From a
Logical Point of View. New York: Harper, 1953.
- - ,-. [STL] Set Theory and ItS Logic. Cambridge, Mass.:
Harvard .University Press, 1963.
Routley, R. [EM]] "Exploring Meinong's Jungle." Unpub-
lished manuscript.
- .- . [AFOMO] "Another 'Fatal' Objection to Meinongian
Objects." Philosophical Studies 24 (1973) :1-2 I.
- - . [DIO] "The Durability of Impossible Objects." Inquiry
19 (197 6) :247-5 1 •
- - - . [SSFD] "The Semantical Structure of Fictional Dis..
course." Poetics, forthcoming.
- - , and Routley, V. [RMTO] "Rehabilitating Meinong's
Theory of Objects." Revue Intemationale de Philosophic 104/5
(1973) :255-65-
Russell, B. [1904 Review] "Meinong's Theory of Complexes and
Assumptions." Mind n.. s. 13 (1904) :2 °4- 1 9, 336-54, 50 9-2-4 .
Reprinted in [EA].
- - - . [EIP] "The Existential Import of Propositions." Mind
n.s, 14 (1905):398-401. Reprinted in [EA] .
._ - . [1905 Review] "Review of: A. Meinong, Untersuchungen
SELECTED BIBLIOGRAPHT 253
zur Gegenstandstheorie und Psychologie;" Mind n.s, 14 (1905):

530-38. Reprinted in [EA].
- - . [OD] "On Denoting." Mind n.s, 14 (1905) :479-93.
Reprinted in [EA] and several other. sources.
- - - . [1907 Review] "Review of: A.. Meinong, Ober die Stel-
lung der Gegenstandstheorie im System der Wissenschaften:" Mind
n.s, 16 (1907) :436-39. Reprinted in. [EA].
_._-. [IMP] Introduction to Mathematical Philosophy. London:
AlIen & Unwin, 1919-
- ..- . [POM] Principles of Mathematics. 2d ed. New York:
Norton, 1964.
- - . [EA] Essays in Ana{ysis. Edited by D. Lackey. New York:
Braziller, 1973.
Ryle, G. [10] "Symposium: Imaginary Objects." Proceedings of
the Aristotelian Society supp.. vol. 12 (1933): 18-43.
- - -.. [ITNT] "Intentionality-Theory and the Nature of

Thinking." Revue Internationale de Philosophic 104/5 (1973):
255-6 5.
Scales, R . [A&E] "Attribution and Existence." Doctoral disserta-
tion, University of California, Irvine, I 969~
Scott, D. [AML] "Advice on Modal Logic . " In K. Lambert
(ed.), Philosophical Problems in Logic. Dordrecht: Reidel,
197°·
Segerberg, K. [TDML] "Two-Dimensional Modal Logic."
Joumal ofPhilosophical Logic 2 (1973) :77-96.
\
Thornason, R., and Stalnaker, R . [M&R] "Modality and Ref-

erence." Nous 2 (1968) :35g-72.
Urmson J. O. [PA] Philosophical Ana?Jsis. New· York: Oxford
University Press, 1956. .
van Inwagen, P.. [COF] "Creatures of Fiction.. " American Philo-

Wellek, R., and Warren, A. [TL] Theory of Literature. New

York: Harcourt Brace & World, 1942/1970.
Wolterstorff, N. [WWA] "Worlds of Works of Art." Journal of
Aesthetics andArt Criticism 35 (1976) :121-32. .
Woods) J. [LF] The Logic of Fiction. The Hague: Mouton, 1974.
Index
Abstraction principles, 72-73, Complete objects, 19-22, 26,

103-04 , 109-IO, 158-59 106-07, 173n ; defined, 19, 106.
Abstract objects, 10-11 See also Incomplete objects
Abstracts, 66-67, 103-05, 1 17, Complex predicates, 103-05, 157,
162-63 163, 185-86, 227-28
'According to the story p • • 't 53-54,
Compossibility, 222
56~60, 175-82 Concrete objects, lQ-11
Account, maximal, 175-211 .Conjunction, nuclear, 21n, 105
,Actual' t .1.3. Consistency of the theory, 90-92
Anselm, 213-15 Constructivist theory of fictional ob-
Authors, utterances by, 187-88 jects, 194-97
Axioms: listed, 7D-77, 12 4-27 , Contradiction, 31; law of, 38,42, 171
158-60; axioms for objects, 74, Contradictory objects, 42
159. 235, 239-40 Correlation of objects with sets of
properties, 17-18, 73, 79, 81,
Barcan formulas, 100--01, 162 92-97, 17 1
Barnes, J-, 214n Counterfactual properties, 186-87
Being, 10, 42; vs. so-being, 42. See "Creating" (objects in stories), 51,
also Sein/sosein 188
Brentano, F4' 48
Bundles of properties, tBn, 93 De dictolde re: distinctions explained,
4&-48; readings contrasted, 1 i.i ,
Castaneda, H-N., 171, 172 112n, 117-18, 122, 149, 163-64,
Categories, 200, 231 166, 214-15. See also de re
Characters: of stories, 180-211; Definite descriptions, 30-3 2 , 43,
of scientific myths, 229-30; of I 11-20, 164-65; Meinongian,
dreams, 240-44. See also Dream 1 18-20. See also RusseIl, theory of
objects; Fictional objects descriptions
Cogito, 217-18n De re, 21n, 35, 55, 58, 224; reading of
Comparatives, 24, 168-70 'in the story', 175, 181, 183. See
eompleteness theorem, 83-85, 224n also De dicta/de re
255
256 Index
Descartes, R., 213, 217-18 Ideal entities, 232

Disjunction, nuclear t 21 n Identity, 27-29; of indiscernibles,
Donnellan, K., 5, 121 28-2 9, 74, 159;~ conditions, 81,
Dream objects, 207-10, 243 110; statements, 142, 156, 166,
241-43. See also Substitutivity of
Essence, 186,224-26 identicals
Essentialismv aaa-ef Immigrant objects, 5 1-60 , 175,
'Exist' t l<r-ll, 22-24 182-85, 189, 202, 204; in dreams,
Existence, 2 13-17; kinds of, I 1 ; 208-09; in visual art, 210-11; in
extranuclear, 25; watered-down, scientific myth, 229
42-44; possible, 186; fictional, Impossible objects, xi, xii, 2, 21-22,
188; and passim 25, 3 1 , 4 2 ; in fiction, 184; in
Existent gold mountain, the, 42-44. dreams, 209; in visual arts, 21 L
See also Gold mountain See also Possible objects
Existential generalization, 33, 35 Incompleteness: in maximal ac-
'Exist in', 50 count, 182; 'of characters, 183-84;
Extranuclear predicates, 22-26; es- of fictional objects, 2°5; of objects
sentially extranuclear, 167-68; in in dreams, 209; of objects in visual
stories, 197-202, 203; and passim. arts, 21 1; of God, 2 12; of neutri..
See also Nuclear/extranuclear dis- nos, 230-31; radical and
tinction nonradical, 183-84- See also In-
corn plete objects
'Fictional': as an extranuclear predi.. Incomplete objects, 1g-2 1 56,
7
. cate, 23, 26 77-78. See also Complete objects;

Fictional objects, 32, 4g-60, 172, Incompleteness
1 73, 175-2 I 1 Inconceivable objects, 40, 236
Free. logic, 8 Indiscernibles: identity of, 28-29,
Frege, G., 18,24, 45,216,244-45 74, 159; in fiction, 190-94
Intentionality, 30-3 1 , 44-48
Gestalt, 208, 211 Intentional predicates and proper-
God, 212-17 ties, 23-24, 44-48, 186, 214
Gold mountain, the, xi, 11, 18,20,
23, 30, 93, 94; as referent of Kant, I., 40, 216
Meinongian definite description, King of France, 2n, 112, 115-16
30, 118-20; yielded by axioms, 74,
159~ not a set of properties, 93. See Leibniz, G., 21g-20
also Existent gold mountain Logical constructions, 92
Logically closed objects, 1 g-2 2,
Hamlet, 201, 211 106-07; defined, 106
Hamlet, 200-01
Holmes, Sherlock, 2, 4, 3 2 , 34; ex- Man in the doorway, the, 27-28,
isting in fiction, 11; as a paradigm 114- 1 5, 165
fictional object, 4g-6o, 170, 172, Meinong, A., xi, xii, 1-24 passim,
180-go, 202-05; not a set of prop.. 30-48, 118, 21Sn
erties, 93; reference to, 1 2 1; hav- Mirroring worlds, 219, 223-24
ing relational properties, 194-97 Modalities, 98-103, 162; reduction
Index 257
of operators to predicates, 223~24. Plantinga, A·., 202, 225

See also Possible objects; Impossi- Platonic forms, 227
ble objects "Plugging u p" relations, 26-2 7,
Modal moment, 44 5g-6o, 64--69 passim, 75-77,
Monads, 21g--23 156--61 passim, 234-40. See also
Myths, 207 Relations and Relational
properties
Names, 2n, 11 1, 120-23, 165-66; Possible objects, 2, 21-25, 28;
bearing a name, 192; causal theory defined, 21, 101. See also Impossi-
of, 1 2 1, 192; as rigid designators, ble objects
122 Possible worlds, g8-1 03, 21g-24;
Native objects, 5 1--60 , 175, 182-85, and fiction, 56, 182, 202
189, 202-06; in dreams, 208--09; "Prejudice in favor of the actual:' 12,
in visual arts, 2 10-1 I; in scientific 18, 114
legends, 229 Properties, 17-29 and passim
Negation: of nuclear properties, Prepositional attitudes, 45, 108-11,
19-20,42n, 1° 5-06, 227- 28 ; 163-64
within categories, 231 Prepositional functions; 1 1-0
Neutrinos, 230 Propositions, ro, 45-46) 108-10,
Nonextensional contexts, 33-35, 180,233-34. 240-4 1
47-48. See also Intentionality Pure reflectors, 224
"Normal science," 2, 5, 36--37
Nuclear/extranuclear distinction, Quarrtifiers, 6, 11-12, 35, 38; substi-
22-26,4 2- 44, 52 - 54, 64t 155, tu tional vs. objectual, 1 1-1 2, 36
160-61 t 166-74, 216, 227-28 Quine, W., xi, 3, 18,27,37
Nuclear abstracts, 163
Nuclear negations, 183, 206 'Real', 13, 50
Nuclear predicates/properties, Reference: to objects, 3, 8, 30, 37,
19-29,74 52, 111-22; failure of, 37, 43, 56,
Nuclear relations, 5g-6o, 194-g7 112-16, 170-,4,2°7,212,22&-29
Nun object, 22, 198 Relations and relational properties,
26-27, Sg-607 75-17, 156-60 pas-
Objectives, 45 sim, 194-g7, 234-39 passim; rela-
Ontological arguments, 212-17 tions encoded in relational prop-
Ontological commitment, 32-38t erties, 2 1 9
20 4 Relativity theory, 1 2 ~ 4 I
Orthodox theory, 85-97; compari- "Robust sense of reality," 2, 4, 12
son with, 92-97 Round square, xi, 2, 11, 21, 38-42,
Orthodox view (of nonexistent ob- 46; as referent of definite descrip-
jects), 1, 6--12, 33-38; 112n, 162, tions, 307 1 1 8--20; not a set of prop--
202, 246; comparison with, erties, 93; yielded by axioms, 159
160-61 . See also Orthodox theory Routley, R., xii, 1 19, 172
Russell, B., 1-8,. 18, 24, 2~, 30; ob-
Paradoxes, 234-40 jections to Meinong's theory, xi, 1,
Pegasus, 2-11 passim, 35, 37~ refer- 10, 2gn, 27, 30-32,38-44, 17 1 ,
ence to, 11 2, 1 2 1 172; theory of descriptions, 2-5, 8,
258 Index
Russell, B. (continued) Substitutivity of identicals, 33-35,

37; objections to Frege's theory, 110-11, 164, 166
242; paradox concerning proposi- Surrogate 'objects, 52, 57-59, 183
tions, 233-34
Tensed properties, I 1, 28n
Satisfaction principle, 30-32 Theoretical entities: of literary criti..
Scope distinctions, 1 17 cism, 202-05, of science, 228-3 2
Sein/sosein, 23n. See also Being 'There is (are)', 00, 6-7, 205
Sets, 31, 93, 101, 190, 195, 235
Sherlock Holmes. See Holmes, Van Inwagen, P., 202-06
Sherlock Vulcan, 228-30
Stories, 175-82; inconsistency in,
177-78, 182, 184, 206; within ,.Watered·down" properties, 44, 65,
other stories, 200-02; with us as 68, 155, 220, 225; axioms concern-
native characters, 218-19- See also ing, 73, 159; used to define nu-
·According to the story . . . '; Fie· clear abstracts, 1°4; used in
tionaI objects fiction, 184, 186, 192,200,201,
Subsistence, 10, 45n 206

Terence Parsons Nonexistent Objects

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Terence Parsons Nonexistent Objects

Uploaded by

Copyright:

Available Formats

Nonexistent

Yale Unioersity Press

Designed by James J4 ]ohnson

Library of Congress Cataloging in Publication Data

without whom I might have been one

I. Initial Sketch of the Theory 17

4. The Language {f} 63

APPENDIX: A SEMANTICS FOR A FRAGMENT OF

I I I: Applications and Discussion

6. Summary of the Formal Development and More

3· DIGRESSION CONCERNING THE AUTHOR 187

8. ALTERNATIVE THEORIES 202

8. Traditional Issues from the Present Perspective 2 12

9· Global Issues 233

Alexius Meinong was an Austrian philosopher and psychologist

constituting one of the clearest examples of philosophical pro-

Are there objects that don't exist? The orthodox, mainstream

'Rut' has an unfavorable connotation; the same point can

In 1900 Bertrand Russell believed in nonexistent objects, a view

3.. The theory of descriptions is essentially a method for paraphrasing

Most of Western philosophy has agreed with Russell ever

names of nonexistents false, a matter that has been controversial

4. See Linsky [R] chap. V for a discussion, and chapter 5, section 4, of

the development of modern logic, he provided a kind of founda-

2. The Current Scene

Let me repeat my view of the current scene. Metaphysically, we

tively common to teach elementary logic in a manner that pre-

(d) There are winged horses-Pegasus, for example.

one example above, the manner in which we use logic to sym-

because I think that any fruitful philosophical theory is going to

My intent is to describe in some detail the ontological commit-

I have had certain goals in mind when working on this project,

here. If I am successful in my enterprise, some people-who begin

exists in mythology, Sherlock Holmes exists in fiction, . Some

One last note: throughout this work I have always used

Initial Sketch of the Theory

I am going to assume that no two existing objects have exactly

Now, make a list of all existing objects. Correlated with each

Real Objects Sets ofProperties

The set of Oa'8 properties

°a+l {goldenness, mountainhood, . . .}

Actually we can dispense with talk of lists and correlations

should keep in mind that this is a somewhat unusual use of the

2. It is possible to develop a different notion of incompleteness which

hazel-eyedness but it does not have the nuclear property of being

I have said so far (e.g., I haven't given a precise definition of

that it did stand for a nuclear property, existence. Now consider

5. This argument parallels one of Russell's objections to Meinong; see

nuclear and extranuclear is not peculiar to Meinong at all; it's

This is because if F is nuclear and F is not a member of X, then

7. This test for distinguishing nuclear from extranuclear properties will

of being a unicorn, this schema can be used to show that being

about by Meinong. This plugging-up process will be discussed in

One of the major sources of opposition to nonexistent objects is

It is clear that the conclusion expr.essed in the last sentence is a

9- These difficulties with nonexistent objects are foreshadowed in Russell

Identity is meaningfully applicable to all objects, existent

10. In a less simplified development of the theory, I would use "tern-

There will be at least a couritably infinite number of these. Are

Meinong and .Motivation

The basic insight behind Meinong's theory of objects was that

I. This involves an interpretation of Russell's argument that goes beyond

jects, a rather severe reaction. Certainly some of the natural "ob-

2. Evidence for the Theory