ABSTRACT OBJECTS

SYNTHESE LIBRARY

STUDIES IN EPISTEMOLOGY,
LOGIC, METHODOLOGY, AND PHILOSOPHY OF SCIENCE

Managing Editor:

JAAKKO HINTIKKA, Florida State University, Tallahassee

Editors:
DONALD DAVIDSON, University of California
GABRIEL NUCHELMANS, University of Leyden
WESLEY C. SALMON, University of Arizona

VOLUME 160
 EDWARD N. ZALTA
Department of Philosophy, Rice University, Houston, Texas

ABSTRACT
OBJECTS
An Introduction to Axiomatic Metaphysics

D. REIDEL PUBLISHING COMPANY

t...AII
A MEMBER OF THE KLUWER " ACADEMIC PUBLISHERS GROUP

DORDRECHT/BOSTON/LANCASTER
 Library of Congress Cataloging in Pnblication Data

Zalta, Edward N., 1952-

Abstract objects.

(Synthese library; v. 160)

Bibliography~ p.
Includes index.
1. Metaphysics. 2. Object (Philosophy) I. Title.
BDll1.Z34 1983 111 82-20549
ISBN-13: 978-94-009-6982-7 e-ISBN-13: 978-94-009-6980-3
DOl: 10.1007/978-94-009-6980-3

Published by D. Reidel Publishing Company,

P.O. Box 17, 3300 AA Dordrecht. Holland,

Sold and distributed in the U.S.A. and Canada

by Kluwer Boston Inc.,
190 Old Derby Street, Hingham, MA 02043, U.S.A.

In all other countries, sold and distributed

by Kluwer Academic Publishers Group,
P.O. Box 322, AH Dordrecht, Holland.

All Rights Reserved

Copyright to 1983 by D. Reidel Publishing Company, Dordrecht, Holland
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical,
including photocopying, recording or by any informational storage and
retrieval system, without written permission from the copyright owner
To my parents
"Sometimes, unexpected flashes of instruction were struck out by the
fortuitous collision of happy incidents, or an involuntary concurrence of
ideas, in which the philosopher to whom they happened had no other
merit than that of knowing their value, and transmitting unclouded to
posterity that light which had been kindled by causes out of his power."

Samuel Johnson
The Rambler
Saturday, September 7, 1751

vii
 T ABLE OF CONTENTS

PREFACE Xl

INTRODUCTION I
1. Theory, Data, and Explanation I
2. The Origins of the Theory 6

CHAPTER I. ELEMENTARY OBJECT THEORY 15

1. The Language 16
2. The Semantics 19
3. The Logic 28
4. The Proper Axioms 32
5. An Auxiliary Hypothesis 37

CHAPTER II. APPLICATIONS OF THE ELEMENTARY THEORY 40

1. Modelling Plato's Forms 41
2. Modelling the Round Square, etc. 47
3. The Problem of Existence 50
Appendix 52

CHAPTER III. THE MODAL THEORY OF ABSTRACT OBJECTS

(WITH PROPOSITIONS) 59
1. The Language 59
2. The Semantics 61
3. The Logic 68
4. The Proper Axioms 73

CHAPTER IV. THE APPLICATIONS OF THE MODAL THEORY 77

1. Truth 77
2. Modelling Possible Wodds 78
3. Modelling Leibniz's Monads 84
4. Modelling Stories and Native Characters 91
5. Modality and Descriptions 99

ix
x T ABLE OF CONTENTS

CHAPTER V. THE TYPED THEORY OF ABSTRACT OBJECTS 107

1. The Language 109
2. The Semantics 113
3. The Logic 121
4. The Proper Axioms 124

CHAPTER VI. APPLICATIONS OF THE TYPED THEORY 126

1. Modelling Frege's Senses (I) 126
2. Modelling Frege's Senses (II) 140
3. Modelling Impossible and Fictional Relations 145
4. Modelling Mathematical Myths and Entities 147

CONCLUSION 154

APPENDICES 158
A. Modelling the Theory Itself 158
B. Modelling Notions 167

NOTES 172

BIBLIOGRAPHY 187

INDEX 190
 PREFACE

In this book, I attempt to lay the axiomatic foundations of metaphysics

by developing and applying a (formal) theory of abstract objects. The
cornerstones include a principle which presents precise conditions under
which there are abstract objects and a principle which says when
apparently distinct such objects are in fact identical. The principles are
constructed out of a basic set of primitive notions, which are identified
at the end of the Introduction, just before the theorizing begins. The main
reason for producing a theory which defines a logical space of abstract
objects is that it may have a great deal of explanatory power. It is hoped
that the data explained by means of the theory will be of interest to pure
and applied metaphysicians, logicians and linguists, and pure and applied
epistemologists.
The ideas upon which the theory is based are not essentially new. They
can be traced back to Alexius Meinong and his student, Ernst Mally, the
two most influential members of a school of philosophers and psychologists
working in Graz in the early part of the twentieth century. They
investigated psychological, abstract and non-existent objects - a realm of
objects which weren't being taken seriously by Anglo-American philoso-
phers in the Russell tradition. I first took the views of Meinong and Mally
seriously in a course on metaphysics taught by Terence Parsons at the
University of Massachusetts/Amherst in the Fall of 1978. Parsons had
developed an axiomatic version of Meinong's naive theory of objects. The
theory with which I was confronted in the penultimate draft of Parsons'
book, Nonexistent Objects, had a profound impact on me. Parsons' work
was a convincing new paradigm of philosophical investigation.
While canvassing the literature during my research for Parsons' course,
I discovered, indirectly, that Mally, who had originated the nuclear/
extranuclear distinction among relations (a seminal distinction adopted
by both Meinong and Parsons), had had another idea which could be
developed into an alternative axiomatic theory. This discovery was a result
of reading both a brief description of Mally's theory in J. N. Findlay's
book, Meinong's Theory of Objects and Values (pp. 110-112) and what
appeared to be an attempt to reconstruct Mally's theory by W. Rapaport

xi
XlI PREFACE

in his paper "Meinongian Theories and a Russellian Paradox". With the

logical devices Parsons had used in his book, plus others learned from
my colleagues or invented on my own, the alternative theory was
elaborated and applied in a series of nine short working papers written
between November 1978 and September 1979 (the third one was co-
authored with Alan McMichael and published; the others are un-
published). Since then, the current work has been thrice drafted - once in
1979, once in 1980, and once in 1981. The first draft assimilated the nine
working papers. The second was submitted as my Ph.D. dissertation at the
University of Massachusetts/Amherst. The third and present draft, which
preserves the essential structure of the second, was written during my stays
at the University of Auckland and Rice University. This final draft is a vast
improvement on its predecessors - it contains both significantly new ideas
and exposition and more crystalline development of the technical material.
Chapters I, III, and V contain somewhat technical presentations of
successively more powerful versions of the theory. I suggest that readers
less technically inclined skip Section 2 of each of these chapters, since
these contain the model-theoretic semantics which will prove useful for
answering questions about the consistency of the axioms, completeness
of the logic, etc. But they are not essential for understanding the primitive
metaphysical and logical notions used in the statement of the axioms.
Sections 1, 3, and 4 of these chapters however, contain short though
valuable expositions of the language, logic, and proper axioms of the
theory, respectively.
The entire project could not have been carried out without the
inspiration and assistance of both teachers and colleagues. Throughout
the writing of the first and second drafts, Parsons served as a sharp critic.
Our conversations every couple of weeks always left me with an idea for
improving my work or with an outline of a problem which had to be
tackled and solved. It is to his credit that he was such a great help, despite
the fact that our theories offered rival explanations to certain pieces of
data. Barbara Hall Partee graciously gave her time in weekly discussions
during the writing of the first draft. Her enthusiasm, encouragement, and
suggestions were invaluable. My colleague, Alan McMichael, also deserves
special mention. Besides teaching me the techniques of algebraic semantics,
and discovering (and helping to solve) a paradox within the theory,
McMichael served as my first critic. During the writing of the first two
drafts, whenever I discovered a new application of the theory or got stuck
on a point of logic, I frequently presented it to Alan. His criticisms and
 PREFACE xiii

suggestions helped me to sharpen up many of the intricate details.

I would also like to thank Mark Aronszajn, Blake Barley, Max Cresswell,
Cynthia Freeland, Edmund Gettier, Richard Grandy, Herbert Heidel-
berger, Larry Hohm, Michael Jubein, Robert Nola, Dana Scott, Krister
Segerberg, and Robert Sleigh. Spirited discussions with these individuals
forced me to think deeply about a variety of metaphysical and logical issues.
Special thanks go to the members of the philosophy department at the
University of Auckland for allowing me to lead a series of staff seminars at
which the details ofthe theory were expounded. It was during those lectures
that many ways were found to improve the second draft.
Finally, thanks goes to Nancy Scott, Helen Bedford, and Jane Addington
for their dedication in typing unfriendly looking manuscripts.

December 1981 E. N.Z.

Houston
 INTRODUCTION

1. THEORY, DATA, AND EXPLANATION

In this book, we shall produce a research program in metaphysics.

Following Lakatos, a research program in metaphysics consists of three
parts:l (1) a theory about what things there are, which stands at the "core"
of the program; (2) a "heuristic" which grounds the principles of in-
vestigation and explanation, organizes the data to be explained, and
provides a problem solving machinery for transforming recalcitrant data
into confirming evidence; and (3) a "protective belt" of auxiliary hypotheses
which grow out of the heuristic to guard the metaphysical theory against
refutation. Our particular research program can be characterized using
this framework.
The metaphysical theory which stands at the core of the program may
be stated roughly as follows: in addition to existing (actual, or real) objects
(like you, me, my desk, sub-atomic particles, etc.), and the properties and
relations they exemplify, there are abstract entities as well. Among the
abstract entities we find abstract objects (or as some might prefer to say,
abstract individuals), abstract properties, and abstract relations. For the
major part of this book, the theory of abstract objects is developed and
applied (Chapters I-IV). At the end, the theory of abstract properties and
relations is developed and applied, using the resources of a new kind of
type theory (Chapters V-VI). For convenience, we refer to the overall
theory as the theory of abstract objects. 2 The first principles of the theory
will tell us not only the precise conditions under which there are particular
abstract objects, but also the conditions under which two such objects
are identical. This will provide us with a clearly defined background
ontology.
The heuristic we associate with this theory revolves around two central
tasks of all scientific research programs: (1) show that accepting the first
principles of the theory allows us to construct explanations of interesting
pieces of data, and (2) show that there is no good evidence for thinking
that there must be other (kinds of) entities than those required by the
theory. Such a heuristic anchors the following two methodological
2 INTRODUCTION

principles of investigation and explanation concerning the history of

philosophy and the philosophy of language:
(A) History of philosophy: If earlier philosophers who postulated
theoretical entities were describing anything at all, they must
have been describing entities which can be found in our
background ontology. So try to construe the discussions of
philosophers who described theoretical entities like Forms,
Monads, Possible Worlds, Senses, etc., as discussions about
existing or abstract objects, properties, or relations.
(B) Philosophy of language: If the terms of a natural language
denote anything at all, they must denote entities found in the
background ontology. So try to show both that abstract and
existing objects, properties, and relations are denoted by terms
of natural language and that there are no true sentences which
contain terms that denote entities other than these.
These principles leave us with philosophically interesting data to be
explained.
The data consists of true sentences of natural language (and their
entailments) and there are two basic kinds: the A PRIORI truths and the A
POSTERIORI truths. There are two kinds of A PRIORI data. The first consists
of metaphysical hypotheses, possibly developed or discussed by earlier
philosophers, which we intuitively believe to be true? For example, here
are two hypotheses developed by earlier philosophers, followed by two
hypotheses about possible worlds which we take to be true:
(i) If there are two distinct F-things, then there is a Form of Fin
which they both participate (Plato, Parmenides, 132a).
(ii) ... each simple substance (i.e., monad) .. . is a perpetual living
mirror of the universe (Leibniz, Monadology, §56).
(iii) A proposition is necessarily true iff it is true in all possible
worlds (Leibniz, "Necessary and Contingent Truths"?).
(iv) There is a unique actual world.
We hope to show that certain abstract objects display features resembling
those of Platonic Forms, that others display features resembling Leibnizian
Monads, while still others display features resembling possible worlds. By
saying, for example, that certain abstract objects display features resembl-
 INTRODUCTION 3

ing those of Platonic Forms, we mean three things: (1) a definition such
as the following,
x is a Form=df"'x""
can be given using only the primitive and defined notions of the theory;
(2) features of the Forms that Plato describes are definable in the theory
as well; and (3) it follows from the first principles of the theory that the
abstract objects which satisfy our definition of "Form" have the features
Plato says HIS Forms are supposed to have. Consequently, we shall suppose
that data like (i)-(iv) have been explained if there are reasonable facsimiles
of them which turn out to be consequences of the theory.
The second kind of A PRIORI data consists of sentences such as the
following:
(v) The round square is round.
(vi) The set of all sets which are not members of themselves is a set of
all sets which are not members of themselves.
(vii) The fountain of youth is a fountain.
Even if we have only a rough, pre-theoretical understanding of what
(v)-(vii) assert, no experience is needed to decide whether they are true.
(v) and (vi) are just part of the evidence we use to establish that the round
square and the set of all sets which are not self-members are impossible
objects. In what follows, we shall try to show that abstract objects can
serve as the denotations of the descriptions in (v)--{vii). And as with all A
PRIORI data, we suppose that (v)-(vii) have been explained if we can deduce
them as consequences of the first principles of our theory.
The A POSTERIORI data also fall into two major groups. The first group
consists of statements we ordinarily make about fictional characters,
mythical figures, dream objects, and the like. Here are some examples:
(viii) Santa Claus does not exist.
(ix) Stephan Dedalus is a fictional character.
(x) In the myth, Achilles fought Hector.
(xi) Some Greeks worshipped Dionysus.
(xii) Ponce de Leon searched for the fountain of youth.
(xiii) Franz Kafka wrote about Gregor Samsa.
4 INTRODUCTION

To construct a PRIMA FACIE case for thinking that fictional characters,

mythical figures, etc., just are abstract objects, we shall focus on a formal
language that we develop in Chapter III. Certain definitions tell us the
conditions under which a given sentence of the formal language is true.
We then translate (viii)-(xiii) into our formal language using names and
descriptions in the language which denote abstract objects (as well as
existing ones). Thus, an explanation of (viii)-(xiii), and others like them,
consists in showing that they can be translated into sentences which
preserve their intuitive truth value and their entailments.
The other group of A POSTERIORI data contains triads of sentences. These
are sentences which involve verbs of propositional attitude and the "is"
of identity:

(xiv) S believes that Socrates taught Plato.

(xv) S does not believe that the son of Phaenarete taught Plato.
(xvi) Socrates is the son of Phaenarete.
(xvii) S believes that Woodie is a woodchuck.
(xviii) S does not believe that Woodie is a groundhog.
(xix) Being a woodchuck just is being a groundhog.

We follow Frege in supposing that the English terms inside propositional

attitude contexts do not have their ordinary denotations and that they
denote their senses instead. This was Frege's explanation of why (xiv) and
(xvi) do not imply the negation of (xv), and why (xvii) and (xix) do not
imply the negation of (xviii). However, the senses of terms denoting objects
will be construed as abstract objects and the senses of terms denoting
properties (relations) will be construed as abstract properties (relations).
To do this, we focus on a formal language developed in Chapter V.
(xiv)-(xix) are translated into our language, using names and descriptions
of the language which denote abstract entities that serve as the senses of
English terms. Thus, our Fregean explanation of the consistency of each
triad lies in showing that the sentences which translate the members of a
given triad are consistent.
These, then, will be the kinds of data and explanation which shall occupy
our attention. In the course of solving problems and broadening the scope
of the theory, we will adopt auxiliary hypotheses in both metaphysics and
the philosophy of language. But we can not describe these even in a rough
way in advance of the rigorous presentation of the theory.
The research program we have just outlined has been designed to
 INTRODUCTION 5

compete with the current programs in metaphysics. No attempt will be

made to provide a list of these alternatives, but it would serve well to
mention just a few. The most influential one has developed around a
theory that many philosophers have attributed to Russell, namely, that
existing objects, and the properties and relations they exemplify, are the
only things there are. 4 We shall sometimes talk as if the established
"Russellian view" on certain issues is the only one to take seriously, partly
because this view has been so widely regarded as true, but partly because
other influential research programs, which initially appear to differ from
the Russellian view, are nevertheless closely allied to it. (For example, one
research program has developed around the metaphysical theory that
individuals, sets, and possible worlds are the only things there are. 5 While
this ontology differs from the classic Russellian ontology, philosophers in
this tradition reconstruct properties and relations out of functions from
possible worlds to sets of sequences of individuals and thereby provide a
link with Russellian metaphysics.) In addition to the Russellian program
and its modern counterparts, there is also the physicalist program, which
guards the theory that the entities described by the correct physics are
the only things there are (the less radical physicalists commit themselves
to sets as well).6 And finally, we should note that there are other programs
besides ours which have been motivated by Meinong's views on what
there is.7
Among a host of competing research programs, Lakatos would
distinguish those that are "progressive" from those that are "degenerating".
Basically, the distinction is that in a progressive program, the addition of
auxiliary hypotheses does not just accommodate known facts and anoma-
lies, but also leads to the discovery of hitherto unknown, novel facts.
Research programs may go through phases of progression, degeneration,
and then progression again, and it is hard to tell whether a current phase
of degeneration will be a permanent one. So the fact that a particular
program is presently in a phase of degeneration is not sufficient reason
for switching to an alternative program. However, it is good reason for
careful scrutiny of the successes of alternative programs.
In what follows, we shall not argue that all other programs are in
permanent phases of degeneration. Nor shall we argue that the data we
have presented is not, and cannot, be assimilated by the other programs.
Enough AD HOC auxiliary hypotheses can be appended to any theory to
enable it to handle the data. Instead, we will try to establish one thesis,
namely, that the research program in metaphysics developed here is
6 INTRODUCTION

progressive one (i.e., our theory helps us to explain our data and, together
with auxiliary hypotheses, predicts hitherto unknown, novel facts). Then,
any philosopher, linguist, or cognitive scientist who agrees that the data
we have chosen are important, interesting, and currently lack natural
explanations, should either consider our program as a viable alternative
or be prompted to find progressive explanations from within current
programs.
The above description should give the reader a rough idea of the nature
of our project. But in the course of setting up our theory, many
philosophical issues will be confronted. For example, we will end up
developing a full-fledged theory of relations (where properties and proposi-
tions turn out to be one-place and zero-place relations, respectively). This
includes a definition which tells us when any two relations (properties,
propositions) are in fact the same. Semantics for our formal languages are
developed in which we may consistently suppose that logically equivalent
relations are distinct. The resulting metaphysical system should be
attractive not only because it might handle important kinds of data which
seem problematic for current traditions, but also because it exhibits many
interesting and philosophically satisfying qualities in its own right.

2. THE ORIGINS OF THE THEORY

The theory we will develop has its origins directly in the naive theory of
nonexistent objects which Meinong and Mally investigated at the turn of
the century. A very simple statement of the theory upon which Meinong
seemed to be relying in his early work is the following, which we call
Naive Object Theory:

(NOT) For every describable set of properties, there is an object which

exemplifies just the members of the set.

For example, there is an object which exemplifies just the properties in

the set which contains only the two properties roundness and squareness.
If we round out (NOT) by assuming "Leibniz' Law", namely that two
objects are identical if and only if they exemplify exactly the same
properties, then such an object would be unique. Maybe Meinong's idea
was that this object is what is being talked about or what is denoted when
the definite description "the round square" is used. (NOT) also guarantees
an object which exemplifies just the properties in the set of properties-
attributed-to-Zeus-in-the-myth. Maybe this object is denoted by the name
"Zeus". It is important here for the reader to try to anticipate how (NOT)
 INTRODUCTION 7

might be used to construct explanations of some of the other data we

presented in Section 1.
(NOT) may be stated precisely in a second order predicate calculus
without directly invoking the notion of a set or the set membership relation.
Such a statement would be preferable to philosophers like myself who
want to initially remain neutral on the ontological status of mathematical
entities like sets, but who are willing to commit themselves to objects (in
general) and relations. A second order predicate calculus is an interpreted,
deductive system built upon a second order language in which the following
three metaphysically primitive notions are embedded:
object: x, y, z, . ..
n-place relation: Fn, Gn, H n, . ..
(where properties=df 1-place relations: pl,Gl,H l , . ..)
Xl"" ,xn exemplify pn: pn Xl ... Xn"

The symbolic representations of the notion of exemplification ("P"x i ... xn")

serve as the atomic statements of the formal language. The language also
utilizes three primitive logical notions:
It is not the case that cp: '" cp.
/fcp, then tf;: cp~tf;.
Every x(every pn) is such that cp: (Vx)¢, (Vpn)cp.
We frequently abbreviate (Vx)¢ as (x)cp and (VF")cp as (pn)cp. The
other logical notions of our basic predicate calculus such as both cp and
tf;("¢ & tf;"), cp or tf;("¢ v tf;"), cp if-and-only-i/tf;("¢ == tf;"), and some x(some
pn) is such that cp("(:Jx)¢), (:JF")¢"), can all be defined in the standard way.
The simple and complex statements which can be constructed out of
these primitive and defined notions reveal the expressive power of the
system.
To see how to express (NOT) precisely in this system it is important
to first look at a general method for describing sets of properties which
employs our second order language. Then we can indicate how to represent
(NOT) without mentioning sets. Consider the following open formula
"Socrates exemplifies Fl". If we let "s" denote Socrates, then we can
represent this condition on properties in our language as "plS". Now we
can form the following description of a set : the set of all properties pI
such that Socrates exemplifies pI, i.e., {plIP1s}. This describes the set of
properties which satisfy (in Tarski's sense) the open condition "pIS." The
set contains properties like being a philosopher, being Greek, being
snub-nosed, etc. Here's another example, where "p" denotes Plato. Take
the open condition "both Socrates exemplifies Fl and Plato exemplifies
8 INTRODUCTION

F I " ("Fls & Flp") and form the set abstract: the set of all properties FI
such that both Socrates exemplifies FI and Plato exemplifies FI, i.e.,
{FIlFls & Flp}. The set described here contains such properties as being
a philosopher and being Greek as well, but it would not contain the
property of being snub-nosed, since Plato did not exemplify that property
(let us suppose). Consider another example using the second order language
with identity. We can form the open condition "either Fl is identical to
the property of being round or Fl is identical to the property of being
square" ("F I =Rl v FI =Sl"). Then form the description: the set of all
properties FI such that FI is identical to roundness or Fl is identical to
squareness, i.e., {FIIFI =Rl v Fl =Sl}. This describes a set which con-
tains just the two properties satisfying the open condition, namely
roundness and squareness. We can even consider formulas without a free
property variable to express vacuous conditions on properties. For
example, if "P" denotes being a philosopher, then every property is a
member of {FliPs}. In this manner, any condition on properties expressible
in our language can be used to describe sets of properties.
Now (NOT) asserts that for each such set of properties, there is an
object which exemplifies all and only the properties in the set. This assertion
can be captured in our second order language without mentioning sets
by using an axiom schema. An axiom schema is basically a rule which
says that every sentence of a certain form shall be considered to be an
axiom. Let ¢ be any condition on properties expressible in our language
(possibly with identity), as in the above examples. Then any instance of
the following sentence schema is to be an axiom:
(NOT) (3x)(F I )(F l x == ¢), where ¢ has no free x's.
Here are four instances of (NOT') which guarantee that there are objects
which correspond to the sets of properties described in the above
examples: 8
(a) (3x)(FI)(Fl x == FI s)
(b) (3x)(FI)(FIX==F I s&Flp)
(c) (3x)(FI)(FIX==Fl=RI VFl=SI)

(d) (3x)(F 1 )(F l x == Ps).

A complete theory of objects may be obtained by adding Leibniz's Law
to the infinite set of axioms generated by (NOT):
(LL) x = y == (Fl)(FI X == FI y).
 INTRODUCTION 9

(LL) ensures that the object yielded by an arbitrary instance of (NOT')

is unique, since there couldn't be distinct objects which exemplify exactly
the properties satisfying the given condition 4>.
Although we shall not specifically attribute the theory which results to
Meinong, it does appear to be the natural way to formalize the principles
upon which he seemed to be relying. Unfortunately, there are lots of things
wrong with the theory, and philosophers since Russell, who have worked
with informal versions of the theory, have been quick to recognize this.
For one thing, (a) yields an object exemplifying just the properties Socrates
exemplifies. By (LL), any such object just is Socrates. So the A PRIORI
metaphysics rules that Socrates has being. Yet the being of Socrates seems
to be a contingent matter. Secondly, (c) is incompatible with the natural
assumption that whatever exemplifies roundness fails to exemplify square-
ness. (c) also implies that Russell never thought about the round square (on
the natural assumption that the property of being thought about by Russell
is distinct from both the property of being round and being square). This is
dubious, at best.
But there are much more serious difficulties with the theory. It implies
falsehoods and is incompatible with a very important principle yielding
complex relations. Consider first the following instance of (NOT), noted
by Russell, where "E!" stands for existence, "G" stands for goldenness,
and "M" stands for mountainhood:
(:Jx)(F)(Fx ==F=E! v F=G v F=M).
This implies the falsehood that there is a golden mountain which exists.
But the most serious problem with (NOT') is that it is inconsistent with
the following abstraction schema for relations:
(:JFn)(Xl) ... (xn)(Fnx 1 •.• Xn == ¢), where ¢ has no free FH's.
Here are two typical instances of this schema:
(:JF)(x)(Fx == ~ Gx)
(:JF)(x)(Fx == Gx & Hx).
The first guarantees that any given property G will have a negation, while
the second guarantees that any two given properties G and H will have
a conjunction. There are many other kinds of complex properties yielded
by this schema as well. But consider, in particular, the following:
(:3F)(x)(Fx == Rx & ~ Rx).
This says that there is a property objects exemplify iff they exemplify
10 INTRODUCTION

redness and it is not the case that they exemplify redness. Call an arbitrary
such property K. The assumption that something exemplifies K produces
an immediate contradiction. But (NOT') ensures just that:
(:Jx)(F)(Fx == F = K).

So if we want to keep our abstraction schema for relations as it is stated,

we have to give up (NOT') (there are other interesting ways to produce
contradictions from (NOT') and the relations schema, but we shall not
discuss them here).
One suggestion by Mally to refine (NOT) was to distinguish two types
of properties - nuclear and extranuclear. 9 The nuclear properties an object
has are its "ordinary" properties and are more central to its nature and
identity than its extranuclear properties. Terence Parsons follows up on
this suggestion. lo He adds this distinction as one new primitive to a
standard second order predicate calculus. He develops the theory and
logic associated with nuclear and extranuclear relations in general. He
restricts the property quantifier in (NOT') so that it ranges just over
nuclear properties. He also restricts the relation quantifier in the abstrac-
tion schema for complex relations so that it ranges just over extranuclear
relations. His theory is based on the following three principles, where "F m '
ranges over extranuclear n-place relations and "f"" ranges over nuclear
n-place relations Y
(I) For every condition on nuclear properties, there is an object
which exemplifies just the properties satisfying the condition

(:JX)(fl)(flx == ¢), where ¢ has no free x's.

(II) Two objects are identical iff they exemplify the same nuclear
properties
x = y == (fl)(flx == fly).
(III) For every extranuclear relation, there is a nuclear relation
which is coextensive with it on the existing objects
(F")(:Jj")(Xl)···(x")(E!Xl & ... &
E !x" -+ (f"x 1 ... x" == F"x 1 ... x")).
So on Parsons theory, there are two kinds of relations and one kind of
object. Principle III yields a nuclear, "watered down" version of each
extranuclear relation, but it is a consequence of the theory that distinct
 INTRODUCTION 11

extranuclear relations sometimes have the same nuclear watered down

version.
Consequently, Parsons avoids generating the above oddities, falsehoods,
and inconsistencies. Although the theory asserts that there is an object
which exemplifies just the nuclear properties Socrates exemplifies and that
this object just is Socrates, it is not a consequence of the theory that
Socrates exemplifies extranuclear existence. The object which exemplifies
just nuclear existence, nuclear goldenness, and nuclear mountainhood,
("the existent golden mountain") provably does not exemplify extranuclear
existence. (Nuclear existence is just the watered down version of extra-
nuclear existence.) The object which exemplifies just nuclear roundness and
nuclear squareness could, and in fact did, exemplify the extranuclear
property of being thought about by Russell (intentional properties are
classic extranuclear properties). This object doesn't violate the principle
that everything exemplifying roundness fails to exemplify squareness
because the principle is false when the quantifier "everything" is allowed to
range over nonexistent objects as well as existing objects (of course, all
existing objects exemplifying roundness fail to exemplify squareness). Also,
the only way to produce the property K described above is to use the
abstraction principle for complex extranuclear relations (i.e., use the
principle we used to produce K in the first place, except now, the variable
"F"" ranges just over extranuclear relations). So K is an extranuclear
property and the theory does not imply that there is an object which
exemplifies being-red-and-not-being-red. And with these obstacles out of
the way, Parsons finds interesting applications for his theory. In particular,
he models fictional characters (and the like), Leibnizian Monads, and
suggests how to model Plato's Forms. 12 These models served as prototypes
for the models we have constructed in our alternative object theory.
Our theory of abstract objects is based on an entirely different suggestion
of Mally's, however. He distinguished two relationships which relate
objects to their properties. On Mally's view, properties can DETERMINE
objects which do not in turn SATISFY the properties. 13 For example, the
properties roundness and squareness can determine an abstract object
which satisfies neither roundness nor squareness. The properties of
existence, goldenness, and mountainhood can determine an abstract object
which does not satisfy any of these properties. The properties which
determine an abstract object are central to its identity. For a recent attempt
to reconstruct Mally'S theory, see W. Rapaport's discussion in "Meinon-
gian Theories and a Russellian Paradox".14
12 INTRODUCTION

In what follows, we construct languages capable of representing the

distinction between satisfying and being determined by a property.
However, we shall employ different terminology. We shall say that an
object exemplifies a property instead of satisfying it. We shall say that
an object encodes a property instead of saying that the object is determined
by the property. The distinction between exemplifying and encoding a
property is a primitive one and will be represented by a distinction in
atomic formulas of the languages we construct. All the primitive notions
that we shall need in order to state the first principles of the theory are
listed and followed by their symbolic representations:
Primitive M etaph ysical Notions
object: x, y, z, ...
n-place relation: Fn, Gn, Hn, .. .
Xl' ... ' xn exemplify Fn: Fnx l ... Xn
X encodes Fl: XF1.

Primitive Logical Notions

it is not the case that ¢: ~ ¢
if ¢, then I/I:¢ -+ 1/1.
every X (every Fn) is such that ¢: (Vx)¢, (Vpn)¢.
Primitive Theoretical Relations
existence: E1.
Using these basic notions, we define a property to be a one-place relation
and say that x is abstract ("A !x") iff x fails to exemplify existenceY We
also say that two objects x and yare identicalE ("x = EY") iff x and y both
exemplify existence and exemplify the same properties (for reasons which
we cannot go into here, in Chapter I we will take = E as a primitive
two-place theoretical relation and cast the preceding definition as a proper
axiom).
Now if we have understood Mally's insight correctly, the main principle
of the theory must assert that for every condition on properties, there is
an abstract object which is determined by just the properties meeting the
condition. Using our new terminology, this can be captured by the
following principle:
(I) For every expressible condition on properties, there is an
abstract object which encodes just the properties meeting the
condition:
(3x)(A!x & (Fl)(xFl == ¢)), where ¢ has no free x's.
 INTRODUCTION 13

The two other principles which serve as the cornerstones of the theory of
abstract objects are: 16
(II) Two objects are identical iff they are identic alE or they are
both abstract and encode the same properties

(III) Two properties are identical iff they are encoded by the
same objects
Fl = G 1 == (x)(xF 1 == xG 1).

Principle (I) gives us "being" conditions for abstract objects. Principle (II)
gives us identity conditions for all objects. And Principle (III) yields identity
conditions for properties (in Chapter I, we will generalize on this definition
to obtain an identity principle for all relations). On our theory, in contrast
to Parsons, there is just one kind of relation and we will avail ourselves
of the abstraction schema for complex relations described above without
restricting the variable "pn" in any way (though, in order to avoid
paradoxes, we shall not allow any new relations to be constructed using
encoding formulas - only the relations constructible in the standard second
order calculus will be found).
With these principles, we will find an abstract object which encodes
just the properties Socrates exemplifies ((3x)(A!x & (F)(xF == Fs))). But,
clearly, this object is not identical with Socrates. We also find an abstract
object which encodes just roundness and squareness ((3x)(A!x & (F)
(xF == F = R v F = S))). But our principles do not imply that this object
exemplifies either of these properties. They are compatible with the claim
that everything whatsoever which exemplifies roundness fails to exemplify
squareness. We also find an abstract object which encodes just existence,
goldenness, and mountainhood ((3x)(A!x & (F)(xF == F = E! v F = G v
F = M))). Although the theory presupposes that this object fails to
exemplify existence, this is compatible with the contingent fact that no
existing object exemplifies all the properties this abstract object encodes
(which is how we will read the ENGLISH nonexistence claim). Finally, the
abstraction principle for complex relations will generate the property K
in exactly the manner described above and the theory will guarantee that
there is an object which encodes K((3x)(A!x & (F)(xF == F = K))). But it
is provable that this object does not exemplify this property. Indeed, it's
provable that no object does, and our principles are compatible with this
result.
14 INTRODUCTION

In Chapter I, we will couch principles (I), (II), and (III) as a proper

axiom schema and two definitions, respectively. The abstraction schema
for complex relations will be a logical theorem schema. We also axiomatize
the other logical and non-logical principles which round out the theory.
This should make the details of an ontology rich with abstract objects
sharp and accurate. Once our background ontology is set, we will go on
to apply the theory in Chapter II.
 CHAPTER I

ELEMENT AR Y OBJECT THEOR Y

The full presentation of the elementary theory of abstract objects shall

occupy the first four sections of this chapter. In each of these sections, we
concentrate on the following major groups of definitions:

1. The Language.
2. The Semantics.
3. The Logic.
4. The Proper Axioms.

The proper axioms are stated in the language. Since the semantics contains
a definition which tells us the conditions under which an arbitrary formula
of the language is true, we will know what is being asserted by our proper
axioms. The logic we associate with the language allows us to prove the
consequences of the proper axioms.
In the course of the definitions which follow, we frequently provide
examples and make extended remarks to explain and motivate unusual
features. In the remarks, we frequently define (with the help of boldface
and lists) certain syntactic or semantic concepts which will help us to
single out classes of expressions or entities which have certain properties.
We use quotation marks to mention expressions of the language. We
generally omit these standard devices (quotation marks, corner marks)
for mentioning and describing pieces of language when the intent is clear.
We use quotation marks inside parentheses (" . . . "), to give readings and/or
abbreviations of formulas. All definitions of the object language appear
with the label " D n."
With the exception of "X' and "z", we use lower case Greek letters to
range over expressions of the formal language. In particular, we use:
1('S to range over names (i.e., constants)
q;, 0/, x, e to range over formulas
o's to range over object terms
pms to range over relation terms
ex, /3, y to range over all variables
r's to range over all terms

15
16 CHAPTER I

v's to range over object variables

nn,s to range over relation variables
/1, ~, , to range over A-expressions.
Finally, we note that in most of Chapter I, we shall not give the intuitive
readings in natural language of the formulas and complex terms of the
object language. That is because our aim is to focus on the expressive
capacity of a formal language, without prejudice as to how English
sentences and terms are to be translated into the language. However, it
will be useful to provide some examples in natural language, since this
will help the reader to picture what the language and theory can, and
ultimately will, say.

1. THE LANGUAGE

We shall utilize a slightly modified second order language. The ONLY

modification is that new atomic formulas have been added - they express
the fact that an object encodes a property. These new atomic formulas
are called "encoding formulas", and whereas the ordinary "exemplification
formulas" which we shall still have around have n object terms to the
right of an n-place relation term, encoding formulas have a single object
term to the left of a ONE-place relation term. These atomic encoding
formulas can combine with other formulas to make molecular and
quantified formulas. The complex formulas which result may be construc-
ted solely out of atomic exemplification subformulas, solely out of atomic
encoding subformulas, or may be of mixed construction. Many of the
interesting definitions, axioms, and theorems are mixed formulas.
Our language will also have one kind of complex term - the A-
expressions. These terms shall denote relations, and they involve the
primitive logical notion "being such that", which logicians represent with
the A. However, only complex formulas constructed solely out of atomic
exemplification formulas can combine with the A to form relation terms
(for reasons to be explained shortly).
The language which results has much more expressive capacity than
the standard second order language with complex terms. The definitions
which precisely describe the language may be subdivided as follows:
A. Primitive Symbols.
B. Formulas and Terms.
 ELEMENT AR Y OBJECT THEORY 17

A. PRIMITIVE SYMBOLS

We have two kinds of primitive object terms: names and variables.

Officially we use the subscripted letters aI' a2, a 3, ... as primitive object
names, but unofficially, we use a, b, c, ... for convenience. Officially, we use
the subscripted letters Xl' X 2 , X 3 , ..• as primitive object variables, but
unofficially we use X, y, z, .... There are also two kinds of primitive relation
terms: names and variables. Officially, we use the superscripted and
subscripted letters P~,P~, ... ,n:::::1 , as primitive relation names (un-
officially: pn,Q", ... ) and F~,F~, ... ,n::::: 1, as primitive relation variables
(unofficially: F", G", ... ). E! is a distinguished one-place relation name;
= E is a distinguished two-place relation name.
In addition we use two connectives, "', and --t ; a quantifier: V; a lambda:
A; and we avail ourselves of parentheses and brackets to disambiguate.

B. FORMULAS AND TERMS.

We present a simultaneous inductive definition of (propositional) formula,

object term, and n-place relation term.
The definition contains six clauses:
(1) All primitive object terms are object terms and all primitive n-
place relation terms are n-place relation terms.
(2) Atomic exemplification: If p" is any n-place relation term, and
are any object terms, P"OI'" On is a (propositional)
0 10 ... ,0"
formula (read: "0 1 , ... ,0. exemplify relation p"").
(3) Atomic encoding: If pI is anyone-place relation term and 0 is
any object term, Opl is a formula (read: "0 encodes property
pI").

(4) Molecular: If cfJ and t/J are any (propositional) formulas, then
( '" cfJ) and (cfJ --t t/J) are (propositional) formulas.
(5) Quantified: If cfJ is any (propositional) formula, and a is any
(object) variable, then (Va)cfJ is a (propositional) formula.

(6) Complex n-place relation terms: If cfJ is any propositional

formula with n-free object variables VI"'" V., then [Av 1 ··· vncfJ]
is an n-place relation term.
18 CHAPTER I

We rewrite atomic exemplification formulas of the form = E0102 as

01= E02' We drop parentheses to facilitate reading complex formulas
whenever there is little potential for ambiguity. We utilize the standard
abbreviations: (¢ & 1/1), (¢ == 1/1), (¢ v 1/1), and (30M. And we define:
Dl x is abstract ("A !x") = dI[Ay '" E!y ]x.
Here then are some examples of formulas: p 3 axb ("a, x, and b exemplify
relation p 3 "); aG ("a encodes property G"); '" (3x)(xQ & Qx) ("no object
both encodes and exempiifies Q"); (x)(E!x -+ '" (3F)xF) ("every object
which exemplifies existence fails to encode any properties"); and (3x)
(A!x & (F)(xF == Fa)) ("some abstract object encodes exactly the properties
a exemplifies").
By inserting all the parenthetical remarks when reading the above
definition, we obtain a definition of propositional formula. In effect, a
formula ¢ is propositional iff ¢ has no encoding subformulas and ¢ has
no subformulas with quantifiers binding relation variables.! Only proposi-
tional formulas may occur in ),-expressions. A-expressions allow us to
name complex relations. 2 We read [AV 1 ••• vn¢] as "being objects VI"'" vn
such that ¢(v 1 , ... , vnY', or as "being a first thing, second thing, ... , and
nth_thing such that ¢". For example: [Ax", Rx] ("being an object x such
that x fails to exemplify R"); [Ax Px & Qx] ("being an object x such that
x exemplifies both P and Q"); [Axx=Eb] ("being identicalE with b");
[AxyPx & Syx] ("being objects x and Y such that x exemplifies P and y
bears S to x"); [Ax (3y)Fxy] ("being an x such that x bears F to something");
[Axyz Gzx & E!y] ("being a first, second, and third thing such that the
third bears G to the first and the second exists"). 3
Since arbitrary formulas ¢ cannot appear after A's, the following
expressions are ill-formed: [AxxP], [AYYP&py], [h(3F)Fx], and [Ax
(3F)(xF & '" Fx)]. The first two are ill-formed because the formula after
the A has an encoding subformula; the third because the formula contains
a quantifier binding a relation variable ("relation quantifier"); the fourth
fails both "restrictions" on propositional formulas. The "no encoding
subformulas" restriction is essential. It serves to prevent paradoxes in the
presence of the proper axioms. For a detailed discussion of the paradoxes
this move eliminates, see Appendix A, part A (especially the discussion
concerning "Clark's Paradox"). The "no relation quantifiers" restriction
is not essential. However, it allows us to effect a huge simplification of
the semantics. Since we shall not critically need to use A-expressions with
relation quantifiers in the applications of the theory, we choose not to
complicate the semantics any further than necessary. We shall show how
 ELEMENT AR Y OBJECT THE OR Y 19

to bypass this restriction once we move to the typed theory of abstract

objects (Chapter V). The semantics for the language which couches the
typed theory more easily assimilates the interpretation of A-expressions
with "higher-order" quantifiers. The net result of these restrictions is that
no relation denoting expression not already found in the standard second
order language can be constructed. Intuitively, this means we will be
working with familiar sorts of complex properties and relations.
These A-expressions widen the possibilities for atomic and complex
formulas: [hyPx & Qy]ab ("a and b exemplify being two objects x
and y such that x emplifies P and y exemplifies Q"); x[AY~ Ry] ("x en-
codes failing-to-exemplify-R"); (:lx)(A!x & (F)(xF == (:lG 2 )((Gab & F =
[AyGyb]) v (Gba & F = [AyGby])))) ("some abstract object encodes just
the relational properties a exemplifies with respect to b").
Finally, we say that, is a term iff, is an object term or there is an n
such that, is an n-place relation term.
In the definitions which follow in Sections 2, 3, and 4, it shall be useful
to have precise definitions for certain syntactic concepts which up until
now, we have used on an intuitive basis: 4 all and only formulas and terms
are well-formed expressions. An occurrence of a variable rx in a well-formed
expression is bound (free) iff it lies (does not lie) within a formula of the
form (Vrx)¢ or a term of the form [Av1 ... rx ... vn ¢] within the expression.
A variable is free (bound) in an expression iff it does (does not) have a free
occurrence in that expression. A sentence is a formula having no free
variables.
Furthermore, a term, is said to be substitutable for a variable rx in a
formula ¢iff for every variable [3 free in " no free occurrence of rx in ¢
occurs either in a subformula of the form (V [3)ljI in ¢ or in a term
[A v1 ... [3 ... vnljl] in ¢. Intuitively, if , is substitutable for rx in ¢, no free
variable [3 in , gets "captured" when, is substituted for rx, by a quantifier
or A in ¢ which binds [3. We write ¢(rx1, ... ,rx n) to designate a formula
which mayor may not have rx 1, ... ,rxn occuring free. Finally, we write
¢~:: . :. :~: to designate the formula which results when, for each i, 1 :-s; i:-s; n,
' i is substituted for each free occurrence of lXi in ¢.

2. THE SEMANTICS

The definitions which help to determine the conditions under which the
formulas of the language are true may by grouped as follows:
A. Interpretations.
B. Assignments and Denotations.
20 CHAPTER I

C. Satisfaction.
D. Truth under an interpretation.
In the definitions which follow, we use script letters as names and variables
for sets, entities, and functions which are all peculiarly associated with
the semantics.

A. INTERPRETATIONS s

<
An interpretation, ,I, of our language is any 6-tuple 1/), flA, ext[!ii, 2,
>
extd' ff which meets the conditions described in this subsection. The
first two members, I/) and flA, must be non-empty classes - they provide
entities for the primitive and complex names of the language to denote
and they serve as the domains of quantification. I/) is called the domain
of objects, and we use o's as metalinguistic variables ranging over members
of this domain. flA is called the domain of relations and it is the union of
a sequence of non-empty classes flAl' flA2' flA 3 ,.··; i.e., flA = Und flA n. Each
flAn is called the class of n-place relations. We use '',zm' as a metalinguistic
variable ranging over the elements of tli n. flA must be closed under all ofthe
logical functions specified in the fourth member of the interpretation (2).
Intuitively, the third, fourth, and fifth members of any interpretation
are functions (or classes of functions) which impose a certain structure on
the elements of I/) and flA. We suppose that for each n-place relation in
flAm there is a set of n-tuples of elements drawn from I/) which serves as
the exemplification extension ("extension[!il") of the relation. Each n-tuple
of the set represents an ordered group of objects which exemplify (bear,
stand in) the relation. The third member of an interpretation is therefore
a function, ext[!ii, which maps each ItEtlin into 'l3(l/)n) ("the power set of
E0 n,,), i.e., ext [!iI: flAn ~ 'l3(l/)n). We call ext [!iI(Jt n) the exemplification extension
of Itn.
The fourth member of any interpretation, 2, is a class oflogical functions
which operate on the members of i?lln and I/) to produce the complex
relations which serve as the denotations for the A.-expressions. Each
complex relation receives an exemplification extension which must mesh,
in a natural way, with the extensions[!il of the simpler relations it may have
as parts.
There are six elements in 2 - the first four are each families of indexed
logical functions: 6 g>2iJ1tt:g i ("i-plug"), iJIt%,I j/" i ("i-universalization"),
~@%j/"i,j ("i,j-conversion"), and flAt! ff 2 i ,j ("i,j-reflection"), where i,j are
elements of the set of natural numbers. The other two members of 2 are
 ELEMENT AR Y OBJECT THEORY 21

particular functions ~(I)%.@ ("conditionalization") and % g~ ("negation").

These six elements of !f work as follows:

(a) f!}!fd/1~ 1 maps (~2U~3U . . . ) x .@ into (~1 U~2U . . .) .

&!fd/1~j' for eachj> 1, maps (~jU~j+l U ...) x .@ into
(~j-l U~ju" .).
:fJ !fd/1~; is subject to the following condition:
ext[1Jl(:fJ !fd/1~;(in,O)) = {<0 1,···, 0 ;-1,0;+ 1"" ,On) I
<0 1,··· 'Oi-l 'O' 0i+ 1"'" On )Ee xtf1Jl{i n)}.

The condition on & !fd/1~i basically says that the extension[1Jl of the new
relation f!} !fd/1~;(in, 0 ) ("the jlh-plugging of i n by 0") includes just those
n - 1 tuples which result by deleting the object ° from the jlh place of every
n-tuple in the extension[1Jl of the original relation i n which has ° in its jIb
place. This ensures, for example, that an object 0 1 which falls in the
extension[1Jl of the property &!fd/1~2(1t2, o S) is such that <OI' OS) is in the
extension[1Jl of i 2 •

(b) d/1% f ill maps (~2 U~3 u ...) into (~l U~2 u . . ).. d/1% ..fY j,
for each j > I, maps (~jU ~j+l u . ..) into (~j-l U~ju " .).
d/1.!V..f Y i is subject to the condition: ext BI (d/1% fil;(in)) =
{<° 1, .. . , ° ;- 1' 0 i+ On) I 1"'"
('<1'0)( <01"'" 0 i-l' 0, 0i+ 1"'" On) Ee xt(in)) }.

The condition on d/1% ..fY ; tells us that d/1% ..fY;(lt n ) ("the jlh_
universalization of in") is an n - 1 place relation which has a given n - I tuple
in its extension[1Jl ifffor every object 0, the n-tuple which results by inserting
° in the jth place of the given n - 1 tuple is in the extensionf1Jl of i n. Intuitively,
d/1% ..fY2 (i 2) is the property of bearing i 2 to everything.

(c) ~(!)%Yi,j' for each j,j, 1 ~ j <j, is a function mapping

(~j U~j+ 1 u ...) into (~jU~j+ 1 u .. .) subject to the condition:
ext[1Jl{~(!)%f jin)) =
{< .0 1' . . . ,.oi - 1' .oj, .oi + l' ... ,,oj - l' ,oj, ,oj+ l' ... ,.on> I
<0 1"" 'Oi" " ,OJ' '' ' , On) EextBl(i n) }.

This says that ~(!)%fi,j(ltn) ("the conversion of Itn about its jlh and /b
places") is an n-place relation which has in its extensionBl all those n-tuples
which result by switching the jtb and r
members of every n-tuple in
<
ext a (ltn). So <0 1' 0 2 )E extBl(~(!) %Y 1,2(2 2 )) iff 0 2, ( 1) Eext BI(i 2).
22 CHAP T ER I

(d) PJU :!J' 5l'i ,j' for each i,j, 1 s i <j, is a function mapping
(~jU~j+l u",) into (~j-l U~jU" ,) subject to the condition:
extfJI(~tff:!J' 5l'i,i2n)) = { <01"'" 0i"' " 0 j -l, 0j+ 1"'" On) I
<01"" 'Oi"'" OJ''' , ,On) Eext fJI(2 n) and 0i = OJ},
When given place numbers i and j, [!lU g; 2';j2n) ("the i,r
reflection of
i n") is an n - 1 place relation which has in its extensionfJI all those n - 1
tuples which result by deleting the P member from every n-tuple in the ex-
tension:Jj> of 2 n which has identical ilh and lh members, This ensures that any
object 0 which falls in the extensionfJI of :aM:!J' 5l' 1,2(22) is such that 0 bears i 2
<
to itself, i,e" 0 , 0 ) Eext.'!Q(22).

(e) ~(9%.@ isa functionfrom(~l U~2 u ...) X (~1 U~2 u" .)into
(~2 U~3 u ...) subject to the condition:
extfJI(~(9%.@ (in,Om)) =
{ <o l, ,, ., on, o ~ , ,,. ,o~ ) I <O l,.,, ,On > ¢ extfJI(in) or
<o~, .. . , o~ ) Eext fJI(om) }.

~(9%.@ maps any n-place relation i n and rn-place relation om to an n + rn

place relation which has in its extensiona any n + rn tuple which either fails
to have an n-tuple from ext a (i n) as its first n members or has an rn-tuple
<
from exta(om) as its second rn members. So 0 1, O2 ) Eexta(1&'(9%.@(i l,ol»
iff 01¢ ext:Jj>(i 1) or 02Eext a (ol).
(f) %tff<;§ is a function from (~lU~2U",) into (~lU~2U",)
subject to the condition: exta(% tff<;§( in» =
{ <0 1"'" On> I<01" '" On ) ¢exta(-i n)}.

% tff<;§(-i n) is an n-place relation which has in its extensiona all of the

n-tuples not in the extensionfJI of i n.
This completes the definitions of the logical functions. They guarantee
that the domain of relations, ~, houses a rich variety of complex relations.
The fifth member of an interpretation is the last function which imposes
a structure on the domains .@ and ~. We suppose that every property
~ 1 has an encoding extension ("extension"","), The encoding extension of
a property is a set of members of .@ which encode the property. The fifth
member of an interpretation is therefore a function, ext""" which maps
~1 into 'l3(.@), i.e. ex<.,, : ~1 ~'l3(.@).
The final member of an interpretation, the :!J' function, maps the simple
names of the language to elements of the appropriate domain. For each
object name K, :!J'(K)E .@. For each relation name Kn, :!J'(Kn)E ~n' Since
"E! " is a simple property name, :!J'(E !)E~l' and so ext a(:!J'(E!)) £ .@. We
 ELEMENTARY OBJECT THEORY 23

call this subset of ~ the set of existing objects ("1&""). We call the complement
of I&" on ~ (i.e., eaa ,gp(.K1&"'9' (~(E!)))) the set of abstract objects ("d").

B. ASSIGNMENTS AND DENOTATIONS

As usual, an assignment with respect to an interpretation § will be any

function, l.ff' which assigns to each primitive variable an element of the
domain over which the variable ranges. And, a denotation function with
respect to an interpretation § and an § -assignment l.ff' will be any function,
d.ff./.?, defined on the terms of the language, which: (1) agrees with ~.ff
on the primitive names, (2) agrees with l.ff on the primitive variables, and
(3) assigns denotations to the complex terms on the basis of the denotations
of their parts and the way in which they are arranged. But consider a
complex term like "[hPx ~ Syx]". Suppose that ~.ff(P) is the property
of being a painting and 31'AS) is the study relation. Our A-expression
would then read: "being an object x such that if x exemplifies paintinghood
then y bears the study relation of it". The denotation of this A-expression
will be assigned in terms ofthe denotations of "P", "S", and "y", and the way
in which these parts of the expression are arranged.
Since the denotation function d .ff,/ (for convenience, we drop the
subscript on the I) must agree with 31'.ff' we know:

d.ff,/(P) = fUI£'ntinfllwod
d.ff,/(S) = the {)tud y relation

d.ff,/ will also agree with 1 on its assignment to "y"; so let us suppose
that d.ff,/(y) = o. However, there are three ways to construct a complex
property which might serve to interpret the way in which these simple
parts are arranged in the A-expression. One alternative is to first plug
d.ff,/(y) into the first place of d.ff,/(S), conditionalize d.ff,/(P) with the one
place property which results, and then reflect the first and second places of
the 2-place relation resulting from the conditionalization. This would give
us:
(]U 31'!£' 1,2 (~(9.K~ (d.ff,/(P),fJJ!£'0l/'9' 1 (d.ff,/(S)' d.ff,/(Y)))).
On the other hand, we might first conditionalize d.ff ,/(P) withd.ff ,/(S) to get
a three place relation, reflect its first and third places to get a 2-place
relation, and then plug d.ff,/(y) into the second place of the result. This
would give us:
fYJ 2iJlt'IJ 2 (fYt,c g;- 2 1,2 (~(1)Jf/'g (dJ,/(P), dJ,/(S))),dJ,I(Y))'
24 CHAPTER I

Finally, we might conditionalize d 5 ,/(P) with d 5 ,/(S), then plug d 5 ,/(y)

into the second place of this 3-place relation, and then reflect the first and
second places of the result. This would give us:
fYl$ $'!l' l,z(@l !l'1llt'lJ z(~(9%f?fi (d5 ,/(P),d5 ,/(S)),d5 ,/(Y))).

That is, the following three properties are all sitting around in P11 and
could equally well serve as the denotation of [hPx ---+ Syx] with respect
to f and/:
fYl$ $'!l' l,z(~(9%§(ftaint{'nflA£N}d, f!jJ !l'1llt'§ 1 (dtad11,0)))
f!jJ !l'1llt'§ z(flItrff $' !!! 1.3(~(9%qJ(ftaintinflAood, dtad11)),0)
flItrff $'!!! l,z(@l !!!Illt'§ z(~(9%qJ(ftaint{'nfllwod, dtad11),0)).
The claim that these three complex properties are in fact the SAME
property is a metaphysical thesis of great interest. The idea is that these
complicated looking script expressions which are displayed immediately
above just represent different decompositions of the same property. Of
course, such a thesis needs to be supported, preferably with a (mathe-
matical) theory which predicts when any two such properties or relations
are identical. 7 But such a theory has yet to be devised. 8
Consequently, we face the question, which of the above three properties
should be assigned as the denotation5 ,/ of our A-expression [Ax Px---+
Syx]? In order to answer this question, we shall develop a mechanical
procedure which selects one of the above properties and which makes a
similar kind of selection for each of the other A-expressions. This mecha-
nical procedure is embodied primarily in a classification which partitions
the A-expressions into seven syntactic equivalence classes. Six of these
classes will correspond to the logical functions found in !l'.9 The seventh
houses all of the "simple" A-expressions. [AxPx ---+ Syx] will be categorized
as a I,2-reflection of the expression [AxuPx ---+ Syu], which in turn
will be categorized as the conditionalization of the two expressions
[AxPx] and [AuSyu]. The first of these is simple and the second will be
categorized as the Pt-plugging of [AwuSwu] by term y. Once the A-
expressions have been partitioned, it will be straightforward to define f-
assignment and denotatio~,/ so that [AxPx---+ Syx] denotes the first of the
above three properties. The definitions off-assignment and denotation5 ,/
follow the partitioning.
Partitioning the A-expressions. We use p,~,' as metavariables ranging
over A-expressions. Suppose p is an arbitrary A-expression. Then
 ELEMENT AR Y OBJECT THEOR Y 25
J.1 = [AV 1 ... vn 4>], for some 4>, VI'.·.' v n. Utilizing the following five major
rules, we then classify J.1 as the iSh-conversion of ~, as the negation of ~,
as the conditionalization of ~ and (, as the ilh-universalization of ~, as the
e,
iSh-reflection of as the ilh-plugging of ~ by 0, or as elementary.
(1) If (3i)(1 ~ i ~ n and Vi is not the t h free object variable in 4> and i is
the least such number), then where Vj is the ilh free object variable in 4>,
J.1 is the i,jlh-conversion of

[AV 1 ··· Vi-lVjVi+l··· Vj-lViVj+l··· vn 4>J.

(2) If J.1 is not the i,fh-conversion of any A-expression, then:
(a) if 4> =( '" t/I), J.1 is the negation of [AV 1 ••• vnt/l]

(b) if 4> = (t/I-+ X), and t/I and X have no free object variables in
common, then where VI' ... ' Vp are the variables in t/I and
vp + 1' ..• ' Vn are the variables in X, J1. is the conditionalization of
[.leVI··· vpt/l] and [AV p + 1 •·· vnX]

(c) if 4> = (Vv)t/I, and V is the ilh free object variable in 4>, then J1. is the
ilh-universalization of [AV l •• · Vi - l VViVi + l ·•• vnt/l].
(3) If J.1 is none of the above, then if (3i)(1 ~ i ~ n and Vi occurs free in
more than one place in 4> and i is the least such number), then where:
(a) k is the number of free object variables between the first and
second occurrences of Vi'
(b) cp' is the result of replacing the second occurrence of Vi with a
new variable v, and
(c) j = i + k + 1,
iSh-reflection of [.leVl ... vi+kVVj ... vnCP'].
J1. is the
(4) If J1.is none of the above, then if 0 is the leftmost object term occurring
in cp, then where:
(a) j is the number of free variables occurring before 0,

(b) cp' is the result of replacing the first occurrence of 0 by a new

variable V, and
(c) i =j + 1,
J.1 is the ilh_plugging of [.leVI ••. VjVV j + 1 ... Vn cp'] by o.
 26 CHAPTER I

(5) If J1 is none of the above, then

(a) ¢ is atomic,
(b) VI"'" vn is the order in which these variables first occur in ¢,
(c) J1 = [AV 1 ·.· vnP"v 1 ••• v.], for some relation term pH, and

(d) J1 is called elementary.

Rules (1)-(5) partition the class of A-expressions into seven equivalence
classes. The reader should verify that: [h Rxb] is the 2nd _plugging of
[AxyRxy] by b; [h(Px~Skx)J is the 1, 2-reflection of [Axy(Px~Sky)J;
[hy (Vw) Bxwy ] is the 2nd -universalization of[.hwyBxwy]; and [hy(Rxx~
Syy)] is the conditionalization of [hRxx] and [AYSyy]; among other
examples.
J-assignments. IO If given an interpretation J of our language, an
J -assignment, I, will be any function defined on the primitive variables
of the language which satisfies the following two conditions:

(1) where Vis any object variable, f(V)E~

(2) where nn is any relation variable, l(nn)E~n"

Denotations. If given an interpretation J of our language, and an

J -assignment f, we recursively define the denotation of term T with respect
to interpretation J and J-assignment 1("dc9'jT)") as follows:
(1) where K is any primitive name, dc9',/(K) = ffc9'(K)
(2) where Vis any object variable, dc9''/(v) = I(v)
(3) where nn is any relation variable, dc9',/(nn) =/(n")
(4) where [AV I ... vnpnVI ... vn] is any elementary A-expression,
d",/([AV I ... vnpnvl'" vn]) = dc9',/(pn)
(5) where J1 is the ith-plugging of ~ by 0,
dc9',/(J1) = [1j> 2'OZt~;(dc9',/(~),dc9',/(o))
(6) where J1 is the ith-universalization of ~,dc9',/(J1)=OZtJVJY/a;'fo"/(~))
(7) where J1 is the i,ph-conversion of ~,
dc9',/(J1) = '?!'@.!VY;jd c9',/(W
(8) where J1 is the i,jth-reflection of ~,
dc9',/(J1) = fJIlrff ff 2';jdc9',/(~))
 ELEMENT AR Y OBJECT THEORY 27

(9) where f1 is the conditionalization of ~ and (,

d'.f(f/) = ~(DJV£0 (d,.lO, d'.f('))
(10) where!1 is the negation of ~,d§,/(!1) = JV'0''§(d§,/(m.
Here are some examples of A-expressions and their denotations:

d$,/([AxRxa]) = & £'Olt'§ 2 (d$,f(R),d$ ,/(a))

d$,/([hSxbd]) =
& £'Oltrg 2 (& £'Oltrg3 (d$ ,/(S),d$, / (d)),d$,/(b))
d$,/([h Px ~Skx])=
fJltlffg; £ ' 1 ,2(~(DJV£0 (d$ ,/ (P),&£'Oltrg 1 (d$, / (S),d$ ,/(k) )))
d$, / ( [..1.xy(\lw) Bxwy ]) = Olt ffyr "f/2(d$ ,/(B))
d$,/([hy Rxx ~Syy])=
C(i(()JV:!l! (fJltcC!!F£'1,2 (d$ ,/(R)), fJltcCg; £'1,2 (d$, /(S)))
d$, / ([AxBx ~(\ly)(Wyx ~ Lmy)]) =
fJltcC!!F £'1 , 2 (C(i(()JV:!l! (d$ ,/(B), OltJVyr "f/ 1 (fJltcC!!F£'1 , 3
(C(i(()JV:!l! (dJ .I(W),&!l'Oltrg 1 (d $,/ (L),d$, ,Am)))))))

C, SATISFACTION 11

If we are given an interpretation yr, and an assignment f, we may define

I satisfies <p, recursively, as follows:
(1) If <p = pno1 ... on ,1 satisfies <p iff <d$,/(01),oo.,d$,/(on)
Eext 9lCd$ ,;Cpn))

(2) If <p =Opl, I satisfies <p iff dJ,/(o) Ee xt"" (d$,/(p1))

(3) If <p = (- t/I), I satisfies <p iff I fails to satisfy <p
(4) If <p = (t/I ~ X), I satisfies <p iff I fails to satisfy t/I or I satisfy X
(5) If <p = (\I(I.)t/I, I satisfies <p iff (\I I')(/~I ~ I' satisfies <p),
where: IV = dil' is an yr-assignment just like I except
perhaps for what it assigns to (1..12

D . TRUTH UNDER AN INTERPRETATION

<p is true under interpretation yr iff every .f-assignment I sa tisfies <p. <p is
false under .f iff no .f-assignment I satisfies <p. Using this definition, we
say that 1> is valid (logically true) itT 1> is true under all interpretations.
28 CHAPTER I

The logical axioms which follow in the next section are all valid. We say
that an interpretation J is a model of elementary object theory iff all the
proper axioms of the theory (Section 4) are true under J.

3. THE LOGIC

The logic for our interpreted language consists of :13

A. Logical Axioms.
B. Rules of Inference.

A. THE LOGICAL AXIOMS

There are an infinite number of formulas which are logically true (valid).
Some of these are designated as logical axioms and they, together with
the rules of inference, store the analytical power of the theory. The logical
axioms are introduced by schemata, which indicate that all formulas of
a certain form are to be axioms. The schemata fall into three groups: the
propositional schemata, the quantificational schemata, and two schemata
governing A-expressions. The second A-schema will be introduced after
some discussion.
Propositional Schemata
LAl: 4>~(ljI~4»
LA2: (4) ~ (ljI ~ X)) ~ ((4> ~ ljI) ~ (4) ~ X))
LA3: (~4> ~ ~ ljI)~ (( ~ 4> ~ljI)~ 4».

Quantificational Schemata
LA4: (CI.)4> ~ 4>~, where 1: is substitutable for CI.
LA5: (CI.)(4>~ljI)~(4>~(CI.)ljI), provided CI. is not free in 4>.
Lambda Schemata
A-EQUIVALENCE: Where 4> is any propositional formula, the following
is an axiom:
(xd··· (X n )([AV 1 ••• Vn 4> JX 1... xn == 4>:::::::::)
Here are some instances of A-EQUIVALENCE: 14
(x)([Ay ~ RyJx == ~ Rx)
(U)(V)([AXYPx &SyxJuv == Pu& Svu).
 ELEMENT AR Y OBJECT THEORY 29

The first says that an arbitrary object x exemplifies failing-to-exemplify-R

iff x fails to exemplify R. The second asserts that any two objects u and
v exemplify being-two-objects-such-that-the-first-exemplifies-P-and-the-
second-bears-the-S-relation-to-the-first iff u exemplifies P and v bears the S
relation to u.
The second A-schema, A-IDENTITY, is stated in terms of some defined
notation. Its being logically true is a consequence of the fact that given an
arbitrary interpretation ~ and assignment /, the denotation.,-,/ of an
arbitrary relation term pn is identical to the denotation,g;,1 of [AV 1 . .. vn
pn v1 ... vnl However, we do not have the primitive logical notion of identity
in our object language to express this fact. Nevertheless, we shall designate a
logically true formula involving pn and [Avl ... vnpnvl ... vn ] to serve as a
definition of identity among relations. This will allow us to interchange pn
and [AV 1 .. . vnpnvl . . . vn] in formulas once we add a proper axiom legitimiz-
ing substitution of identicals in the next section. The formula which is to
serve as the definiens for relation identity itself involves the defined notion
of identity among properties. Consider D 2 :
D2 Fl = G 1 = df(x)(xF 1 == xG 1 ).
At first glance, it will not be apparent why this should serve as a good
analysis of property identity, but at least it has the merit of not being
intuitively false (like a certain other analysis of property identity, namely,
that properties are identical iff they are exemplified by the same objects). We
will try to provide more justification for this definition in Section 4. Notice
that any formula of the form [AVp1V] = pl will be logically true. In any
interpretation, the denotation of the terms flanking the identity sign are
identical, so they must have the same encoding extension.
There is a natural way to generalize the definition of property identity to
obtain an analysis of relation identity. Consider D3:
D3 Fn= Gn =df (where n> 1)
(Xl)···(Xn _l)([AyFnYX1···Xn_l]=[AyGnYX1···Xn_l] &
[AY FnX1YX 2 •• .x n - 1 ] =
[AY GnX1YX 2 •.. x n - 1] & ... &
[AY Fnx 1 .. . x n- 1y] = [AY Gnx1 ... xn-1Y ]).
This definition may be read in the following, intuitive manner: relations
Fn and Gn are identical iff the one-place properties which result no matter
how n - 1 objects are plugged into them (provided F nand Gn are plugged up
in the same way) are identical (i.e., encoded by the same objects). Given
30 CHAPTER I

this definition, formulas of the form [Av l ... Vn pnvl ... vn] = pn are logically
true (and I do not see a way to derive them). So we officially add
A-IDENTITY as a logical axiom schema governing our structure:

A-IDENTITY: where pn is any relation term and vl ' ... , vn are any object
variables, the following is an axiom:

B. RULES OF INFERENCE

(1) Arrow Elimination ("--->E"): from ¢ and ¢--->t/J, we may infer t/J
(2) Universal Introduction ("UI"): from ¢, we may infer (rx)¢.
Officially, these are all the rules we'll need. Open formulas are assertible.
We define the universal closure of ¢(rxl, ... ,rxn) to be (rxl) ... (rxn)¢. It is
straightforward to show that open formulas are true iff their universal
closures are. In the usual manner, a proof will be any finite sequence of
formulas ¢l"'" ¢n such that, for each i, either ¢i is a logical axiom or ¢i may
be inferred from some of the preceding formulas by a rule of inference. ¢ is a
theorem of logic (logical theorem) iff there is a proof of which ¢ is the last
member. We write "f-- ¢" to indicate that ¢ is a theorem of logic. ¢ is a
proof-theoretic consequence of (derivable from, provable from) a set r of
formulas iff there is a sequence of formulas ¢l""'¢n such that ¢ = ¢n and,
for each i, either (a) ¢i is in r, or (b) ¢i is a logical axiom, or (c) ¢i may be
inferred from some of the preceding formulas by a rule of inference. Such a
sequence is called a proof of ¢ from r and to indicate this, we write either
"f--r¢" or "rf-- ¢".
If the set of formulas r constitute the proper axioms of some theory,
and if r f-- ¢ but it is not the case that f-- ¢. then we say that ¢ is a proper
theorem of r. It is important to distinguish the logical theorems of a
theory from the proper theorems. The logical theorems are derivable from
the logical axioms and rules of inference alone, whereas the proper
theorems depend on one of the proper axioms. We may define this notion
of dependence as follows. Let t/J be a formula in a set r of formulas.
Assume that we are given a proof ¢l"'" ¢n from r, together with the
justification for each step of the proof. We then say ¢i depends upon t/J
in this proof iff: (a) ¢i is t/J, and the justification for ¢i is that it belongs
to r, or (b) ¢i is justified as a direct consequence by ---> E or UI of some
preceding formulas of the sequence, where at least one of these preceding
formulas depends on t/J.
 ELEMENT AR Y OBJECT THEORY 31

It will be convenient to employ the many standard derived rules of

inference. For example, we call the rule of inference derivable from ~ E
and LA4 universal elimination ("UE"). Standard formulations of the
existential introduction and elimination rules ("EI" and "EE"), the
quantifier negation rules ("QN"), and the introduction and elimination
rules for "', &, v, and == are employed. And we shall use conditional
and indirect proof techniques. The proofs sketched in the text are all
constructed with the aid of these derived rules and proof techniques.
By using UE on the universal quantifiers of the instances of A-
EQUIVALENCE, we obtain biconditionals. Rules of inference governing
the biconditional allow us to introduce (eliminate) A-expressions into
proofs when the right (left) side of the biconditional is added as a premise.
We may shorten this procedure by formulating two rules of inference
derived from A- EQUIVALENCE, AI and AE: where ¢ is any propositional
formula with object terms ° 10 "" Om and VI, ... , Vn are object variables
°
substitutable for I, ... , Om respectively, then the following are rules of
inference:
(1) A-Introduction ("AI"): from ¢, we may infer
[AV I .. . vn¢:::::::::]OI ... On"
(2) A-Elimination ("AE'): from [.leVI'" Vn¢::::::~:]OI'" Om we
may infer ¢.
Also, since [.Ie VI ••• Vn ¢] is an n-place relation term, it is subject to existential
introduction. We get an important logical theorem schema by applying
EI to A-EQUIVALENCE:

THEOREM(S) ("RELATIONS"): where ¢ is a propositional formula

which has no free F"'s, but has x 10 .•• , Xn free, the following is a theorem:

The instances of this schema tell us what complex properties and relations
there are. Here are some examples:
(a) (y)(3F)(x)(Fx == Gyx) (by UI)
(b) (3F)(x)(Fx == '" Gx)
(c) (3F)(x)(Fx == Gx & Hx)
(d) (3F)(x)(Fx == Gx v Hx)
32 CHAPTER I

(e) (3F)(x)(y)(Fxy = Gyx)

(0 (3F)(x)(Fx = (Vy)Gxy).
Axioms (a)-(f) assert, respectively, that for every object y and two place
relation G, there is a property which results by plugging y into the first
place of G, that every property has a negation, every two properties have
a (non-disjoint) conjunction and disjunction, every two place relation has
a converse, every two place relation has a universalization on its second
place.
Note that RELA nONS, Dz, and D3 jointly constitute a full-fledged
theory of relations. We no longer need to suppose that relations are
"creatures of darkness". They have precise "being" conditions and
precise identity conditions. It is not a consequence of our theory that
equivalent relations are identical; we cannot prove (Fn)(Gn)(Xl)·· . (x n )
((F"x 1 ... Xn = G"X 1 ... Xn) ~ F" = Gn). So it does not follow from the fact
that being a rational animal and being a featherless biped are exemplified
by exactly the same objects that these two properties are identical.
We should also note that there are two senses of "F is a equivalent to
G" when F and G are properties. One sense is that F and G are exemplified
by the same objects. The second sense is that F and G are encoded by
the same objects. We have stipulated that properties equivalent in the
latter sense are identical. In what follows, we always use "equivalent" in
the former sense.
We call the slightly modified second order language, together with its
semantics and logic, the object calculus (with complex relation terms). The
object calculus is the formal system in which the proper axioms of the
elementary theory of abstract objects are stated.

4. THE PROPER AXIOMS 15

We have now embedded our primitive metaphysical notions in the atomic

formulas of the language and embedded our primitive logical notions in
the complex formulas and terms of the language. To state the theory of
abstract objects, we shall also need to use our two primitive theoretical
relations, existence and E-identity. The theory has four axioms, two of
which are schemata and which involve defined notions. Though these
axioms are not logically true, we nevertheless suppose them to be true A
PRIORI. The first two axioms are non-schematic and express truths about
existing objects. The first schema utilizes all the defined notions of identity
constructed in both Sections 3 and 4 and tells us about the behavior of
 ELEMENT AR Y OBJECT THEORY 33

any entities which satisfy the definitions. The second schema tells us what
abstract objects there are. Since the schemata indicate that all sentences
of a certain form are to be axioms, we end up with a denumerably infinite
number of proper axioms. These axioms, plus the definitions in terms of
which they are stated, constitute the first principles of the elementary
theory of abstract objects.
The first axiom tells us that two objects bear the identityE relation to
one another iff they both exist and exemplify the same properties :16

AXIOM 1. ("E-IDENTITY"): X=EY== E!x&E!y&(F)(Fx== Fy).

The second axiom tells us that no existing objects encode properties:

AXIOM 2. ("NO-CODER"): E!x-+ ~(3F)xF.

The theory does not assert that there are any existing objects. Instead,
these first two axioms are meant to capture natural assumptions we make
about existing objects, should there be any.
In a sense, our first axiom tells us the conditions under which existing
objects are identical. Recall tha1 Dl (Section 1) says that abstract objects
are objects which exemplify the property of non-existence. Since this
partitions the domain of objects into disjoint classes, the following
definition is a completely general definition of object identity:
D4 X=Y =dJX=EY v (A!x&A!y&(F)(xF==yF)).

E-IDENTITY, D 2 , D 3 , and D4 allow us to prove one of the laws of

identity as a theorem schema:

THEOREMS ("IDENTITY INTRODUCTION"): rJ. = rJ., where rJ. is any

variable.
Proof. If rJ. is an object variable x and E!x, then since we have
(F)(Fx == Fx) from propositional logic and UI, we may use E-IDENTITY
to prove x = EX. SO X = x, by D 4. If ~ E lx, then x is abstract and similar
techniques get us the right hand disjunction of D 4 . If rJ. is a one-place
property variable F\ we easily get (x)(xF 1 == XF1). So by Db Fl = Fl. And
a generalized version of this procedure gets us F n = Fn. ~

In what follows, we abbreviate "IDENTITY INTRODUCTION" as

"==I."
34 CHAPTER I

We may complete the presentation of our THEORY of identity by

introducing the third axiom of the theory of abstract objects. Since all of
the definienda in D2 , D 3 , and D4 have the form a = /3, we assert that the
following axiom is true:

AXIOM 3. ("IDENTITY"): rx=fJ ~ (c/>(rx,rx)==c/>(rx,fJ)), where c/>(rx,fJ) is

the result of replacing some, but not necessarily all, free occurrences of a
by /3 in </J(a, a), provided f3 is substitutable for a in the occurrences of a it
replaces. 1 7

The rule of inference derivable from ~ E and IDENTITY is called identity

elimination (" = E").
The schema for abstract objects generates the most important set of
axioms of the theory. In effect, the schema guarantees that for every
expressible set of properties, there is an abstract object which encodes just
the members of the set. 1S However, the schema does this without a
commitment to sets. We generally use open formulas with one free property
variable with this axiom, though sentences will work as well (they express
vacuous conditions). Metalinguistically, it is legitimate to talk about the
set of properties satisfying a given condition, but in the object language,
our schema says something more like: for every condition on properties,
there is an abstract object which encodes just the properties which meet
the condition :19

AXIOM(S) 4. ("A-OBJECTS"): for any formula </J where x is not free, the
following is an axiom:
(:lx)(A!x & (F)(xF == </J)).
Some examples will help. If we let "F = R v F = S" be our formula </J, and
suppose that "R" denotes roundness and "S" denotes squareness, then
our axiom guarantees that there is a "round square" as follows:
(:lx)(A!x&(F)(xF==F=R v F=S)).
Suppose ao is such an object. It is easy to see that ao must be unique.
For suppose some other distinct abstract object, say a 1 , encoded exactly
roundness and squareness. By D 4 , it would follow that either a 1 encoded
a property ao did not, or vice versa, contrary to hypothesis.
Let us use the standard notation "(:I !x)IjJ" ("there is a unique x such
that 1jJ") to abbreviate (:lx)(IjJ&(y)(IjJ~~y=x)). Then, given D 4 , the
following is a consequence of A-OBJECTS:
 ELEMENTARY OBJECT THEORY 35

THEOREM(S) ("UNIQUENESS"): for any formula <p where x is not

free, the following is a theorem:
(3 !x)(A!x & (F)(xF == cjJ)).
Proof. An arbitrary instance of A-OBJECTS asserts that there is
an abstract object which encodes EXACTLY the properties which satisfy the
given formula. But there could not be distinct such objects, since distinct
abstract objects must differ with respect to at least one of the properties
they encode. ~

Another instance of the schema for objects says that there is an "existent
golden mountain". Suppose "G" denotes goldenness and "M" denotes
mountainhood. We then have:
(3x)(A!x & (F)(xF =-F=G v F=M v F=E!)).
It follows that there is an abstract object which encodes a property it fails
to exemplify. It is a contingent fact that there does not exist an object
which exemplifies all the properties that this object encodes.
By letting cjJ = r F=I=F', we obtain the empty object - it fails to encode
any properties. By letting <p = r F = F', we obtain the universal object - it
encodes every property.
Suppose "as" denotes Socrates. Then the following instance of A-
OBJECTS yields an A-object which encodes exactly the properties
Socrates exemplifies:
(3x)(A!x & (F)(xF =- Fa s )).
We might call this object Socrates' blueprint, and call Socrates the
correlate of the blueprint. 20 We define these terms as follows: 21
Ds x is the blueprint of y and y is the correlate of x ("Blue (x,y)"
and "Cor(y, x)") = dJ(F)(xF =- Fy).
A-OBJECTS guarantees that every object, existing or abstract, has a
unique blueprint:
(y)(3 !x)(A!x & (F)(xF =- Fy)).
This follows by UI on the instance of UNIQUENESS which results when
"Fy" is the formula <p.
Given any object b, A-OBJECTS yields an object which encodes all the
properties b fails to exemplify. Given any two objects band c, A-OBJECTS
yields an object which encodes (1) just the properties band c have in
36 CHAPTER I

common, (2) just the properties exemplified by either b or c, and (3) just
the relational properties b has with respect to c. This last object is yielded
by the following instance:
(3x)(A!x&(F)(xF == (3G 2 )((Gbc&F= [AxGXC]) v

(Gcb & F = [Ax GcxJ)))).

These examples give one a pretty good idea of what A-OBJECTS says.
We use A-OBJECTS to justify the definition we have proposed for
property identity. Suppose that instead of defining identity between
properties as in D 2 , we had added primitive identity formulas between
property terms. Then the following would have been a consequence of
A-OBJECTS:
(G)(H)((x)(xG == xH) ~ G=H).
To see this, suppose arbitrary properties P and Q were encoded by exactly
the same objects, but were distinct. By A-OBJECTS, it would have followed
that there is an object which encodes just P, without encoding Q, contrary
to hypothesis.
Since we would have had this consequence had property identity been
primitive, there was every reason to just define identity among properties.
Semantically, our definition ensures that two properties which have the
same extensionsJ are identical. But as we have seen our theory does not
commit us to the view that properties exemplified by the same objects
(i.e., which have the same extensionY/) are identical. An overriding reason
for choosing the style of semantics we have employed is that properties
and relations are not identified with their extensionsY/. The semantics does
not force upon us a view to which the theory is not committed.
E-IDENTITY, NO-CODER, IDENTITY, and A-OBJECTS are
jointly called the elementary theory of abstract objects. Good evidence for
thinking that the theory is consistent may be found in Appendix A, parts A
and B, where the reader will find a full discussion of the solutions to the
paradoxes which have been avoided as well as a model in Zermelo-
Fraenkel set theory of the monadic portion of our theory. One cannot
just model A-objects as sets of ordinary properties. That is because sets
of ordinary properties cannot exemplify the very same properties which
serve as their elements. That would be a violation of type, just as in ZF, no
set of sets can be an element of one of its members. But in object theory, A-
objects may exemplify the very same properties which they encode. For
 £LIlM£NTARY ODJ£CT TH£OR Y 37
example, any A-object which encodes [Ax ~ E!x] also exemplifies this
property.22 And in the next section, we will try to harness an entire range
of intuitions about which other properties A-objects exemplify. These will
be properties drawn from the same stock of properties which A-objects
encode.
Before we turn to this next section, there is an interesting consequence
of the theory of which the reader should be warned. Some complex relations
do not have unique constituents. Specifically, it is provable, given any
two place relation R, that for some objects a 3 and a4 , a3 =/= a4 , that
[AyRya 3 ] = [AyRya 4 J. Here is how:
Let R be any two place relation. By A-OBJECTS, (3x)(A!x & (F)(xF ==
(3u)(F = [AY Ryu] & ~ uF))). Call this object a3 and suppose that a3 fails
to encode [AY Rya 3 J. So by definition of a3, ~ (3U)([AY Rya 3J = [AY RyuJ &
~ u[AyRya 3J). That is, (u)([AyRya 3J = [AyRyuJ --+ u[AyRya 3 J). So by
instantiating to a 3 , it follows that a3 does encode [AY Rya 3J, contrary
to hypothesis. So suppose a3 encodes [AY Rya 3 J. Then (3U)(AY Rya 3 J =
[AyRyuJ&~u[AyRya3J). Call such an object a4 • So [AyRya 3J=
[AyRya 4 J, but a3 =/=a4 , since a3 encodes [AyRya 3J and a4 does not. Qg23
The reader is asked to postpone judgement about the seriousness of
this result until after the applications have been considered.

5. AN AUXILIARY HYPOTHESIS

In Chapter II, we shall put the theory we've now formulated to work.
For these applications, we add to our primitive vocabulary abbreviations
of the "gerundive versions" of standard English transitive verbs, intransi-
tive verbs, predicate adjectives, and predicate nouns. 24 By the "gerundive
version" of these words, I mean the phrases constructed out of English
gerunds which can appear in the subject places or direct object places of
English sentences. Here are some examples:

A. Transitive verbs Gerundive Version

kick (the) kicking (relation)
worship (the) worshipping (relation)
hate (the) hating (relation)
B. Intransitive verbs
run (the property of) running
walk (the property of) walking
38 CHAPTER I

C. Predicate adjectives
red (the property of) being red
courageous (the property of) being courageous
happy (the property of) being happy
D. Predicate nouns
horse (the property of) being a horse
person (the property of) being a person
building (the property of) being a building

We abbreviate these gerundive versions in our language with single letters,

appropriately chosen. For example, "K" might abbreviate "the kicking
relation", etc. (Also, we shorten our readings of A-expressions so that they
more closely resemble their natural language counterparts. So instead of
reading "b[Ax ex & Px]" as "b encodes being an object x such that x
exemplifies being courageous and x exemplifies being a person", we read
it as "b encodes being a courageous person".)
These additions to our primitive vocabulary are supposed to reveal our
pretheoretic conceptions about what simple properties and relations there
are. By adding these properties and relations to our system, A-OBJECTS
provides us with a rich variety of abstract objects which encode familiar
sorts of simple and complex properties.
These additions also make it possible to state an auxiliary hypothesis
of the elementary theory - an hypothesis to which we shall appeal on
occasion in the applications. Despite its rather vague character, it grounds
a wide range of intuitions some of us may share about abstract objects.
Pretheoretically, we have a pretty good idea of what properties existing
objects exemplify. And the theory tells us the conditions under which both
existing and abstract objects encode these properties. But other than being
abstract (i.e., [h ~ E !x]), we have not said anything about which
properties abstract objects exemplify.
Some of us may share the following intuitions. Abstract objects DO
NOT EXEMPLIFY the following properties: being round, having a shape, being
red, having a color, being large, having a size, being soft, having a texture,
having mass, having spatio-temporallocation, being visible, being capable
of thought (this is NOT to say that they are not thought of), being capable
of feeling, etc. In addition, it might seem that no two abstract objects
could ever meet each other, kick each other, kiss each other, etc. I am
sure the reader can provide many more examples. If these intuitions are
 ELEMENTARY OBJECT THEORY 39

true, then by A-EQUIVALENCE, A-objects exemplify the negations of

these properties and relations. I think we have intuitions to this effect as
well.
These properties and relations are "ordinary" properties and relations
of existing objects. Mally, Meinong, Findlay, Parsons, and others call
them "nuclear" relations (Routley calls them "characterizing" relations).
They are to be distinguished from "extranuclear" relations such as being
abstract, being thought about, being written about, being worshipped,
being more famous than, etc. Note that many of these are "intentional"
relations. We can easily imagine that abstract objects exemplify these
extranuclear relations. 25
We shall not pursue this distinction among relations in any detail. We
mention it because there will be occasion to appeal to the above intuitions
and it would be nice to ground them all in some general principle.
Consequently, we suppose that no abstract objects exemplify nuclear
relations.
We incorporate this hypothesis into elementary object theory by
assuming that we can somehow characterize nuclearity in our language. 26
Suppose that for each n, n~ 1, "Nuclear (F")" is some open formula with
one free n-place relation variable Fn. We then stipulate:

AUXILIARY HYPOTHESIS: (F")(XI)'" (xn)(A !XI & ... & A !x n & Nu-
clear (P) ~ ~ F"x i ... xJ

We trouble the reader with this hypothesis because it seem likely that
some such set of truths like these govern abstract objects A PRIORI.
 CHAPTER II

APPLICATIONS OF THE ELEMENTARY THEOR Y

It wiil be important to have definite descriptions (which involve the

primitive logical notion the) to facilitate the following applications. Let
us stipulate that where 4> is any formula with one free x-variable, (IX)4>
("the object x such that 4>") is to be a complex object term of our language.
Some examples might be: (ly)(E!y & Typ) ("the object which exists and
taught Plato") and (IX) (A !x&(F)(xF == F=R v F=S)) ("the object
which encodes just roundness and squareness"). Semantically, we interpret
descriptions like (IX)4> as denoting the unique object which satisfies 4>, if
there is one, and as not denoting anything if there is not one. To guarantee
that descriptions work in our system just as we would expect them to
A PRIORI, we add a proper axiom schema which asserts that atomic formulas
or defined identity formulas 1f; in which there occurs a description (IX)4>
are true iff there is a unique object satisfying 4> and there is something
which satisfies both 4> and 1f;.1

DESCRIPTIONS: where 1f; is any atomic formula or defined object

identity formula with one free object variable v, the following is a proper
axiom:
1f;~'x)t/> == (:3 !y)4>~ & (:3y)( 4>~ & 1f;~).

Here are a few examples:

E!(IX)Txp == (:3 !y)Txy & (:3y)(Txy & E !y)

b=(IX)PX == (:3 !y)Py&(:3y)(Py&b = y).

The first of these might say: the teacher of Plato exists iff there is a unique
teacher of Plato and something which is a teacher of Plato exists. The
second says: object b is identical to the object exemplifying P iff there's a
unique thing exemplifying P and something exemplifying P is identical with
b. Given DESCRIPTIONS, we may easily derive instances of the following
schema: r=(lx)4>-+4>~(let1f; ='r=v'; then 1f;~X)t/>='r=vx)4>', which is
the left side of DESCRIPTIONS).
There are modifications and restrictions that must be incorporated into

40
 APPLICATIONS OF THE ELEMENTARY THEORY 41

the definitions of Chapter I in order to accommodate terms that might fail

to denote. A detailed description of these appears in the Appendix to this
chapter and interested readers are directed there. For the most part, these
modifications and restrictions will not affect what follows, since we shall
utilize only descriptions which provably have denotations. So the above
discussion is all that is necessary for understanding the following
applications.

1. MODELLING PLATO'S FORMS 2

In this section, we construe certain assertions by Plato as consequences

of the theory. Most philosophers today regard Plato's Forms as first level
properties of some sort and view participation as just exemplification. But
this view of Plato from within Russellian background theory turns Plato's
major principle about the Forms into a triviality.
Plato's major principle about the Forms is the One Over the Many
Principle. It is stated principally in Parmenides (132a).3 The following
characterization is, I think, a faithful one:

(OMP) If there are two distinct F-things, then there is a Form of F

in which they both participate.
According to the orthodox view,
The Form of F =dfF
x participates in F = dfFx.
So translating (OMP) into a standard second order predicate calculus,
we would get:
x =1= y&Fx&Fy~(jG)(G=F &Gx& Gy).

But the consequent of this conditional just follows from the antecedent
by existential introduction. Clearly, we do not want to attribute such a
triviality to Plato. 4 Yet it is difficult to conceive of it as an interesting
metaphysical truth from within the Russellian framework.
In object theory, however, we may think of Forms as just a special kind
of A-object. When (OMP) is translated into our language, it turns out to
be an interesting theorem. To see this, consider the following series of
definitions and proofs: 5
D6 x is a Form of G ("Form (x, G)") = dfA !x&(F)(xF == F = G).
42 CHAPTER II

So a Form of G is any abstract object which encodes just G. So we have:

THEOREM 1. (G)(3x)Form(x,G).
Proof. By A-OBJECTS.
In fact, given UNIQUENESS (I., Section 4), it also follows that:
THEOREM 2. (G)(3!x)Form(x,G).

Given Theorem 2, we know that the description (lx)(A!x & (F)(xF ==

F = G)) ("the Form of G") always has a denotation. For convenience, let
us use "cJ)G" to abbreviate (lx)(A!x & (F)(xF == F = G)). We then have as a
simple consequence of DESCRIPTIONS:

THEOREM 3. cJ)GG ("The Form of G encodes G").

Proof By DESCRIPTIONS, cJ)GG == (3 !y)(A !y&(F)(yF == F=G)) &
(3y)(A!y & (F)(yF == F = G) & yG). The right side of this biconditional is
easily obtainable from Theorem 2. ~

Now we can define participation:

D7 Y participates in x ("Part(y, x)") = dJ(3F)(xF & Fy).
So something participates in the Form of G just in case there's a property
the Form encodes which the object exemplifies. All objects which exemplify
redness participate in the Form of Redness ("cJ)R").
These definitions validate (OMP). The translation of (OMP) into our
language turns out to be a theorem:

THEOREM 4. x=/=y & Fx & Fy~(3u)(U=<DF & Part(x,u) & Part(y,u)).

Proof Assume a =1= b, Pa, and Pb, where a,b are arbitrary objects and
P is an arbitrary property. By = I, we have <Dp=cJ)p.6 By Theorem 3 and
the above assumptions, we have cJ)pP & Pa. So (3G) (cJ)pG& Ga), i.e., Part
(a, <Dp). By the same reasoning, Part(b,cJ)p). So cJ)p=cJ)p & Part(a,cJ)p) &
Part(b,cJ)p). So, (3u)(u=cJ)p&Part(a,u) and Part(b,u)). ~

Another theorem quickly falls out of these definitions:

THEOREM 5. Fx == Part (x, cJ)F)'

Proof. (~) Assume Fx. By Theorem 3, Part (x, cJ)F)' (+-) Assume
 APPLICATIONS OF THE ELEMENTARY THEORY 43

Part (x, <l'>F). Call the property <l'>F encodes and x exemplifies, G. Since <l'>F
encodes just F, it must be that G = F. So Fx. IXI

So in our system, the notions of exemplification and participation are

distinct (unlike the orthodox view) though nonetheless equivalent. This
should preserve at least some of the intuitions of orthodox theorists.
On our theory, some Forms participate in other Forms, and indeed,
some Forms participate in themselves. Consider the property [Ax '" E!x]
("E!"). Let us call this property: Platonic existence. Since all A-objects fail
to exist, they all exemplify Platonic existence. In particular, we have:

THEOREM 6. (x)(r3F)(x=<I'>F)~E!x).

So the Forms exemplify a kind of existence which is different from the

existence exemplified by actual objects. 7 But now consider <l'>E!, which we
may call Platonic Being, or Reality. From Theorems 5 and 6 it follows
that:

So all Forms participate in Platonic Being.8 In particular, <l'>E! participates

in itself, justifying our claim that some Forms participate in themselves.
To reach this conclusion, we might also have used the AUXILIARY
HYPOTHESIS and the assumption that being blue, for example, is a
nuclear property. It would follow that all A-objects fail to exemplify this
property. So all A-objects would exemplify [Ax '" Bx]("B"), where "B"
denotes being blue. Then by Theorem 5, all Forms participate in <1'>13. So
would <1'>13.
Consider now the Third Man Argument. This is a puzzle which
commentators say Plato produces in the Par men ides (132affV The puzzle
is that several of Plato's principles about the Forms seem to be jointly
inconsistent. We have seen two of these principles: (OMP) (Theorem 4)
and the Uniqueness Principle (Theorem 2). There are two others: the
Self-Predication Principle and the Non-Identity Principle:
(SP) The Form of F is F
(NI) If something participates in the Form of F, then it is not
identical with that Form.
44 CHAPTER II

We can prove a contradiction if we assume that there are two distinct

F-things x and y. By (OMP), there is a Form of F in which x and y
participate. By (SP), the Form of F is an F-thing. By (NI), it is distinct
from x and y. But then, (OMP) guarantees that there is another Form of
F in which x and the first Form participate. Then (NI) yields the conclusion
that the latter Form must be distinct from the first. But this violates the
uniqueness principle, which says that the Form of F is unique.
On the theory we have presented, (NI) must be false. We can derive its
negation as a theorem:

THEOREM 8. ~ (x)(Part (x, <1>F) ~ x +- <1>F).

Proof. Consider <1>£!. IX!

So by rejecting (NI), we dissolve the puzzle.

However, it is worthwhile to examine (SP). If we translate (SP) into our
language as <1>F exemplifies F, then it must be false. This time, consider
<1>E!. But if we translate (SP) into our language as <1>F encodes F, then it
turns out to be Theorem 3. Does the word "is" in the (SP) principle mesh
the distinction between exemplifying and encoding a property?
Of course we cannot generalize on this one example, but we can look
for further evidence for thinking that the "is" of English is ambiguous.
Maybe we have an option of translating a sentence involving the pre-
dicative "is" as either an exemplification or an encoding formula. And in
case there is such an ambiguity, let us now stipulate that whenever we
use the word "is" in its predicative sense in what follows, we shall mean
"exemplifies". 1 0
We may conclude, with respect to the Third Man Argument, that our
theory rules that (OMP) and (U) (Uniqueness Principle) are true, that
(NI) is false, and that (SP) has a true reading and a false one. Since we
abandon the (NI) principle, further research should be directed toward
the question of how deeply Plato was committed to it.
Finally, we discuss the Sophist. The four assertions by Plato
in that work that we discuss are ones which, taken together, are somewhat
mysterious. Many scholars regard Plato's theory of Forms as his attempt
to reconcile two major philosophical schools of thought. The first was the
school of Parmenides, founded as the view that the world had to be
considered as a whole without parts, without motion and change, and
without generation and decay. The opposing school (Thales, Anaximander,
Anaximenes, Heraclitus, Empedocles, and the Atomists) denied this and
 APPLICATIONS OF THE ELEMENTARY THEORY 45

attempted to isolate the elementary parts of the world, the interaction of

which was responsible for motion, change, generation, and decay. Plato's
Forms were entities he postulated to capture certain truths of the
Parmenidean school - they were changeless, motionless, and eternal. Yet
Plato allowed that there were ordinary objects which moved, changed,
came into being, and passed away. But, apparently, he supposed them to
have a lesser degree of reality.
Plato's attempt to capture the Parmenidean truths was not completely
successful. Some Forms gave him trouble, especially the ones which
reflected some of the more mundane things in the world. He could never
quite accept the fact that there were Forms with respect to hair, dirt, or
mud. And the Form of motion - did it move? If so, how could it remain
a Form? Forms were supposed to be motionless. Given the (SP) principle,
how could there be a real Form of Motion if it did not move? And how
do the Forms of Motion and Rest interact with each other?
In this context, the following four assertions by Plato in the Sophist
seem mysterious:
(1) Rest and Motion are completely opposed to one another (250a).
(2) Rest and Motion are real (250a).
(3) Reality must be some third thing (250b).
(4) In virute of its own nature, then, reality is neither at rest
nor in movement (250c).
To analyze these assertions, we need the following definitions and
(reasonable) assumptions, where "M" denotes being in motion.
Ds Being at rest ("R") = dJ[AX ~ Mx]
Ai Nuclear(M)
A2 M=f=R&M=f=E!&R=f=E!.
(1) may be interpreted as a true statement about the Forms.
Consider (1a):
(la) (x)(Part (x, <D M) == ~ Part (x, <DR))'
This is provable, given D s, A-EQUIVALENCE, and Theorem 5. That is,
by Ds and A-EQUIV, something exemplifies being in motion iff it fails to
exemplify being at rest. So by Theorem 5, something participates in <DM
iff it fails to participate in <DR'
46 CHAPTER II

There is also an uncharitable way to interpret (1) as a statement about

the Forms. Consider (1b):
(1b) ~ RI1>M & ~ MI1>R ("The Form of Motion does not exemplify
being at rest and the Form of Rest does not exemplify being
in motion").
This is false, since by AbDS' the AUXILIARY HYPOTHESIS, and
A-EQUIV, the Form of Motion does exemplify the negation of a nuclear
property.
Consider (2), "Rest and motion are real". (2a) seems to be a good
candidate for translating it:
(2a) E !I1>R & E !I1>M ("The Forms of Rest and Motion exemplify
Platonic existence").
(2a) is a theorem. We also know that both I1>R and I1>M participate in I1>E!'
If we define "blend with" as "participate in", we get that both of these
Forms blend with Being or Reality. (3) could be read as (3a):
(3a) I1>R of I1>E! & I1>Mof 11>£!.
This is provable from assumption A 2 •
Finally, we consider (4), "In virtue of its own nature, reality is neither
at rest nor in motion". (4) is another example of a sentence which turns
out false when we read the copula "is" as exemplification and true when
read as encoding. Consider (4a):
(4a) ~ RI1>E! & ~ MI1>E!'
Since we have defined Platonic Being, or Reality, as I1>tb (4a) captures
(4) when "is" is read as "exemplifies". (4a) is false since I1>t! exemplifies
being at rest. But consider (4b):
(4b) ~ I1>E!R & ~ I1>E!M.

The key to seeing that this might be right comes from the following
definition:
D9 The nature of I1>F = dfF.

The nature of a Form is the property it encodes. Thus, we read "in virtue
of its own nature" as a clue to thinking that Plato is going to conclude
something about the fact that E! is central to the identity of 11>£!.
 APPLICA nONS OF THE ELEMENTARY THEORY 47

Assumption A2 tells us that the nature of <DE! is distinct from the natures
of <DR and <D M. SO (4b) is derivable.
Assertion (4) has always been rather puzzling to me, and I think it is
interesting that the distinction between exemplifying and encoding a
property has helped us to find a true reading for it.
Is there, after all, some unity to the history of philosophy? Do we have
here a PRIMA FACIE link between Plato, Meinong, Mally, and the theory
of abstract objects? Maybe further investigations along the above lines will
help us to answer these questions.

2. MODELLING THE ROUND SQUARE, ETC.

In our first encounter translating certain theoretical statements of natural

language into the language of the theory, we discovered that a few of them
containing the copula "is" turned out true when translated using an
encoding formula yet turned out false when translated using an exempli-
fication formula. In this section, we look at a class of English sentences
which exhibit this feature. These sentences can be recognized by the facts
that: (1) they have the form "The F 1,F2, ... ,Fn is F/, (1 :::;i:::;n), and
(2) there is not (or could not be) an object which jointly exemplifies
F l' F 2' ... , Fn" Here are some examples:

(1) The set of sets which are not members of themselves is a set
of sets which are not members of themselves (The F is F).
(2) The round square is round (The F, G is F).
(3) The existent golden mountain is existent (The F, G, H is F).

These sentences seem to be true A PRIORI. But if we translate the

description in (2), for example, as "the object which exemplifies roundness
and squareness", then the description would fail to denote. It would then
be hard to see how to account for the intuitive truth value of the sentence.
And similar remarks apply to (1) and (3).
However, if we translate the description in (2) as "the object which
encodes just roundness and squareness", and read the "is" as "encodes",
we end up with the truth: the object which encodes just roundness and
squareness encodes roundness. In a similar manner, we read (1) as: the
object which encodes just being a set of sets which are not self-members
48 CHAPTER II

encodes being a set of sets which are not self-members. And we do

something similar for (3). The suggestion, then, is to translate 'The F 1 , ... ,Fn
is F;" as "the object which encodes just F 1, ... ,Fn encodes F/'.
To make this suggestion precise, we must focus on an interesting class
of descriptions. These are descriptions of the form: (lx)(A!x & (F)(xF == X)).
We call this class of descriptions A-object descriptions, and the reason
they are interesting is that whenever X is a formula with no free x's, the
resulting description always has a denotation. This is a consequence of
the UNIQUENESS theorem schema for objects. In fact, UNIQUENESS
and DESCRIPTIONS allow us to prove an interesting set of theorems
governing the A-object descriptions:

THEOREM(S) ("A-DESCRIPTIONS"): xi' == (lx){A!x & (F)(xF == X))G.

Proof. (-+) Suppose G satisfies X. By UNIQUENESS, there is a unique
A-object, say b, which encodes exactly the properties which satisfy X. So
b encodes G. So there is a unique A-object which encodes exactly the
properties satisfying X and something which encodes exactly the properties
satisfying X also encodes G. So by DESCRIPTIONS, the A-object which
encodes exactly the properties which satisfy X encodes G. 11 (~) By
reversing the reasoning. I8l

Using this theorem schema, it now becomes possible to prove certain

facts regarding the objects denoted by A-descriptions. Consider (lx)(A!x &
(F)(xF== F=R v F=S), where "R" denotes roundness and "s" denotes
squareness. If we let X=(F=R v F=S), then X: and X~. So by A-
DESCRIPTIONS, (IX) (A !x & (F)(xF == F = R v F = S)) encodes both R
and S, as we might have expected. In general, when X=(G=F1 v G=F 2
v ... V G=Fn), it is provable that:
(lx)(A!x & (G)(xG == G=F 1 v ... v G=Fn))Fi'
where 1 :s; i:S; n, given A-DESCRIPTIONS.
This provides us with the key to the proper translation of our data.
For simplicity, let us shorten A-descriptions by using restricted variables
to range over A-objects. In fact, throughout the remainder of this work,
we use z-variables to range over A-objects. 12 So our A-object descriptions
now have the form: (lz)(F)(zF == X). Where "s" denotes being a set, "E"
denotes the membership relation, and where the other abbreviations are
obvious, we may translate the descriptions in (1)-(3) as (a)-(c), respectively:
 APPLICATIONS OF THE ELEMENTARY THEORY 49

(a) (lz)(F)(zF == F = [Ax Sx & (Y)(YEX == Sy & y¢y)])13

(b) (lz)(F)(zF==F=R v F=S)
(c) (lz)(F)(zF == F=E! v F= G v F=M).
And in general, the descriptions in the class of English sentences we have
singled out are to be translated as:
(IZ)(G)(zG== G=F 1 v G=F 2 v ... v G=Fn).
In the metalanguage, we will signal the fact that we intend this reading
of the English definite article by writing "the",,".
Now let (IZ)1jJ 1, (IZ)1jJ 2, and {IZ)1jJ 3 abbreviate the descriptions in (a)-(c),
respectively. We then translate (1)-(3) into our language as (1)'-(3)',
respectively:
(1)' (lZ)1jJ 1 [Ax Sx & (Y)(YEX == Sy & y¢y)]
(2)' (lZ)1jJ 2R
(3)' (lZ)1jJ3E!
(1)' -(3)' are all theorems, hence the A PRIORI character of the English. In
!Igu_eraL,ollLtranslation of "The F 1 " •• ,Fn is F/', where there is not (or
could not be) an object which jointly exemplifies F 1, ... ,Fm will always
be a theorem of the following form:
(IZ)(G)(zG == G=F 1 v ... v G=Fn)Fi.
There is a closely related use of the English definite article. Here are some
examples:
(4) The even prime number greater than two is not odd.
(5) The set of all non-self-membered sets is a set.
(6) The existent golden mountain has a shape.
(7) Necessarily, the teacher of Aristotle is a teacher.
These sentences will be represented with the help of a slightly modified
A-description. "The even prime number greater than two" shall be
translated as "the A-object which encodes being an even prime number
greater than two or any property implied by this property". To represent
and interpret this reading of the definite article, we must define "F implies
50 CHAPTER II

G" as necessarily, everything exemplifying F exemplifies G. So we postpone

further investigation until the modal theory has been developed.

3. THE PROBLEM OF EXISTENCE

The property of existence has puzzled philosophers for years. The assertion
that some particular thing fails to exemplify existence (or being) strangely
carries with it a commitment to the existence (or being) of the very thing
which serves as the subject of the assertion. This is partly a result of trying
to keep the theory of language as simple as possible - we try to account
for the truth of a simple sentence by supposing that the objects denoted
by the object terms are in an extension of the relation denoted by the
relation term. But when we have a true non-existence claim, talk about
"the object denoted by the object name" seems illegitimate.
Although the theory we have developed is rather flexible on this issue,
our discussion of the matter will be slightly complicated by the fact that
we have taken "existence" as a primitive theoretical notion. The reason
this may confuse things is that this primitive notion is not necessarily the
notion to use to translate the English word "exists", as it occurs in the
data. To see this, consider first the fact that the following two sentences
are theorems:
(1) [Ay~E!Y](lZ)(F)(zF==F=GvF=M)

(2) ~ E !(lz)(F)(zF == F = G v F = M).

That is, it is provable both that the"", golden mountain exemplifies
non-existence and that it fails to exist. However, neither (1) nor (2) would
be an acceptable translation of (3):
(3) The golden mountain does not exist.
(3) has at least one reading on which it is contingent and not knowable
A PRIORI. SO that eliminates both (1) and (2) as acceptable translations.
However, an abstract object x can "exist" in the sense that some existing
object exemplifies all the properties x encodes:
DlO x exists 2 ("E! !x") = df(3y)(E!y & (F)(xF -+ Fy»
It cannot be known A PRIORI that thes1 golden mountain fails to exist z. That
is, we may read (3) as the contingent (3'):
 APPLICATIONS OF THE ELEMENTARY THEORY 51

(3') '" E! !(lz)(F)(zF == F = G v F = M)

The English word "exists" is therefore properly translated as "E! !".

Meinong could have truthfully responded to Russell that the"" existent
golden mountain is existent but doesn't existz .
It is important to realize that we were not forced to have theorems like (1)
and (2) above. We could have designed things so that it was provable that
everything whatsoever exists. Instead of taking existence as our primitive
theoretical relation, we could have started with the notion of being abstract
("A !"). We could have then defined:

x is concrete ("C !x") = df [)'Y '" A !yJx

x exists ("E !x") = dfA!x v C Ix.

We could have then revised NO-CODER as: C!x~ "'(:IF)xF. Finally,

we could have relabeled "= E" as "= 0" changed E- IDENTITY to C-
IDENTITY (i.e., x =cy == C!x & C!y & (F)(Fx == Fy», and redefined gen-
eral identity (i.e., X= Y =df X =cy v (A!x & A!y & (F)(xF == yF»). Leaving
A-OBJECTS as it stands, we could call the result of all these changes
VERSION 2.
On VERSION 2, it is provable that everything whatsoever exists.
VERSION 2 can do all the work the original theory can do. That is
because the exemplification/encoding distinction, and the distinction
between two types of objects, remain intact. On VERSION 2, we still
have to analyze (3) in a manner analogous to the above. To translate "the
golden mountain does not exist" properly, we have to suppose that "the
golden mountain" denotes the"" golden mountain and that the sentence
claims about this object that there are (exist) no concrete objects which
exemplify all the properties it encodes.
So the theory is really pretty flexible on the question are there objects
which fail to exist? It is a question of how you prefer to use the word
"exists". But I think that philosophers who insist that VERSION 2 is the
only correct version of the theory are mistaken. The theory remains useful
no matter which of the two versions you adopt. We have taken the present
course because it leaves us with a formal language which can be used to
investigate the claim that there is a distinction in natural language between
the quantifiers "there is" and "there exists". Some philosophers, myself
included, believe that there is an exploitable difference in meaing between
52 CHAPTER II

these two quantifiers of English. Our view can be made precise by

investigating a language in which this difference in meaning might be
represented. The language we have now is such a language. We use ''(:h)¢''
to express the fact that THERE IS an x such that ¢, and use "(:3x)(E!x & ¢)"
to express the fact that THERE EXISTS an x such that ¢. In a theory which
supposed that all the things there are exist, there is no natural way to do
this. But this is not an overriding reason for keeping things as they are.
Meinong claimed that "the Object as such stands beyond being and
non-being" and that "the Object is by nature indifferent to being".14 I
am not a Meinong scholar, so I do not suppose that I know what Meinong
meant by this "doctrine of aussersein", and I do not suppose that he had
these two versions of our theory before his mind when he said things like
this. Nevertheless, something like these cryptic utterances of Meinong are
relevant here. It just does not matter whether you conceive of A-objects
as existing or as failing to exist.
Maybe the word "exists" is an ambiguous word, one of the senses of
which is a property which has a negation that also turns out to be a sense
of the wordY To make this idea plausible, we could stick with the original
version of the theory, and read "E!" as "real existence" and "[)"x ~ E !x]"
as "Platonic existence". Now we have two kinds of existence, with A-objects
exemplifying the latter kind. This reading of [h ~ E!x], besides working
to our advantage in Section 1, is further justified by the facts that in the
modal theory which follows, A-objects end up having being in every
possible world and the class of A-objects stays fixed from world to world.
Platonic beings are necessary beings, and A-objects turn out to be necessary
beings. 16 They, therefore, exhibit a more perfect kind of existence. 17
So talking in terms of two kinds of existence is yet a third way of
approaching the problem of existence. This means that we really do not
have to commit ourselves on the question: Do A-objects fail to exist?
Three equivalent versions of the theory decide the question in different
ways. The version one prefers to go with will be mostly a result of a
decision about which of the various senses of the word "exist" one prefers
to use. 1S We shall use it to mean "having a location in space-time."

APPENDIX TO CHAPTER II

In this appendix, we describe the modifications and restrictions which

have to be incorporated in the system of Chapter I in order to
accommodate descriptions.
 APPLICATIONS OF THE ELEMENTARY THEORY 53

(A) The first thing to do is to revise the simultaneous recursive definition

offormula and n-place relation term so that the descriptions get generated
in a recursive clause defining new object terms. This is relatively
straightforward and the result will be very similar to the definition in III.,
Section 1, B.
(B) Next, we semantically interpret these descriptions by adding a clause
to I., Section 2, B., Denotations. The clause should read:

dJ./((IX)<P) ={o iff (3/')(/'xI & I'(x) =0 & I' satisfies <p &
(/")(/" xl & I" satisfies <p ~ I" = I'))
undefined, otherwise.

This guarantees that (IX)<p denotes the unique object 0 satisfying the
description, if there is one, and denotes nothing at all if there is not one.
Note that if a description fails to denote, the failure of denotation is
inherited by any complex term in which it occurs. So if "(IX)GX" does not
denote, neither will "[AY TY(IX)GX]" nor "(ly)TY(IX)GX".
(C) We must next prevent the base clauses in the definition of sa tisfaction
from being undefined. So I., Section 2, c., clauses (1) and (2) need to be
redesigned and should read as follows:
(1) If <p = pn01 ." On' I satisfies <p iff
(301)'" (:Jo nK3-t n)(01 = d J./(Ol) & ... & on = d J./(on)
&1," =d J,/(pn) & (01)'''' on> Eext&l!(1,")).
(2) If <p = Opl, I satisfies <p iff(30)(:J1,1)(0 = d J,/(o) &,z1 = d J,/ (pi)
& OEext", (-t 1)).
So if an assignment lis to satisfy an atomic formula, all the terms in the
formula must have a denotation. The other clauses in the definition of
satisfaction are acceptable as stated.
(D) Fourthly, we must modify one of our logical axioms so that we
may invalidate the following proof of a proper theorem which would be
false in some models of the theory:
(i) (X)(x=x) = I (Proper Theorem)
(ii) (IX)GX = (IX)GX LA4
(iii) (3y)(y= (IX)GX) EI.
Existential Introduction ("EI") is a typical derived rule of inference and
can be used to move from line (ii) to line (iii).19 Although line (iii)
54 CHAPTER II

abbreviates a longer formula, it will be false in models of the theory in

which (IX)GX fails to denote. The problem here is that LA4, as it now
stands, allows us to instantiate universals to terms which might fail to
denote. This gets us into trouble when a true universal claim gets converted,
by LA4, into a formula in which atomic formulas appear containing the
non-denoting term.
Let us say that a term 'i contains a definite description iff either 'i is a
definite description or a definite description occurs somewhere in 'i. For
example, the following three terms all contain descriptions: (IX)GX,
(IX) TX(ly)Hy, [Ay TY(IX)GX]. Now the only terms of our language that
might fail to denote in an arbitrary interpretation are those which con-
tain descriptions. All primitive names are guaranteed a denotation, since
the domains of interpretation must be non-empty and the ff function is
a total function from the set of primitive names into the appropriate
domains. Also, any A-expression which doesn't contain a description is
guaranteed a denotation. The A-expressions are partitioned and to each
equivalence class there corresponds a unique clause in the definition of
denotation J ./. So if each term in the A-expression has a denotation, the
logical functions in !f? guarantee that a relation with the appropriate
structure will be found in the domain of relations.
LA4 will never get us into trouble therefore if we require that only
terms which are guaranteed denotations (i.e. terms which do not contain
descriptions) may instantiate universal claims. But we also want to
instantiate universal claims to terms which contain descriptions whenever
we know such terms have a denotation. This happens whenever such
terms appear in true atomic formulas. So if a universal claim is true, then
if a term containing a description appears in a true atomic formula, the
result of instantiating that term into the universal claim should also be
true.
These remarks can be incorporated into our system by revising LA4
into the following two axiom schemata:
LA4a: (IX)¢ ....... ¢~, where 'i is substitutable for IX and 'i contains no
descriptions
LA4b: (IX)¢ ....... (ljIp ....... ¢~), where 'i is substitutable for both IX and [3,
and ljI is any atomic formula.
The reader may wonder here whether we have considered the option
standardly taken in predicate calculi containing descriptions. In a typical
 APPLICA nONS OF THE ELEMENT ARY THEORY 55

second order predicate calculus where identity formulas are primitive and
interpreted in the usual way, where the domains of interpretation must
be non-empty, where the usual two identity axioms have been added to
the logic, and where the only terms which might fail to denote are those
which contain descriptions, the normal way of invalidating inferences like
the above is to revise LA4 to the following two schemata:
where r is substitutable for!Y.
(!Y.)c/>~«3p)p=r~c/>~),
(3P)P =r, where r contains no descriptions.
Instances of the first would be logically true in our system, strangely
enough, despite the fact that the identity symbol is defined in various
ways. But some instances of the second would not be logically true.
Consider instances in which the quantified variable is an object variable,
for example, (3x)x = c. This abbreviates (3x)(x = EC v (A!x & A!c & (F)
(xF == cF))). Since = E denotes a primitive relation of the theory, consider
interpretations in which the proper axiom E-IDENTITY is false. For
example, consider an interpretation in which = E denotes an irreflexive
relation and c denotes an existing object. Why should (3x)x = EC be true
in that in terpreta tion ?
Consequently, this revision of LA4 will not help us, and we shall adopt
LA4a and LA4b as the official logical axioms of our system. It should be
easy to see that they are both logically true. LA4a blocks the undesirable
inferences in our system, while LA4b allows us to instantiate universal
claims with terms containing denoting descriptions. 2o Of course we can
always "instantiate" definitions with descriptions, even if the descriptions
fail to denote. Definitions are not universal claims. They are metalinguistic
conventions for abbreviations. Strictly speaking, only meta variables
ranging over the appropriate terms should be used in introducing the
definitions. However, we employ object language variables for convenience,
since it makes it easier to read the formulas.
(E) The next modification we need to make in order to successfully
incorporate descriptions concerns our A-EQUIVALENCE schema. We
must require that none of the A-expressions used in the schema are
constructed out of propositional formulas c/> in which there occur
descriptions. This prevents the following derivation of a formula which
is not valid:
(i) (x)([AY FY(IU)GU v ~ FY(lU)GUJX == FX(IU)GU v ~ FX(lU)GU)
by A-EQUIVALENCE
56 CHAPTER II

(ii) [Ay FY(lU)GU V ~ FY(lu)Gu]a == Fa(lU)Gu V ~ Fa(lU)Gu

by LA4a.
(ii) will be false in interpretations where "(lU)GU" fails to denote, since the
right side of the biconditional would be true while the left side false. That
is because the failure of the description to denote is inherited by the
A-expression. But for the atomic formula constituting the left side of the
biconditional to be true, both the object term and the complex relation
term must have a denotation.
Consequently, A-EQUIVALENCE should be reformulated as follows:

A-EQUIVALENCE: where <P is any propositional formula with no

descriptions, the following is an axiom:
(Xl)··· (X n)([AV 1 ... Vn <p ]X1 ... Xn == <p~::::::~:).

Then, by using LA4b, we may construct the following derivation:

(i) (u)(x)([AyFyu V ~FyuJx==Fyu v ~Fyu)
A-EQUIVALENCE and UI
(ii) t/lvu)Gu--+([AY FY(lU)GU v ~ FY(lu)Gu]a == Fa(lu)Gu v
~ Fa(lu)Gu) by UE and LA4b, where t/I is atomic.
(F) Next, we investigate the logical axioms which must be added to the
logic if we are to be able to derive logical truths which arise specifically
as a result of the semantic interpretation of descriptions. The question is
complicated by the following two facts: (1) that identity is a primitive
logical notion in the semantics and crucially appears in the clause assigning
a denotation to descriptions (see part B, above), and (2) that identity is
not a primitive logical notion of the object language, and as we have it
defined, it works like it should only in the presence of the proper axioms
E-IDENTITY and IDENTITY. Were it not for these facts, we could have
just added an axiom like DESCRIPTIONS and be done with it. But, as
noted in note 1, DESCRIPTIONS is not logically true. And for the reasons
mentioned in that note, I do not think that any other formulas utilizing
defined identity which might prove useful for our logic are going to be
logically true. So which formulas without identity are the logical truths
specifically relating to descriptions from which all others can be derived?
Clearly the following two formulas are both logically true: H(lX)GX--+
(3y)(Gy & Hy) and (H(lX)GX & F(lX)GX)--+(3y)(Gy & Hy & Fy). So we
need to add the following general schema :21
 APPLICATIONS OF THE ELEMENTARY THEORY 57

L-DESCRIPTIONS 1 : where ljJ is any atomic formula or conjunction of

atomic formulas, the following is an axiom:
ljJ~X)<I> ~(3y)(<p~ & ljJ~).

But this schema, by itself, does not seem to be sufficient. We do not seem
to be able to derive H(IX)GX~(y)(Gy~Hy). So we probably need to add:

L-DESCRIPTIONS 2 : where ljJ is any atomic formula, the following is an

axiom:

But even this does not seem to be sufficient, since it does not look like
we will be able to derive H(IX)GX ~ ~ «3y)(Gy & Fy) & (3y)(Gy & ~ Fy)).
Again, it looks like the following schema should be added:

L-DESCRIPTIONS 3 : where ljJ is any atomic formula with Vi free and X is

any formula with V2 free, the following is an axiom:
ljJ~~)<I>~ ~«3y)(<p~ & X~,) & (3y)(<p~ & ~ X~,)).

Of course, only a completeness proof will verify that these three logical
axioms will be sufficient. I suspect, however, that there are still underivable
logical truths involving descriptions. Given our present concern with
metaphysics, I think that we may feel secure that our proper axiom
DESCRIPTIONS will allow us to prove all of the consequences of the
theory which we will need in the applications.
Logicians should be interested in the project of developing a complete
logic for descriptions in a language without identity. They may end up
with a rather inelegant group of logical axioms. But that would simply
be because the expressive power of the object language falls far short of
the expressive power of the language used in the semantics, which takes
identity as an extra primitive. But I prefer keeping the object language
elegant and complicating the logic to complicating the object language
with the addition of primitive identity. We must certainly be justified in
searching for the smallest set of primitives powerful enough to derive a
given set of interesting results. Identity is a concept which we can
ANALYZE in terms of our other primitives. I take the definitions of identity,
D 2 , D 3 , and D4 to be insightful. We no longer have to wonder what
philosophers and model-theoretic logicians mean when they appeal to this
58 CHAPTER II

notion. And if in the course of trying to define as much as possible in

terms of a few potent primitives we have to add several logical axioms to
guarantee completeness, so be it (there are also other reasons for not
adding identity as a primitive to the object language, for example, we would
have to place restrictions on A-formation and A-EQUIVALENCE, to avoid
McMichael's paradox (see Appendix A, part A)).
(G) We close the appendix with a few remarks on the proper axiom
DESCRIPTIONS. Note that it allows us to prove the important set of
proper theorems: t/I~X)<I> ~(3Y)(Y=(IX)¢), where t/I is any atomic formula
or any defined identity formula with one free object variable v. For if
t/I~!X)<I>, then by DESCRIPTIONS, (3!y)¢~. If we call an arbitrary such
object b, then we know ¢~&(u)(¢~~u=b). Now if we can show
b = (lX)¢, then since there are atomic formulas in the defined notation,
we may use EI to reap our result. 22 We prove b=(lX)¢ by using
DESCRIPTIONS, this time deriving the right side of the biconditional
with t/I as an identity formula. That is, we try to derive (3 !y)¢~ & (3y)(¢~
& b = y). We already have the first conjunct. The second is easily
obtained given that we know ¢~ and that b=b is introducable by =1.
So (3y)( ¢~ & b = y). So by DESCRIPTIONS, b = (zx)¢, and by EI,
(3y)(y = (IX)¢).
DESCRIPTIONS forces all of the models of the theory to be such that
objects which satisfy D4 (Chapter I, Section 4) are ident£·cat. 23 To see
this, consider the following instance: H(lX)GX == (3 !y)Gy & (3y)(Gy & Hy).
Whenever the right side is true, some object, say 0, exemplifies G and
everything else which exemplifies G is either E-identical with 0 or is abstract
and encodes the properties 0 encodes. Semantically, there may be some
didtinct object other than 0 which in fact does this. But in all the models
of the theory, the left side of DESCRIPTIONS will then force that object
to be ident£·cat to 0, given the way we have interpreted descriptions in
part B above.
The above remarks on incorporating descriptions should give the reader
a fairly good idea of how our system adjusts to the acquisition of terms
which may fail to denote.
 CHAPTER III

THE MODAL THEORY OF ABSTRACT OBJECTS

(WITH PROPOSITIONS)

1. THE LANGUAGE

A. PRIMITIVE SYMBOLS

To the language of Chapter I, we add the "D"-operator (to express the

English sentential adverb "necessarily") and names of (and variables
ranging over) propositions. By allowing the superscripts on the primitive
relation terms to reach zero, we obtain names and variables for
propositions. For convenience, we use pO,Qo,Ro, ... and FO,Go,Ho, ... as
names and variables, respectively, for propositions. Officially, however, our
new list of primitive symbols is as follows:
(1) Primitive object terms
Names: a l ,a 2 , •••
Variables: Xl' X 2 ,.·.
(2) Primitive n-place relation terms
Names: P~,P~, ... , =E,E! n20
Variables: F~, F~, ...
(3) Connectives: ~, ~
(4) Quantifier: '<j
(5) Lambda: A
(6) Iota: I
(7) Box: 0
(8) Parentheses and brackets: (,), [,].

B. FORMULAS AND TERMS

We simultaneously define (propositional) formula, object term, and n-place

relation term, inductively, as follows:
(1) All primitive object terms are object terms and all primitive
n-place relation terms are n-place relation terms.
(2) If pO is any zero-place relation term, pO is a (propositional)
formula.

59
60 CHAPTER III

(3) Atomic exemplification: If p" is any n-place relation term and

are any object terms, P"Ol ... on is a (propositional)
01' ... '0"
formula.
(4) Atomic encoding: If p1 is anyone-place relation term, and °
is any object term, Op1 is a formula.
(5) Molecular, Quantified, and Modal: If cjJ and IjJ are any
(propositional) formulas, and rx is any (object) variable, then
("'cjJ), (cjJ-+IjJ), (Vrx)cjJ, and (DcjJ) are (propositional) formulas.
(6) Object descriptions: If cjJ is any formula with one free object
variable x, then (lX)cjJ is an object term.
(7) Complex n-place relation terms: If cjJ is any propositional
formula, and V 1 , .•• , Vn are any object variables which mayor
may not be free in cjJ, then [AV 1 , ••. , vncjJ] is an n-place relation
term (n ~ 1) and cjJ itself is a zero-place relation term (in what
follows, it will sometimes be convenient to regard cjJ as a
degenerate A-expression, [AcjJ], when n = 0).
In addition to the standard abbreviations for the connectives and
quantifiers, we use OcjJ to abbreviate '" D '" cjJ. However, we now define:

D1 X is abstract ("A!X")=df[AyO",E!y]x
D2 x is a possibly existing object = dfOE !x.

So abstract objects are just not the kind of thing that could exist. Here
are some examples of schemata and formulas: DQ ("it is necessary that
Q"); D(3x)(A!x & (F)(xF == cjJ)) ("necessarily, some abstract object encodes
exactly the properties satisfying cjJ");O(3y)(F)(xF -+ Fy) ("possibly, there is
an object which exemplifies every property x encodes"); and (x)(OE!x-+
'" (3F)xF) ("possibly existing objects fail to encode any properties").
We say that a formula cjJ necessarily implies a formula t/J ("cjJ =>t/J") iff
D(cjJ -+ t/J). cjJ is necessarily equivalent to IjJ ("cjJ ¢>1jJ") iff D(cjJ == t/J).
There are two kinds of complex terms - object descriptions and complex
n-place relation terms. Modal formulas may appear in both. For example,
(lx)(A!x & (F)(xF == F = R)) is an object description which reads: the
abstract object which encodes just R.1 The inductive clause for complex
n-place relation terms differs from its counterpart in the elementary theory
in three important respects: (1) it allows modal formulas to appear after
 THE MODAL THEORY OF ABSTRACT OBJECTS 61

A's if the formula is propositional, (2) it allows A's to bind variables which
are not free in the ensuing formula, and (3) it allows propositional formulas
themselves to be relation terms. Here are some examples of new complex
n-place relation terms: [AXY DQbJ ("being a first thing and a second thing
such that necessarily, b exemplifies Q"); [Ax D(E!x -t Px)] ("being an x
such that necessarily, if x exists, x exemplifies P"); DGb ("b exemplifies
G essentially").
As before, 'L is a term iff either 'L is an object term or 'L is an n-place
relation term, for some n.

2. THE SEMANTICS

A. INTERPRETATIONS

An interpretation, ~, of our modified second order modal language is any

octuple, <1II,wo,~,~, ex"-", 2', ext d , $'), which meets the conditions
described in this subsection. The first member of ~ is a non-empty class, 111,
called the class of possible worlds. 2 The second member of ~, wo' is chosen
from 111 and is called the actual world. The third member, f0, is a non-empty
class and is called the domain of objects. The fourth member, ~, is also a
non-empty class, and is called the domain of relations. ~ is the union of a
sequence of non-empty classes ~o'~1'~2,···,i.e., ~ = Un;o.o~n. Each ~n
is called the class of n-place relations (we call ~ 1 the class of properties,
and ~o the class of propositions). ~ must be closed under all the logical
functions specified in the sixth member of the interpretation (2').
The fifth, sixth, and seventh members of ~ impose a structure on 111,
f0, and ~. The fifth member of ~, ext"" is a function which maps ~n X 111
into 'P(f0n) ("the power set of f0 n,,), where n ;::: 1, and which maps ~o x 111
into {T, F}. We index the function to its second argument and call ext ,Jin)
the exemplification extension ("extension",") of It n at w.
The sixth member of ~, 5£', is a class of logical functions which operate
in a manner similar to their counterparts in the semantics of the elementary
theory. However, we: (1) add two additional functions, "f/ dC{}i ("i-vacuous
expansion") and Y gC{} ("necessitation"), (2) constrain the extensions", of
the complex relations resulting from all the logical functions at every
possible world, and (3) allow f?J5£'U/t'§i and U/ty~"f/i to operate on
properties, allow C{}@Yf0 and Y g'§ to operate on propositions, and allow
"f/ d'??i and Y gC{} to operate on all relations. The definitions which make
these three major changes precise go as follows:
62 CHAPTER III

(a) [JJ> 2lJlrfJ i ,

for each i, i?: 1, is a function mapping
(~iU~i+ 1 u ... ) x q; into (~i-1 U~iU",) subject to the
conditions:
(1) for n> 1, ext "J[JJ> :l't1!1<§;(z", 0» =
{< 0 1"", 0i-1' 0i+ 1,··· ,On> I
<01"'" 0i-1' 0, 0i+ 1"'" 0n)Eext ",(in)}

(b) t1!1%yfYi,foreach i, i?: I, is afunctionmapping(~iu~i+ 1 u ...)

into (~i-1 U~iU",) subject to the conditions:

(c) rt'(9%Yi,j' for each i,j, 1 s i <j, is a function mapping

(~jU ~j+ 1 u ... ) into (~jU ~j+ 1 u ... ) subject to the condition:
exi,Jrt'(9%Yi )in)) =
{<01"'" 0i-1' OJ, 0i+1"" ,OJ-I' 0i' 0j+ 1"", On) I
<01,···,01"'" OJ''''' On)Eext",(in)}.
(d) ~tff;y; 2 i,j' for each i,j, 1 s i <j, is a function mapping
(~jU~j+1 u ... )into(~j_l U~ju ... )subjecttothecondition:
ext<)~tff;y; 2 i,j(i n = »
.. , .on> I
{< 01'· .. , 0i' ... , ,oj-I' ,oj + 1"
<01"" 'Oi"" ,OJ''' .,0n)Eext,Jin) and 0i = OJ}.
(e) Y drt'i' for each i, i?: 1, is a function mapping (~i-1 U ~i u ... )
into (~iU~i+1 u ... ) subject to the conditions:
(1) for n?: 1, ext,JY drt'i(i n = »
{<01"'" 0i-l, 0, 0bOi+ 1"'" on>1
<01"'" 0i"'" on)Eext",(in)}
 THE MODAL THEORY OF ABSTRACT OBJECTS 63

(f) C{[(CVV9& is a function mapping (gilo u 9~\ u ... ) x (gilo U gill u ... )
into (gilo u gill u ... ) subject to the following conditions:
(1) for n ~ 1, m ~ 1, -ext w(C{[(9J11'9& (in, er)) =
{< <
01"'" 0., O~ , ... ,O~) 1 01"'" On) ¢-ext ,)in) or
< o~, ... , O~) E-ext ,jam)}

(2) for n = 0, m ~ 1, -ext jC{[(9J11'9&(iO ,am)) = { <01"'" om) 1

<
-eX()iO) =F or 01"'" om)E-ex()a m )}
(3) for n ~ 1, m=O, -ext.jC{[(9J11'9&(i n,aO)) = {<0 1"" ,Om) 1
<Ol, ... ,On)¢-extu,(i n) or -ex(JaO)=T}
(4) for n=O, m=O, -ext,jC{[(9J11'9&(iO,aO)) =
T iff -ex() iO) = F or -ext",.(0°) = T
{
F otherwise.
(g) JII' 0"~ is a function mapping (gilo u gill u ... ) into (gilo u gill u ... )
subject to the conditions:
(1) for n c.1, -ext)JII' 0"~(in)) = {<01'"'' 0n)I<Ol' ... ,on)
¢-extw(in)}

(2) for n = 0, extw(JII' 0"~(iO)) = { T iff ext ~(iO) = F

F otherWIse.
(h) JII' 0"c{[ is a function mapping (gilo u gill u ... ) into (gilo u gill u ... )
subject to the conditions:
(1) for nc.1, -extw(JII'0"C{[(i n )) = {<ol, ... ,on)1
(£0')«01"'" on)Eext "Ain))}

(2) for n = 0, ext)JII' 0"C{[(iO)) = { T iff (£O')~-ext"AiO) = T)

F otherWIse.

This completes the definitions of the logical functions. 3 The seventh

member of J, -extd, is a function which maps gill into 'll(9&). ext d(i 1) is
called the encoding extension ("extensiond " in the metalanguage) of i 1 .
The final member of J, the :F function, maps the simple names of the
language to elements of the appropriate domain. For each object name
K, :F(K)E9&. For each relation name Kn, :F(Kn)EgIln' We call ex(j:F(E!))
the set of existing objects at £O("0"w"). We call extwo(:F(E!)) the set of
64 CHAPTER III

existing objects (i.e., Iff = df r&"wo). We call {ol(3w)(oEext j~(E!)))} the set
of possibly existing objects ("q>Iff"). The complement of q>1ff on [ij) is called
the set of abstract objects ("d").

B. ASSIGNMENTS AND DENOTATIONS

Partitioning the A-expressions. Since we have A-expressions in the modal

language which were not part of the elementary language, we must
incorporate rules to classify the new possibilities. These new rules
correspond to 11 dl1i and JV 1ff11- they help to classify A-expressions with
vacuously bound A-variables and with D's.
The following six major rules partition the class of A-expressions into
nine equivalence classes. If J1 is an arbitrary A-expression, [AV 1 ··· Vn cP J, J1
is classified as follows:
(1) If (3i) (1 sis n and Vi does not occur free in cP and i is the least such
number), then J1 is the ilh-vacuous expansion of UVI ... Vi - 1 V i + 1.·· Vn cP]'
(2) If J1 is not an ilh-vacuous expansion, then if (3i) (1 sis n and Vi is
not the ilh free object variable in cP and i is the least such number), then
where Vj is the ith free object variable in cP, J1 is the iSh-conversion of
Uv 1 •·· Vi - 1Vj Vi+l··· Vj-lViVj+l··· vncP].

(3) If J1 is neither of the above, then

(a) If cP = ('" 1/1), J1 is the negation of [AV 1 •.. vn 1/1 J
(b) If cP = (1/1 ---" X), and 1/1 and X have no free object variables in
common, then where VI' ... ' Vp are the variables in 1/1 and
Vp + 1' ... ' Vn are the variables in X, J1 is the conditionalization of
[AV 1 ••• vpl/lJ and [AV p + 1 ••• vnxJ

(c) If cP = (Vv)I/I and vis the ilh free object variable in cP, then J1is the
ilh-universalization of [AV 1 ••• Vi - 1 VV iVi+l··· vnl/lJ
(d) If cP = (01/1), then J1 is the necessitation of [AV 1 ••. Vn 1/1].
(4) If J1 is none of the above, then if (3i) (1 ::;; is n and Vi occurs free in
more than one place in cP and i is the least such number), then where:
(a) k is the number of free object variables between the first and
second occurrences of Vi'
 THE MODAL THEORY OF ABSTRACT OBJECTS 65

(b) cp' is the result of replacing the second occurrence of Vi with a

new variable V, and
(c) j = i + k + 1,
Jl is the i,r-reflection of [AV l ··· vi+kVVj'" vnCP'].
(5) If Jl is none of the above, then if 0 is the left most object term
occurring in cP, then where:
(a) j is the number of free variables occurring before 0,

(b) cP' is the result of replacing the first occurrence of 0 by a new

variable V, and
(c) i=j+1,
Jl is the ith-plugging of [Av 1 ..• VjVV j + I ... vncp'] byo.
(6) If Jl is none of the above, then
(a) cP is atomic
(b) V l ,···, Vn is the order in which these variables first occur in <p.
(c) Jl = [AV 1 ... vnpnvl'" vn], for some relation term pn, and

(d) Jl is called elementary.

In addition to the examples we saw from the elementary theory, we

now have: [AxD(Px~ ~Qx)] is the necessitation of [A.XPx~ ~ Qx];
[AxWyPX~Qy] is the 2nd -vacuous expansion of [AXyPX~Qy];
[AxVWyPX~Qy] is the 2nd -vacuous expansion of [AxWyPX~Qy]; etc.
J-assignments. If given an interpretation J of our language, an
,§ -assignment, f, will be any function defined on the primitive variables
of the language satisfying the following two conditions:
(1) where Vis any object variable, /(V)E.@
(2) where nn is any relation variable, /(nn)E~n'

Denotations. If given an interpretation J of our language, and an

J -assignment f, then we recursively define the denotation of term r with
respect to interpretation J and J-assignment /("dY;,/(r)") as follows:
(1) where K is any primitive name, dY;,f(K) = ff(K)
66 CHAPTER III

(2) where rt. is any primitive variable, dJfjrt.) = I(rt.)

(3) where (lX)¢ is any object description,
d Jf,/«lX)¢) ={<> iff (3/')(1'.< I & I'(x) = <> & I' satisfies ¢ with
respect to too & (/")(/" xl & /"satisfies ¢ with
respect to too ~/" =1'))
undefined, otherwise.
where satisfaction is defined as in subsection c,s
(4) where [Avl ". vnpnvl ... vnJ is any elementary A-expression,
dJf,t<[AV 1 .,. vnpnvl . . . vnJ) = dJf,/(pn)
(5) where f1. is the jIb-plugging of ~ by 0,

d Jf ,/(f1.) = &,!i'OZt~i(dJf,/(~),dJf,/(o))
(6) where f1. is the jtb-universalization of ~,
d Jf ,/(f1.) =OZtJV.ftYi(dJf,/(m
(7) where f1. is the j,/b-conversion of ~,
d Jf ,/(f1.) = rtJ(9 .lVYijdJf,/(m
(8) where f1. is the i,/b-reflection of ~,
d Jf ,/(f1.) =!!llrff:F!i'ijdJf,/(W
(9) where f1. is the itb-vacuous expansion of ~,
d Jf ,/(f1.) = Y drtJi(dJf,/(m
(10) where f1. is the conditionalization of ~ and"
d Jf ,/(f1.) = rtJ(9.IV!!fi (d5 ,/( ~),d5, /( 0)
(11) where f1. is the negation of ~,
d Jf ,/(f1.) = .lVrff~(dJf,/(~))
(12) where f1. is the necessitation of ~,
d Jf ,/(f1.) = .lVrffrtJ(d5,/(~))
(13) where ¢ is any propositional formula, d Jf ,/(¢) is defined as
follows:
 THE MODAL THEORY OF ABSTRACT OBJECTS 67

(a) if ¢ is a primitive zero-place term, d,F,/( ¢) is defined above

(b) if ¢ = pn o1 ... on ,d,Fj¢) =
f/' 20/1r:§ 1 (f/'20/1r:§ 2 (. . . (f/'20/1r:§ n(d,F ,/(pn), d,F ,I (On)), ... ),
d,F ,/(02))' d,F ,/(01))
(c) if ¢ = ('" l/I),d,F,,(¢) = .!V'@"r:§(d.?',/(l/I))
(d) if ¢ =(l/I ~ X), d.?',,(¢) =~(!7JV£0(d.?"/(l/I),d.?',/(X))
(e) if ¢ = (Vv)¢,d.?'j¢) =o/1JV§Yl(d.?',/([AVl/l]))

(f) if ¢ = (Dl/I),d.?',f(rP.) = .!V'@,,~(d.?',/(l/I))·

Here are some examples of A-expressions and their denotations with respect
to a given § and l
d,F'/([hD(Px~ '" Qx)])=JVg'~(d,F,,([AXPX~ '" Qx]))
d,F,/([hvwy Px ~QyJ) ="f/:s.::I~ 2 (d,F./( [AXWYPx ~ QyJ))
d,F,/([hwy PX~QyJ)=Yd~ 2 (d,F.,([hy Px ~QyJ))
d,Fj[hy Px ~QyJ)=~(!7JV£0(d,F./(P),d,F.,(Q))
d,F./([Jey 0 Gb J) = ifd~ 1 (JVg'~(gp2OU~ 1 (dJ'./(G),
d,F.,(b))))
d,F,/(D (E!b & Gb)) = JV g'~(JV@"r:§(~(!7JV£0 (&,20/1r:§ 1
(d.fi./(E !),d.fi./(b)),JV g'r:§(f!jJ2OUr:§ 1 (d.fi./(G),d.fi,/(b)))))).

C. SATISFACTION

If we are given an interpretation §, and an §-assignmentf, we may define

I satisfies ¢ with respect to world w as follows:
(1) If ¢ is any primitive zero-place term,fsatisfies ¢ with respect
to w iff extw(d.fi./(¢)) = T
(2) If ¢ = pn01 ... on.! satisfies ¢ with respect to w iff (301)'" (30n)
(3-in)(01 = d.fi.,(Ol) & 00' & on = d.fi./(on) & -in = d,F./(pn)
& <01" 00, On) Eext,j-in))
(3) If ¢ = Op 1, I satisfies ¢ with respect to w iff
(30)(3-i 1)(0 = d ,F,/(o) & -it = d ,F./(pl) & o Eext",,(-i 1))
(4) If cfJ = ( '" t/t), I satisfies cfJ with respect to w iff I fails to satisfy
t/t with respect to w
68 CHAPTER III

(5) If ¢ = (ljJ --+ x),1 satisfies ¢ with respect to £0 iff! fails to satisfy ljJ
with respect to £0 or I satisfies X with respect to £0
(6) If ¢ = (Vr;.)ljJ, I satisfies ¢ with respect to £0 iff
(V1')(fV --+ I' satisfies ljJ with respect to £0)
(7) If ¢ = (O ljJ), I satisfies ljJ with respect to £0 iff (w')(1 satisfies
ljJ with respect to £0').

D. TRUTH UNDER AN INTERPRETATION

¢ is true under interpretation .§ iff every .§-assignment I satisfies ¢ with

respect to £00. ¢ is false under .§ iff no .§-assignment I satisfies ¢ with
respect to £00. The definitions of valid (i.e., logically true) and model remain
the same.

3. THE LOGIC

A. LOGICAL AXIOMS

The logical axiom schemata fall into five groups: the propositional
schemata, the quantificational schemata, the modal schemata, the
schemata governing A-expressions, and the schemata governing des-
criptions. The presentation of the schemata governing A-expressions will
be interrupted by two definitions, in terms of which the second A-schema
will be constructed. We define a modal closure of ¢ to be any formula
obtained by prefixing any finite number of (or possibly zero) boxes to ¢.
Then, with the exception of the schemata governing descriptions, all the
modal closures of any instance of the following schemata shall be the
logical axioms of our system:

Propositional Schemata
LAl: ¢ --+ (ljJ --+ ¢)
LA2: (¢ --+(ljJ --+ X» --+((¢ --+ ljJ) --+(¢ --+ X»
LA3: (~ ¢ --+ ~ ljJ) --+ (( ~ ¢ --+ ljJ) --+ ¢).

Quantificational Schemata
LA4: (a) (r;.)¢ --+ ¢~, where 7: contains no descriptions and is sub-
stitutable for rx
 THE MODAL THEORY OF ABSTRACT OBJECTS 69

(b) (rx)<p-+(tj;!r-t<P~), where tj; is any atomic formula, and r

both contains a description and is substitutable for r!., ~
LA5: (rx)(¢-+tjJ)-+(¢-+(rx)tjJ), provided rx is not free in ¢.

Modal Schemata
LA6: O¢-+¢
LA7: O(¢ -+tj;) -+(O¢ -+ Otj;)
LA8: 0¢-+00¢
LA9: O(rx)¢==(rx)O¢
LA1O: (x)(F) (OxF -+ OxF).

A-Schemata
A-EQUIVALENCE: where ¢ is any propositional formula containing no
descriptions, the following is an axiom:
(Xl)··· (Xn)( [AV l ... V n¢ ]Xl ... Xn == ¢~::::::~~).
To more efficiently state the second A-schema, A-IDENTITY, we utilize
the following two definitions:
F l -- Gl -df
-

D4 pn=Gn=df (wheren>l)
(xl)···(xn-l)([AyFnyxl···xn_l]=[AyGnyxl···xn_l] &
[AY FnXlyx2··· x n- l ] =
[AyGnxlyx2···xn_l] & ... &
[AY pnx l ... xn-lY] = [AY Gnx l ... xn-lY]).
Given D3 and D4 , we have:

A-IDENTITY: where pn is any relation term and Vb ... , Vm V~, ... , V~ are
distinct object variables not free in pO, the following is an axiom:
[AV l ... VnP"V l ·· .Vn] = pn & [AV 1... VnPO] = [AV~ ... V~p0].

Description schemata
L-DESCRIPTIONS l : where t/t is any atomic formula or conjunction of
atomic formulas, the following is an axiom:
t/t~x)'" -+(3y)(¢~ & l/In.
70 CHAPTER III

L-DESCRIPTIONS 2 : where !/J is any atomic formula, the following is

an axiom:
!/J~x)1> -> (y)( <Pi -> !/J~) .

L-DESCRIPTIONS 3 : where !/J is any atomic formula with vl free and X

is any formula with V2 free, the following is an axiom:
!/J~:)1> -> ~ ((:3y)(<pi & X~2) & (:3y)(<pi & ~ x~,))·

The propositional and quantificational schemata have been carried over

almost intact from the elementary theory. The exception is LA4, which
has been modified to accommodate non-denoting descriptions and terms
which may contain such descriptions (for discussion on this matter, see
the Appendix to Chapter II, part D). These first five axioms have greater
significance than their elementary theoretic counterparts, due to the
presence of new kinds offormulas and terms. For example, OP -> (Q -> OP)
o
is an instance of LA I; and (F)Fc-> [Ay Ty]c) is an instance ofLA4a.
LA6- LA8 are the standard three propositional modal axioms of Ss.
LA9 is the second order version of the Barcan formula. Both our object
quantifiers and relation quantifiers are unrestricted (they range over
everything in their respective domains). Since boxes are interpreted
semantically as universal generalizations over the domain of worlds,
commuting a box with a universal quantifier is just as valid as commuting
two universal quantifiers. Also, diamonds commute with "existential"
quantifiers. LAlO is a new logical axiom which governs the modal logic of
encoding. It guarantees that objects encode their properties "rigidly"
(should they encode any). That is, if they encode a property at some
possible world, they encode that property at all possible worlds. To see
that LAlO is logically true, note that the encoding extension (ext",,) of
a property is not relativized to a world (Section 2, A., clause 7). So the
conditions for satisfaction for encoding formulas (Section 2, c., clause 3)
are totally independent of the worlds. So if an encoding formula is true
at some world, it is true at every world.
Then we have our two A.-schemata. A.-EQUIVALENCE, though it has a
minor restriction (see the Appendix to Chapter II, part E, for details), has
greater significance than its counterpart in the elementary theory, due to
the presence of A-expressions which are vacuous expansions or necessita-
tions. The first conjunct of A-IDENTITY has the same significance as its
counterpart in the elementary theory, however, the second conjunct of
 THE MODAL THEORY OF ABSTRACT OBJECTS 71

A-IDENTITY is new. Intuitively, [h 1 PO] and [JeX2PO] are both names

of the same vacuous property. In every interpretation, they will denote
semantically identical properties, so the encoding extensions of such
properties must be the same. This is what is asserted by the second conjunct
of A-IDENTITY, except that it generalizes to the case where relations are
denoted by A-expressions with more than one vacuously bound variable.
Finally, we have the three description schemata which we discussed in
the Appendix to Chapter II, part F. Since our definite descriptions are
rigid designators, instances of these schemata aren't necessarily true. They
are paradigm examples of logical truths which aren't necessary.

B. RULES OF INFERENCE

Officially, we need only two rules of inference:

(1) Arrow Elimination ("~E"): from 41 and 41~l/J, we may
infer l/J
(2) Universal Introduction ("UI"): from 41, we may infer (cx)41
The notion of proof, logical theorem, and provable from all carryover
from the elementary theory. Using these notions, we may state a restricted
version of the rule of necessitation which is easily derivable: 6 •7

Box Introduction ("01"): If we are given a proof of 41 from a set of

formulas r, then if in this proof 41 does not depend on any unmodalized
formula (i.e., formula not beginning with a box), then n---
041 (dependence
is defined in Chapter I, Section 3).

We will appeal to this rule on numerous occasions in Chapter IV. And

we will also use the many standard derived rules of the second order
modal predicate calculus. Our derived rules of A-Introduction and
A-Elimination are formulated as in Chapter I, with the restriction that
definite descriptions not occur in 41.
Note that the RELATIONS theorem schema (I., Section 3) is now
derivable without the restriction that Xl' ... ,x n be free in 41, but must be
restricted to 41's which contain no descriptions:

RELA TIONS: where 41 is any propositional formula with no free F"'s

and no descriptions, the following is a logical theorem:
72 CHAPTER III

Of course if ¢ contains a definite description, then by LA4b, instances of

the above schema follow from the assumption that some atomic formula
containing the description is true.
In addition to the examples of this schema offered in Chapter I, we
now have further examples:
(a) (:IF)D(x)(Fx == DGx)
(b) (:IF)D(x)(Fx == D(E!x~ Gx))
(c) (:IF)D(x)(Fx == Gb)
(d) (:IF)D(x)(Fx == DGb)
(e) (:IF)D(x)(y)(Fxy == DGb).
(a) tells us that for any property G, there is a property of exemplifying G
essentially; (b) tells us that for any property G, there is a property of
necessarily exemplifying-G-if-existing; (c) and (d) assert, respectively, that
there is a property objects exemplify just in case b exemplifies G and just
in case necessarily b exemplifies G; (e) asserts that there is a two-place
relation objects bear to one another just in case necessarily b exemplifies
G.
Note also that while the following is, strictly speaking, not an instance
of RELA nONS, it is nevertheless easily derivable:

PROPOSITIONS: Where ¢ is any propositional formula with no free

FO's and no descriptions, the following is a logical theorem:
(:lFO)D(FO == ¢).
And we may also define the conditions under which propositions are
identical:
Ds FO = GO = dJ [AY FO] = [AY GO].
That is, propositions FO and GO are identical iff the property of being
such that FO is encoded by all and only the objects encoded by the property
of being such that GO. This definition turns out to be extremely useful in
Chapter IV, Section 2, where we prove that there is a unique actual world.
RELATIONS, PROPOSITIONS, D 4 , and Ds comprise a complete
modal theory of n-place relations. It is an important feature of this theory
that relations with the same exemplification extensions at each possible
world may nevertheless be distinct. For example, it is consistent with our
theory that the properties of being an equilateral Euclidean triangle and
 THE MODAL THEORY OF ABSTRACT OBJECTS 73

being an equiangular Euclidean triangle are distinct, even though they

have the same exemplification extensions at each possible world. And the
properties of being-blue-or-not-blue and being-green-or-not-green may be
distinct, though logically equivalent.
We call the metaphysical system which consists of the interpreted modal
language, together with its logic, the modal object calculus (with pro-
positions, and complex terms).

4. THE PROPER AXIOMS

We have again embedded our primitive metaphysical notions in the atomic

formulas of the language, and embedded the primitive logical notions
(including the new primitive, necessarily) in the complex formulas and
terms. We now use our primitive theoretical relations (existence and
E-identity), to state the theory of abstract objects (and in the course of
doing so, produce a theory of identity as well). The theory has five axioms,
three of which are schematic. We assert that the modal closures of the
first two un schematic axioms, the modal closures of all the instances of
the first two schemata (IDENTITY and A-OBJECTS), and the un-
modalized instances of the third schema (DESCRIPTIONS) are all true A
PRIORI:

AXIOM 1. ("E-IDENTITY"): x=Ey==(>E!x&(>E!y&O(F)(Fx==

Fy).

AXIOM 2. ("NO-CODER"): (>E!x-> o "-'(3F)xF.

In order to state the third axiom, we need the following definition:
D6 X =y =dfx=EY v (A!x & A!y & o (F)(xF == yF)).
Since the definienda in D3-D6 all have a special logical form, we have:

AXIOM 3. ("IDENTITY"): a=[3->(4)(a, a) == 4>(a, [3)), where 4> (a, [3) is the
result of replacing some, but not necessarily all, free occurrences of a by
[3 in 4> (a,a), provided [3 is substitutable for a in the occurrences of a it
replaces.

AXIOM 4. ("A-OBJECTS"): for any formula 4> where x is not free, the
following is an axiom:
(3x)(A!x & (F)(xF == 4>)).
74 CHAPTER III

AXIOM 5. ("DESCRIPTIONS"): where ljJ is any atomic or defined

object identity formula with one free object variable v, the following is an
aXIOm:

ljJ~'X)<P == (3 !y)¢~ & (3y)(¢~ & t/J~).

Given our discussion of the axioms and theorems of the elementary theory,
these axioms should be straightforward. Semantically, each possible world
will look somewhat like a model of elementary object theory. At each
world, there are objects which exist there and which fail to exist there.
But from the point of view ofa given world, say the actual world, the objects
which fail to exist divide up into two mutually exclusive classes-the objects
which necessarily fail to exist and the objects which exist at some other
possible world. So from the point ofview of the actual world, E-IDENTITY
and NO-CODER govern the objects which either exist at this world or exist
at some other world.
The IDENTITY axiom has greater significance than its counterpart in
the elementary theory. This is due to the presence of the many new kinds
of terms in the language. The following are both instances of ID ENTITY:
FO = GO --+ (DF O == D GO)
Fl = G1 --+([AyDFa]b == [AyDGa]b).
A-OBJECTS also has greater significance since it now yields objects which
encode vacuous and modal properties. Since the modal closures of
A-OBJECTS are axioms, the following counts as an axiom schema:
D(3x)(A!x & (F)(xF == ¢», where ¢ has no free x's.
Semantically, this tells us that given a world to and a condition on
properties ¢, there is an abstract object at to which encodes just the
properties satisfying ¢ at to. A formula like "Fs" ("Socrates exemplifies
F") is satisfied by different properties at different worlds. At each world,
then, there is an A-object which encodes just the properties Socrates
exemplifies at that world. A formula like "F=R v F=S" is satisfied by
the same two properties, roundness and squareness, at each world. Given
LAlO, and definition of identity, the.,.,. round square of one world will be
identical with the.,.,. round square of any other world. Intuitively, all
of the A-objects from each of the worlds can be grouped into one set, the
set of A-objects, which stays fixed from world to world. In the future,
when we use restricted z-variables, they will range over this set.
 THE MODAL THEORY OF ABSTRACT OBJECTS 75

DESCRIPTIONS has been added to our list of axioms and it has a few
interesting and important features we should consider. Its instances are
paradigm cases of A PRIORI truths which are not necessarily true. Our axiom
guarantees that the descriptions in our language that appear in atomic
or object identity formulas behave according to our A PRIORI intuitions (see
note 1 of Chapter II for reasons why instances of DESCRIPTIONS are
not logically true). But it is easy find worlds such that the left side of a
given instance of DESCRIPTIONS is true there while its right side is
false there. That is because the descriptions of our language are rigid
designators. The left side of a given instance of DESCRIPTIONS, which
will say essentially "the thing which satisfies <P t/I's", will be true at a world
Wi just in case there is a unique object satisfying <p at the base world Wo
which satisfies t/I at Wi' But there need not be an object which satisfies
both <p and t/I at Wi or which uniquely satisfies ¢ at Wt. But that's what
it would take for the right side of the instance of DESCRIPTIONS to be
true at Wi' So DESCRIPTIONS is not necessarily true, and given our
restricted version of the rule of necessitation, we cannot produce the
modalized instances of DESCRIPTIONS as (proper) theorems (for further
discussion on this matter, see note 6).
E-IDENTITY, NO-CODER, IDENTITY, A-OBJECTS, and DES-
CRIPTIONS jointly constitute the modal theory of abstract objects.
Evidence for thinking that the theory is consistent may be found in
Appendix A, part C, where the reader will find an extensional model of
the monadic portion of the theory described in ZF. It is provable that
some propositions as well as some complex relations do not have unique
constituents. 8 But such a result might seem insignificant when compared
to the potential the theory has for applications.
In these applications, it will be important to distinguish three senses of
the phrase "possible object". On one sense of this phrase, objects which
satisfy D2 (Section 1) are possible objects, whereas abstract objects are
not. We always use "possibly existing object" to indicate this sense of
"possible object".
The other two senses of the phrase are ones in which abstract objects
are possible objects. Consider D 7 , where "z" is a restricted variable ranging
over abstract objects:
D7 z is strongly possible ("SPoss(z)") = df O(:3x)(F)(zF == Fx).
We always use the phrase "strongly possible object" to indicate this sense
76 CHAPTER III

of "possible object". For example, Socrates' blueprint is strongly possible,

and so is the blueprint of Socrates' blueprint. 9
The third sense of "possible object" we distinguish requires a preliminary
definition.
Ds X is weakly correlated with z ("WCor(x,z)") = df(F)(zF ~ Fx).
For example, abstract objects which encode just some of the properties
a given object exemplifies are incomplete blueprints of the object - the
object is weakly correlated with them. We now have,
D9 z is weakly possible ("WPoss(z)") =dfO(:lX) WCor(x,z).
Weakly possible A-objects are "possible objects" in the sense that they
do not encode any contradictory properties. F and G are contradictory
properties iff it is not possible that some object exemplify both of them.
The notion of weak correlation we defined in DB was used in Chapter II,
Section 3. Recall that we defined existence 2 for an object x ("E ! !x") as
(:ly)(E!y & (F)(xF ~ Fy». So abstract objects can "exist" in the sense that
they have a weak correlate which exists. To say that an abstract object x
"might have existed" is to say that it is possible that x have a weak correlate
which exists ("OE! !x").
We shall keep these distinctions straight in the applications which follow.
To prepare for these applications, we add to our primitive vocabulary the
usual abbreviations of standard English gerunds. Also, we adopt a modal
version of our AUXILIARY HYPOTHESIS - A -objects necessarily fail to
exemplify nuclear relations. 1 0
 CHAPTER IV

THE APPLICATIONS OF THE MODAL THEORY

In much of this chapter, we shall be speaking in the object language.

When doing so, everything we say may be analyzed in terms of our four
metaphysical primitives (object, n-place relation, exemplifies, encodes), six
logical primitives (not, if-then, every, necessarily, the, being such that), and
two primitive theoretical relations (exists, E-identical). All of the definitions
constructed and theorems proved in what follows may be ultimately
analyzed in terms of these primitives. We begin with a definition of truth.

1. TRUTH

Since propositional formulas are also terms which denote propositions,

we shall follow Ramsey in supposing that the predicate "is true" and the
operator "it is true that" are definable by elimination. l The language we
developed in the previous chapter allows us to incorporate Ramsey's
suggestion through the formulation of the following definitions:
D lO FO is true =dJFo
Dll It is true that FO =dJFo.
Ramsey's idea works fine as long as we are interested in just the truths
relative to our world. A less mundane notion of truth is the notion of
truth at a particular world. We shall produce a definition of this notion
in the next section, once we have modelled possible worlds. But before
we do so, we require a few more preliminary definitions.
We shall say that a property Fl is constructed out of a proposition FO
iff Fl is the property of being such that FO:
D12 Fl is constructed out of FO ("Const (Fl, FO)") = dJ
Fl = [Ax F°].
We then define a vacuous property to be one which is constructed out of
some proposition:
D13 Fl is a vacuous property ("Vac (Fl)") = dJ (:JFO)Const(Fl, FO).

77
78 CHAPTER IV

Examples of vacuous properties are: being such that Carter is President,

being such that Fischer defeated Spassky, being such that Nixon did not
resign the Presidency, being such that a Luxembourgian was the first man
on the moon, being such that every man loves every fish, etc. A-
EQUIVALENCE guarantees that necessarily, an object x exemplifies a
vacuous property like [AY PO] iff pO is true. So if po is true, everything
exemplifies [AYPO], and if pO is not true, nothing does. Consequently,
vacuous properties are either "full" (everything exemplifies them) or
"empty" (nothing exemplifies them). Indeed, some properties will be
necessarily full and others will be necessarily empty. Being such that either
Carter is President or Carter is not President ([AY Pc v ~ Pc]) is an
example of the former; being such that both Carter is President and Carter
is not President ([ AY Pc & ~ Pc]) is an example of the latter.
Finally, it will be important to define conditions under which we can
say that an abstract object encodes a proposition. Consider D 14 :
DI4 Z encodes pO ("LzFO") =dfz[AyP°].
That is, an abstract object Z encodes a proposition pO iff Z encodes being
such that pO.

2. MODELLING POSSIBLE WORLDS 2

Possible worlds will be abstract objects which encode only vacuous

properties and which meet two other conditions. For one thing, they must
be maximal, i.e., for every proposition pO, either they encode pO or they
encode the negation of pO.
DIS Z is maximal ("Max (z),,) = df(FO)(LzPO v L z ~ pO).

So if an object z is maximal, it must encode, for every proposition pO,

either being such that pO or being such that it is not the case that pO.
Secondly, worlds must in some sense be possible objects. One way to
make this requirement precise would be to stipulate that worlds must be
weakly possible (i.e., as in III, Section 4, D9). This would require that it
be possible that some object exemplifies every property the world encodes.
However, a more elegant way of ensuring that inconsistent propositions
will not be encoded by the same world is to stipulate that if an object Z is to
be a world, then it must be possible that every proposition Z encodes is
true. 3
We can formalize all these conditions on worlds in the following
definition:
 THE APPLICATIONS OF THE MODAL THEORY 79

z is a possible world = df(F 1 )(zF -+ Vac(F)) & M ax(z) &

O(F°)(LzF O-+ FO).
Although this definition would serve us well, there is a more elegant
definition which is equivalent:
D 16 z is a possible world ("World (z)") =df(F)(zF -+ Vac(F)) &
O(F°)(LJo := FO).

That is, an object z is a world iff every property it encodes is vacuous and
it is possible that z encodes all and only true propositions. Given D 16 ,
we can prove that worlds are maximal:

THEOREM 1. (z)(W orld(z) -+ M ax(z)).

Proof. Suppose z 5 is an arbitrary world. By definition, O(F°)(LzsFO :=
FO). We want to conclude that for an arbitrary proposition QO, that LzsQo
or Lzs ~Qo. We do this in two stages : in stage (A), we prove that
O(LzsQO v L z ~ QO), and in stage (B), we use a theorem of Ss (which
distributes a <> over a disjunction) and our new logical axiom LA I 0, to
prove that ~zs QO or ~zs ~ QO.
(A) In this stage, we rely on the following theorem of Ss: D(¢ -+ 1/1)-+
(O¢-+Ol/l). If we let ¢='(FO)(L.z,F°:=Foy, and 1/I=ILzsQoV
~zs ~ QO" then by establishing that 0 (¢ -+ l/I), we can apply the Ss theorem
using the fact that O(FO)(L.z,F° := FO) and reap our initial result. So we
first establish that (¢ -+ 1/1), and then use 01. So assume ¢, and instantiate
the quantifier to both QO and ~ QO. SO ~zsQ°:= QO and ~zs~Q°:= ~Qo.
Since QO v ~ QO, it follows that ~zsQo v ~zs "" QO, i.e., 1/1. So O(¢ -+ 1/1), by
01. And by the Ss theorem, we have our initial result: <)(~zsQo v ~Z5 ~ QO).
(B) It is also a theorem of Ss that O(¢ v I/I)-+(O¢ v 01/1). By letting
¢, 1/1 be the disjuncts of our initial result, it follows that O~zs Q v
O~zs ~ QO. By LAlO, it follows that 0 ~z5QO v D~zs ~ QO, since ifpossibly
an object encodes a property, it does so necessarily. And by another
theorem of Ss, O(L.zsQo v L.zs~QO). By LA6, L.zsQo v ~zs~Qo, i.e.
Max(zs)· So every D 16 -world is maximal. I8l

The proof of Theorem 1 could be simplified a great deal if we think

model-theoretically, using the notion of a possible world as a primitive.
If we signify that we are appealing to the semantically primitive notion
of a world by shifting type styles and writing "to£!-dd", the proof of
Theorem 1 would read as follows:
80 CHAPTER IV

Suppose Z5 is an arbitrary world. By D 16 , at some possible wo-ztd, say £01'

Z5 encodes there all and only the propositions true there. For any
proposition FO, either FO is true at £0 1 or ~ FO is true at £0 l ' So for all
propositions FO, either Z5 encodes FO at £01 or Z5 encodes ~ FO at £01' But
by LAlO, if an abstract object encodes a property at some possible Wo'll d,
it encodes it at all possible wolttdJ. So (at the ac//k£[t wOi{d, £00) for all
propositions FO, either Z5 encodes FO or Z5 encodes,...., FO. So every world is
maximal. [2g

Readers who find the semantically primitive notion of a wod d an intuitive

one will find it much easier to construct model theoretic proofs like the
above for the theorems which follow. The proofs are a good deal simpler,
since the theorems turn out to be almost immediate consequences of the
axioms and definitions. However, in the context of the present work, the
presentation of such model-theoretic proofs has a curious disadvantage.
It fosters the wrong impression. It would encourage the reader to suppose
that there are A-objects which can represent the worlds. But the POINT
of the theorems which follow is to show that the worlds just AREA-objects.
Consequently, I shall not present any further proofs model-theoretically.4
Those who prefer to think model-theoretically need not attend to the
proofs offered in the text to verify that the following claims are in fact
theorems. But these readers should not suppose that the model-theoretic
results could be considered to be a substitute for the metaphysical results.
They are simply a device for quickly verifying that the metaphysical claims
do follow.
Theorem 1 is instrumental for showing that D 16 implies our first
definition of 'possible world'. It is a straightforward proof-theoretic exer-
cise to show that the first definition implies D16" 5
Let us say that propositions FO and GO are inconsistent iff it is not
possible that both FO and GO be true. We may then say that an A-object
Z is consistent iff it is not the case that Z encodes inconsistent propositions:

D17 Z is consistent ("Cons (z),,) = df

~ (3FO)(3GO)( ,....,O(F° & GO) & LzFO & LzGO).
We then have:

THEOREM 2. (z)(World(z)~ Cons(z)).

That is, every world is consistent.
Proof. Assume for reductio that World(z4)' ,....,O(p° & QO), L Z4 PO, and
 THE APPLICATIONS OF THE MODAL THEORY 81

Lz QO. Note that byOI and LAlO, these last two assumptions must be
4

necessary truths, i.e. DL z4 PO and DLz4 QO. So by a theorem of Ss,

D(L z4 PO & L z4 QO). Now by DI6,O(FO)(Lz.F° '= FO). Now let ¢=
'(FO)(Lz.F° '= For' and t/I =1 po & QO,. We shall want to show that
D(¢--t/l). Clearly, D((L z4 PO & L Z4 QO)--(¢--t/l)). But since D(L z4 PO &
L z4 QO), it follows by a theorem ofS s (namely, that D(x--e) & Dx-- De)
that D(¢ -- t/I). Now by I.theusing Ss theorem that D(¢ -- t/I) -- (O¢--
Ot/l), it follows that O(po & QO), contrary to hypothesis. [6J

Another result quickly falls out of our definitions:

THEOREM 3. (z)(World(z)--(FO)(GO)((LJo & FO ==>GO)--LzGO)). That

is, all the necessary consequences of propositions encoded in a world are
also encoded in that world.
Proof. Assume World (Z3)' L Z3 PO, and that pO ==>Q o (i.e., D(p0 __ QO)).
Again, DL z3 PO, by 01 and LAlO. So if ¢ =1 (FO)(Lzf O'= FO), and
t/I =ILz 3 QO" we easily get D(¢--t/l) by theorems of Ss. So by again
appealing to the theorem that D(¢ -- t/I) --(O¢ -- Ot/l), we get OLz 3 QO,
since O¢ follows from the definition of a world. So by LAlO and LA6,
Lz 3 QO. 0

Theorems 1,2, and 3 should give us a good grasp on the inner workings
of the theory, as well as the import of the second clause of D 16 . They
should also help us to see that the following definition is justified:
D 18 FO is true at z =djWorld(z) & LJo.
So whenever z is a world, the propositions true at z are just the propositions
z encodes. This definition suggests what it is for a world to be actual:
D 19 z is an actual world ("Worlda(z)")=dj World (z)&(FO)(LzF°'= FO).
That is, an actual world is any world such that all and only true
propositions are true at that world. We now get the following result:

THEOREM 4. (z)(z')(Worlda(z) & Worlda(z') -- z' =z).

That is, there is at most one actual world.
Proof. Suppose, for reductio, that World a(zl) and Worlda(zz), where
ZI =/=zz' Since ZI and Zz are distinct A-objects, they must differ with respect
to at least one encoded property. Since they are both worlds, any such
property must be vacuous. So without loss of generality, suppose LZlQo &
82 CHAPTER IV

~ ~Z2QO. By Theorem l,z2 must be maximal. SO ~Z2 ~Qo. But since both
Z1 and Z2 are actual, every proposition they encode must be true.
Contradiction. ~

We also get:

THEOREM 5. (:lz)Worlda(z).
Proof. By A-OBJECTS, there is an abstract object which encodes a
property F iff it's a vacuous property constructed out of a true proposition,
i.e., (:lz)(F)(zF == (:lFO)(FO & F = [AY FO])). Call this object zoo To show Zo
is an actual world, we show that it satisfies both clauses of Dig. So we
show (a) World (zo), and (b) (FO)(~zoFO == FO):
(a) Clearly, every property Zo encodes will be vacuous. So we want to
show that possibly, all and only the true propositions are encoded in zoo
Consider an arbitrary proposition QO. (~) Suppose ~zoQo. Then by
definition of Zo, (:lFO)(FO & [AyQO] = [Ay FO]). Call this proposition RO.
Since [AyQO] = [AyRO], it follows from the definition of proposition
identity (III, Section 3, Ds) that QO= RO. Since RO is true, so is QO. (<--)
Suppose QO is true. Then Zo[AY QO], i.e., ~zoQo.
Since we have established that ~zoQo == QO, for an arbitrary proposition
QO, it follows that O(FO)(~zoFO == FO).
(b) Clearly, Zo encodes all and only the propositions which are true, as we
have just shown. ~
With Theorems 4 and 5, we have proven that there is a unique actual
world (from A PRIORI assumptions alone). We are entitled to name this object
and we do so as follows: wa=df(lz)Worlda(z).
It should be interesting that there is an actual world even though it does
not exist. If we had proven that the actual world existed, then we
would have just proven that something exists from A PRIORI assumptions;
yet A PRIORI theories should not have contingent consequences. Also, if the
actual world were an existing object like you, me, or some sub-atomic
particle, it would fail to encode any properties (by NO-CODER). There
would be no reason to think that its vacuous properties were any more
crucial to its identity than other properties it exemplified (like not being
a cat, being non-red). Recall here Wittgenstein's dictum that the world is
just all that is the case. 6 These considerations make it easy to see that
the word "actual" as it occurs in the English phrase "the actual world"
does not mean "existing". (It is intriguing that the word "is" in
 THE APPLICATIONS OF THE MODAL THEORY 83

Wittgenstein's dictum may again be read as "encodes", since given our

derivative use of this notion, w. encodes just all that is the case.}
These remarks on existence and actuality bring us to the following,
important definition:
D 20 x exists at Z = df World (z) & LzE !x.
That is, an object x exists at a world iff the proposition that x exists is true
at that world. Note that we cannot prove from our A PRIORI assumptions
that anything exists at w". This is a fortunate result since, as we noted
earlier, it must surely be a contingent matter that something exists (at the
actual world). Let us say that a world of existing things is any actual
world such that something exists:

D21 Z is a world of existing things =dj World,,(z) & (:lx)E!x.

Clearly, there could be at most one world of existing things - all such
worlds would have to be identical with w•. If we add the contingent
assumption that something exists, it follows that w" is THE world of existing
things. So we must add a contingent assumption to prove that there is a
unique world of existing things, though it is true A PRIORI that there is a
unique actual world.
We might note in passing that it is consistent with our definition of
existing at a world that objects exist at more than one world. However,
some philosophers apparently like to work with a notion of existing at a
world on which objects can exist at at most one world. 7 We could
accommodate the views of these philosophers were we to define existence at
a world as follows:
x exists at z =df World(z) & (FO)(LzFO == [JeyFO]x).
Using this definition, we would get the result that individuals are
world-bound. For suppose b exists at worlds Zl and Z2,Zl =1= Z2. If [,l.yQ] was
the vacuous property distinguishing Zl and Z2' it would follow that b both
exemplified and failed to exemplify this property. So b cannot exist at
both, on this definition of "exists at". Counterpart theorists may then
prefer to use this latter definition in their investigations.
One of the most important theorems to fall out of our theory verifies
a now common philosophical intuition - necessary propositions are true
at all possible worlds. Let "w" be a restricted variable ranging over the
worlds:
84 CHAPTER IV

THEOREM 6. (FO)(DFO == (w)LwFO).

Proof. (-+) Assume DQo, where QO is arbitrary and show Lw,Qo, where
W 1 is arbitrary. Let 4> ='(FO)(Lw,FO == For' and l/J ='Lw,Qo" and by
now familiar reasoning, it follows that OL w , QO, using the fact that
D(4)-+l/J)-+(04>-+Ol/J). SO Lw,Qo, by LAlO and LA6. (.-) Assume
(w)~wQo, for an arbitrary QO, i.e., (x)(World(x)-+LxQO). Let us show that
this must then be necessary. By the Barcan formulas, it. suffices to show:
o
(x) (World(x) -+LxQO). Suppose not. Then O(World(b) & ""' ~bQO), where
"b" is arbitrary. Hence, OWorld(b) & O""'~bQo. Since OWorld(b)
-+ W orld(b) (exercise) and 0 ""' ~bQO -+ ""' ~bQO (LAlO), we get W orld(b) &
""' LbQo. So, ""' (World(b) -+ LbQO), contrary to hypothesis. Hence,
D(w)LwQo. Now if we can show that D«w)LwQo -->QO), then by LA7, we
are done. Well if (W)LwQO, then Lw. QO, and by definition of wa ' QO. SO
(W)LwQ°-->Qo, and hence D«w)LwQ°-+QO). ~

We conclude this section on worlds with a proof of a lemma which will

be instrumental in Section 3. Again, let "w" be a restricted variable ranging
over the worlds.

LEMMA. (F°)(w)(x)(Lw[AyFO]x == LwFO).

That is, for any object x, x exemplifies being such that FO at W iff FO is
true at w.
Proof. Let QO, W 1 ' and b 1 be an arbitrary proposition, world, and object,
respectively. Note that by A-EQUIVALENCE, (x)( [AyQO]X == QO), and so
[AyQO]b 1 == QO. Thus, by 0 I, D([AyQO]b 1 == QO), i.e., [AyQO]bc=Qo. (-+)
Assume Lw,[AyQO]b 1 • Then by Theorem 3, it follows that Lw,Qo. (.-)
Assume L w, QO• Then by Theorem 3 again, it follows that Lw,[AyQO]b 1 • lEI

3. MODELLING LEIBNIZ'S MONADS 8

The investigation of monads is as philosophically satisfying as the

definition of truth and the investigation of worlds. Although it is unclear
what Leibniz intended his monads to be, they have traditionally been
regarded as properties of some sort. 9 However, we model them here as
abstract objects which are strongly possible (III, Section 4, D7VO Strongly
possible abstract objects have correlates "at" some possible world. For
example, Socrates' blueprint is a monad since it has a correlate at the
actual world. Intuitively, "compossible" monads have correlates at the
same world. So your blueprint and my blueprint are compossible. A monad
"mirrors" the world at which it has a correlate by encoding the vacuous
 THE APPLICATIONS OF THE MODAL THEORY 85

properties the correlate exemplifies - properties constructed out of the

propositions true at that world.
To make these ideas precise, we utilize the following definitions. As
with the previous lemma, we use "w" as a restricted variable ranging over
the abstract objects which satisfy the definition of a world:

D22 x is a correlate of z at w ("Cor (x, z, w)") = df(F)(r.wFx == zF).

That is, x is a correlate of z at w iff x exemplifies at w exactly the properties

z encodes.
D Z3 z appears at w ("Appear(z, w)") =df(3x)Cor(x,z, w).
DZ4 z is a monad ("Monad (z)") =df(:lw)Appear(z, w).
D Z5 z mirrors w ("Mirror (z, w)") = df(F°)(LwFO == LzFO).
Using the lemma at the end of Section 2, we now get the following result:

THEOREM 7. (z)(w) (Monad(z) & Appear(z, w)--+ Mirror (z, w)).

That is, every monad mirrors any world where it appears.
Proof. Suppose Z7 is a monad and Z7 appears at w7 • We want to show
for an arbitrary proposition QO, that LW7 QO == LZ7 QU. (--+) Suppose LW7 QO.
Since Z7 appears at w7 , it has a correlate there. Suppose b 7 is a correlate of
Z7 at w7 • SO Z7 encodes exactly the properties b7 exemplifies at w7 . In
particular, Z7 encodes [Jey QOJ iff b 7 exemplifies [Jey QOJ at w7 • By our
assumption, Lw 7 QO. So LW7[Jey QOJb 7, by the lemma proved at the end of
Section 2. So LZ7 QO.
(~) Suppose Lz 7 QO. Again, let b 7 be an object which exemplifies at W 7
exactly the properties Z7 encodes. Clearly b 7 must exemplify [Jey QOJ at
w7 • So by the lemma, Lw 7 QO. ~

Another interesting fact about monads is provable with the help of

Theorem 7:

THEOREM 8. (z) (Monad(z)--+(:l!w)Appear(z, w)).

That is, every monad appears at a unique world. 11
Proof Suppose Z8 is a monad. So there is a world, say w j , where it
appears. We want to show that WI is unique, so for reductio, suppose Z8
appears also at wz, W z f WI. Since the worlds are distinct, there must be some
vacuous property which distinguishes them. Without loss of generality,
suppose Lw1QO and ~ Lw 2QO. And since Wz is maximal, LW2 ~ Q O. But by
86 CHAPTER IV

Theorem 7, Z8 mirrors both worlds. So Lz 8 QO and LZ8 ~ QO. But this is

impossible, since QO and ~ QO would both be true in any world where Z8
has a correlate. 0

Since we know that every monad appears at a unique world, we are

entitled to talk about THE world where it appears. Let us use "m" as a
restricted variable ranging over the objects satisfying the definition of
monad. We then define:
D 26 wm =dJ(lw)Appear(m, w).
Theorems 7 and 8 allow us to say that every monad mirrors its worldY
Here now is a definition of com possibility.
D27 m1 is compossible with m2 ("Comp (m1' mJ") = dJ
(3w) (Appear(mp w) & Appear (m2' w)).
With these definitions, we have the following lemma: 13

LEMMA. (m 1)(m 2)(Comp(m 1,m 2 ) == wm1=wmJ

That is, two monads are compossible iff the worlds where they appear
are identical.
Proof. (--) Since m1 and m2 are compos sible, call the world where they
both appear woo By Theorem 8, W o = wm1 and W o = wm2 • So wm1 = wm2 • (+-)
Clearly, if the worlds where they appear are identical, there is a world
where they both appear. 129

With the help of this lemma, we get the following result: 14

THEOREM 9. (m 1)(m 2)(m3)(Comp (m1,m 1) & (Comp(m 1,m 2)--

Comp(m2' md) & (Comp(m 1,m 2) & Comp (m 2, m3)-- Comp(m 1,m 3))).
That is, compossibility is an equivalence notion among the monads.
Proof. Clearly, compossibility is reflexive and symmetrical. To show
transitivity, suppose Comp(m1' m2) and Comp(m2' m3). By the previous
lemma, wm1 = wm} and wm2 = wm,. So wm1 = wm,· 129

It should also be clear that by defining "embedding" as follows:

D 28 Zl is embedded in Z2 ("Embed(zl,z2)")=dJ(F 1)(z l F--z 2F),
we can prove that every monad has the world where it appears embedded
in it:
 THE APPLICATIONS OF THE MODAL THEORY 87

THEOREM 10. (m) Embed (w""m).

Proof. m mirrors its world Wm by encoding all the vacuous properties Wm
encodes. So wm must be embedded in m since the vacuous properties
exhaust the properties Wm encodes. I2$l
Consequently, every monad will be maximal with respect to the
propositions. But an even stronger claim is warranted - monads are
complete:
D 29 z is complete ("Com (z)") = df(F)(zF v zF),
where F =df[h ~ Fx].
THEOREM 11. (z)(Monad (z)--+ Com (z)).
Proof Clearly, if mo is a monad, then some object is its correlate at
the world where it appears. That object must exemplify there, for every
property, either it or its negation. Consequently, mo will encode, for every
property, either it or its negation. I2$l
Theorems (1)-(11) outline a certain picture of objects, monads, and
worlds. I believe that this picture is informative, and even insightful, in
its own right, independently of any potential it might have for understand-
ing Leibniz. The model has been described at a level of generality which
allows us to add a few constraints and investigate the submodels which
result. For example, we might want to investigate the structure which
results upon adding the hypothesis that there is at most one world where
nothings exists. Or we might want to look at the model which results
upon adding the hypothesis that abstract objects necessarily exemplify
any property they possibly exemplify which is not a relational property
with respect to possibly existing objects (the second hypothesis might
imply the first, but I do not think they are equivalent). And I think there
would be some interest in an investigation of E-monads, i.e., monads which
have correlates that exist at the worlds where the monads appear.
Finally, we consider how useful our model is for understanding the
work of Leibniz. Leibniz thought propositions were composed of concepts
and logical relationships. By concepts (or notions), Leibniz meant proper-
ties, things which mayor may not be exemplified by individuals. In addition
to the general concepts of being human, being red, etc., Leibniz supposed
there to be individual concepts. The concept Socrates, the concept
Alexander, the concept Adam, etc., are all examples of individual concepts.
These are not to be identified with properties like being identical E to
Socrates ([AX X = ES]), since these properties have individuals as constitu-
88 CHAPTER IV

ents. Leibniz preferred to develop a calculus of propositions and concepts

which did not have individuals as constituents (§12, "Elements of a
Calculus").
Leibniz took the logical relationship of concept containment to be
crucially involved in the analysis of categorical propositions. Containment
is involved in universal affirmative categorical propositions like "every
pious man is happy" (§7, "Elements of a Calculus"). The analysis of this
sentence is: the concept pious man contains the concept of being happy.
Leibniz seems to extend this kind of analysis to singular affirmative
propositions. In the "General Inquiries" (§ 16), he practically identifies the
sentential form "A is B" with the form "A contains B". In the "Discourse
on Metaphysics" (§8), he analyzes (1),
(1) Alexander is a king,
as: the concept Alexander contains the concept of being a king. He says
in this section,
"It is the nature of an individual substance, or complete being to have a notion so complete
that it is sufficient to contain, and render deducible from itself, all the predicates of the
subject to which this notion is attributed".

And later in the same work (§13) he says,

"the notion of an individual substance contains, once and for all, everything that can happen
to it".

Leibniz is not just claiming that the copula "is" should be read as
"contains", but rather that the very structure of singular affirmative
propositions is a relationship whereby the subject concept contains the
predicate concept. And so we find in the "Correspondence with Arnauld"
(May 1686), the following data destined for this analysis:
(2) Adam is the first man.
(3) Adam lived in a pleasure garden.
(4) Adam contributed a rib to Eve.
(5) Adam had two sons.
This analysis becomes puzzling when we consider how strong a relation
Leibniz intended containment to be. In the "Elements of a Calculus"
(§§7, 17), he says that "the subject concept, taken absolutely, ... , always
contains the predicate concept". Remarks such as this, and the general
tone of the discussion in the "Correspondence with Arnauld" leads us to
believe that if concept containment holds between two concepts, then it
 THE APPLICATIONS OF THE MODAL THEORY 89

holds in all possible worlds and is not relativized to any circumstance.

This puzzled Arnauld, as well as many subsequent philosophers, since it
seems to analyze our contingent data (1)-(5) as necessary truths.
It should be clear how we should proceed to unravel the puzzle. The
idea is that the Leibnizian concept is a conflation between property and
A-object, that proper names are ambiguously used sometimes to speak
about the individual and sometimes to speak about the blueprint of the
individual, and that encoding should represent containment. We may
suppose: (1) that Leibniz's general concepts just are properties, (2) that
Leibniz's individual concepts are our monads and in particular, that the
concept Adam is just Adam's blueprint, and (3) that Leibniz's analyses
utilizing concept containment, in the case of singular affirmative proposi-
tions, can be understood in terms of encoding. 15 We make these supposi-
tions precise by constructing the following definitions, utilizing "K" to
range over proper names.

D30 the concept F =dfF

D31 z is an individual concept ("IC(z)") = dfMonad (z)
D32 the concept K ("K") = df
(lz)(F)(zF == FK), i.e.,
(lZ) Blue (Z,IC)
D33 the concept IC contains the concept F = dfKF.

With these definitions, our theory begins to generate many predictions.

For example, it predicts:

THEOREM 12. IC(K).

That is, the concept K is an individual concept. So we are not just assuming
pretheoretically that the concept Adam is an individual concept, but rather
we prove it from more general assumptions.
Proof. For an arbitrary property P, we know KP == PK, by D32 and
A-DESCRIPTIONS. By the definition of W a , PK == LwfK. So KP == LwfK.
So there is a world where K appears, and hence, Monad (K). 1.81

The theory also predicts that Leibniz's analyses of (1)-(5) are necessary
truths. For example, his analysis of (1) is (1)':
(1)' The concept Alexander contains the concept of being a king
By D 33 , this just means:
90 CHAPTER IV

(1)" iiK ("the blueprint of Alexander encodes being a king")

By LAIO, this is a necessary truth.
The theory also predicts that all individual concepts are complete.
In addition to this positive evidence, there are passages in the letters
to Arnauld which support our suppositions. The word "Adam" is
sometimes used to talk about Adam ("the actual Adam") and sometimes
used to talk about the complete concept of Adam (one of "the many
possible Adams"). Leibniz was even aware of this problem (§14, letter,
May 1686; §12, letter, July 14, 1686). This ambiguity might also explain
why Leibniz calls monads both living and perpetual in §56 of the
Monadology. This could be symptomatic of conflating the blueprints of
persons with the persons themselves. On our understanding of monads,
we can see how they could be perpetual. By the AUXILIARY HYPO-
THESES, we suppose that monads do not have spatio- temporal location
and therefore are not subject to the laws of generation and decay. But it
is difficult to see how monads could be thought of as "living".
It also seems appropriate to suggest that our notion of mirroring could
serve to represent both Leibniz's notion of mirroring (in the M onadology)
and his notion of expression (in his letters to Arnauld). Leibniz says
repeatedly in the letters to Arnauld that every individual substance
expresses (in its concept) the universe into which it enters (May 1686; July
14, 1686). The concepts of individual substances must surely be individual
concepts (i.e., monads), and these "express" (i.e., encode all the vacuous
properties encoded by) the world into which it enters (i.e., where it appears).
These suggestions should anchor our model in the traditions of
Leibnizian scholarship. Any decision about its merit must be the outcome
of future discussion. But one word of warning is in order. When we claim
that the above model may be useful for "understanding Leibniz' ideas",
we are not claiming that the model is what Leibniz intended or had in
mind. The point of the exercise, as we see it, is to first theorize about the
way the world is and to then apply the theory by predicting some of the
things that Leibniz seemed to want to say. The model is to be judged by
how well it succeeds in helping us to explain why Leibniz may have said
some of the things he in fact said. And in the course of modelling his
ideas, we have found further evidence for supposing that the "is" of natural
language has a reading on which it means "encodes". Indeed, our work
suggests not only that there is a lexical ambiguity in the copula, but also
that there may even be a structural ambiguity in the form of singular
affirmative prediction itself. In the next section, we find an entire range
of data which could be explained by this latter hypothesis.
 THE APPLICATIONS OF THE MODAL THEORY 91

4. MODELLING STORIES AND NATIVE CHARACTERS

By adding a few primitives to the language of Chapter III, we may model

stories, and certain characters in them, as A-objects. First we add
abbreviations for any proper name of English which denotes an object
which, pretheoretically, we judge to be a story (for example, novels, myths,
legends, plays, dreams, etc.), or an author or character thereof (where we
take characters to be any story object, not necessarily animate). So we
shall have object names in our language which abbreviate "The Tempest",
"Shakespeare", "Prospero", "The Brothers Karamazov", "Alyosha", "The
Clouds", "Strepsiades", "Socrates", "Ulysses", "Joyce", "Bloom", "Dublin",
etc.
Secondly, we add the name of a new primitive relation which is of
central importance to our investigations - the authorship relation. The
formula "Axy" shall say that x authors y, and we trust that our readers
have at least an intuitive grasp on what it is to author something.
Consequently, we may define:

D34 Z is a story ("Story (z)") =dJ

(F)(zF ---+ Vac (F)) & (:3x)(E!x & Axz).

That is, stories are abstract objects which encode only vacuous properties
and which are authored by some existing thing. Hence, it is a contingent
matter that there are any stories. Lots of A-objects might have been stories,
however. To say this is to say that they encode just vacuous properties
and that POSSIBLY there exists an object which authors it.
Stories do not have to be consistent, nor do they have to be maximal.
But stories and worlds do have something in common - they encode only
vacuous properties. It is therefore appropriate to use our defined operator
"L" to talk derivatively about the propositions the stories encode. In fact,
if z is a story, then we may utilize "L z" as our translation for the English
prefix "according to (in) the story z". So when z is a story, "LzFo" says
that FO is true according to z. This allows us to prove an interesting
consequence of D34 which helps us to identify a given story: a story z is
just that abstract object which encodes exactly the properties F which are
constructed out of propositions true according to the story. That is,

THEOREM ("STORIES"): (z)(Story (z) ---+ t = (lz')(F)(z' F;:

(3FO)(L z FO &F= [AyFO])W 7.

For example, Little Red Riding Hood is a story, so it is that abstract

92 CHAPTER IV

object which encodes exactly the vacuous properties constructed out of

propositions true according to Little Red Riding Hood. Although this is
not a definition of "Little Red Riding Hood", we can identify this story
in so far as we have a good pretheoretical idea about which propositions
are true according to it. Fortunately, the data begins where our suggestion
ends, for we suppose that the data are intuitively true English sentences
of the form "according to the story, ... ". For example,18
(1) According to The Tempest, Prospero had a daughter.
(2) According to The Iliad, Achilles fought Hector.
(3) In The Brothers Karamazov, everyone that met Alyosha loved
him.
(4) In The Clouds, Strepsiades converses with Socrates.
(5) In Joyce's Ulysses, Bloom journeys through Dublin.
Thus, STORIES helps us to understand which A-objects might be denoted
by the underlined terms in the above sentences. We next try to identify
the denotations of some of the other terms.
We can say what it is to be a character of a story. Let us use "s"
variables as restricted variables ranging over stories:

D35 X is a character of s ("Char(x,s)") =dfC3P)L,FX.

That is, the characters of a story are the objects which exemplify properties
according to it. As we noted previously, the characters of a story are any
story objects, not just real or imaginary persons or animals. Note also
that this definition allows existing objects to be characters of stories - we
can tell stories (true or false) about existing objects, just as we can about
non-existent ones. 19
Of the non-existent characters in a given story, some will have originated
entirely from that story. We call these the "native" characters, and they
are to be distinguished from the other non-existent characters which may
have been borrowed or imported from other stories. But the non-native
non-existent characters are nevertheless "fictional", since, presumably, they
are native to (originate from) some other story.
We may define the notions of being native and being fictional by utilizing
a higher order primitive relation - one which could be analyzed in the
context of some other work. This is the relation that two propositions pO
and GO bear to one another just in case pO occurs (obtains, takes place)
 THE APPLICATIONS OF THE MODAL THEORY 93

before GO. We shall represent the fact that FO occurs before GO as

"Fo < GO". This relation helps us to be more specific about what it is to
originate in a story:
D36 x originates in s ("Origin(x,s)") =df
Char(x,s) & A!x & (y)(y')(s')(Ays & Ay's' & (Ay's' < Ays)
~ '" Char (x, S/)).
That is, x originates in s iff x is an abstract object which is a character
of s and which is not a character of any earlier story. We then define
being native and being fictional as follows:
D37 x is native to s ("Native (x, s)") = dfOrigin (x, s).
D38 x is fictional ("Fict(x)") =df(3s)Native (x, s).

So fictional characters are native to (originate in) some story. Clearly,

fictional characters may be characters of stories to which they are not
native. Sherlock Holmes is not native to The Seven Per Cent Solution. 2o
Nor is the monster Grendel, in John Gardner's recent account of the
Beowulf legend from the monster's point of view (Grendel). For simplicity,
we shall suppose that Achilles and Hector are native to The Iliad, even
though they may instead be native to some earlier epic of which no copies
have survived. Also, in what follows, we shall suppose that Prospero is
native to The Tempest, Alyosha and Raskolnikov are native to The
Brothers Karamazov and Crime and Punishment, respectively, Bloom is
native to Joyce's Ulysses, and Gregor Samsa is native to Kafka's
Metamorphosis.
It would be a philosophical achievement of great importance were
someone to discover a way of identifying fictional characters in general.
The best we can accomplish here is to present a means of identifying the
characters native to a given story. The identifying properties of native
characters are exactly the properties exemplified by that character in the
story. So we may utilize the following axiom which identifies the native
characters of a story as specific A -objects :21

AXIOM ("N-CHARACTERS"): (x)(s)(Native(x,s)~

x= (zz)(F)(zF == LsFx)).

For example, since Prospera is native to The Tempest, Prospera is that

abstract object which encodes exactly the properties Prospera exemplifies
according to The Tempest. This tells us an important fact about the
94 CHAPTER IV

Ls-operator and native characters - the Ls-operator "transforms" a pro-

perty a native character exemplifies according to story s into one which
the character encodes. That is, it is a theorem that: 22

(x)(s)(Native (x, s) --+ (F)(xF == LsFx)).

So if according to the play, Pro spero had a daughter, it follows that he
encodes having a daughter.
This theorem assumes greater significance in the presence of the
following axiom schema which also should govern the Ls-operator:

AXIOM(S) ("Ls-SUBSTITUTION"): where ¢ is any propositional

formula in which there occurs an object term 0 for which x is substitutable,
the following is an axiom: (S)(Ls¢ --+ LsE-h ¢~]o).

For example, in the myth, Achilles fought Hector. It therefore follows

from Ls-SUB both that in the myth, Achilles exemplifies the property of
fighting Hector and that in the myth, Hector exemplifies the property of
being fought by Achilles. From the supposition that Achilles and Hector
are both native to the myth in question, we may deduce that they encode
these properties, respectively, by N-CHARACTERS.23
With these definitions, axioms, and consequences, we can translate a
wide variety of data. We begin with (1)-(5) above. The translation
procedure is straightforward - since the Ls-operator is defined only on
proposition terms, we translate the English "in the story" using the
operator, and translate the rest of the sentence just as we would into an
ordinary predicate calculus:

(1)' LTempes, (3y)Dyp

(2)' LIliadFah
(3)' LBK(x)(Mxa --+ Lxa)
(4)' LCloudSCSl S2
(5)' LUlysseJbd.

There is an interesting class of sentences relevantly similar to (1) which

we should discuss briefly. These true sentences begin with the story prefix
and involve the predicative copula "is". For example, (6) and (7):
(6) According to Crime and Punishment, Raskolnikov is a student.
(7) In the Conan Doyle novels, Holmes is a detective.
 THE APPLICATIONS OF THE MODAL THEORY 95

Frequently, there are contexts in which it is acceptable to drop the story

prefix and just use the remainder of the sentence. We can think of the
resulting sentences "Raskolnikov is a student", "Holmes is a detective",
as true if we suppose that the English copula "is" can be read as "encodes".
We can therefore assimilate another phenomenon which is compatible
with our earlier discovery about the ambiguity of "is".24 In fact, it should
be clear that all data like (1)-(5) (and not just those involving the copula
"is") are subject to a structural ambiguity involving predication itself. In
the context of the story operator, that data must be translated as an
exemplification predication. Outside such a context, they must be under-
stood as encoding predications.
I think we can partially accommodate the views of philosophers who
object to (4)' and (5)' by arguing that the real Socrates and the real Dublin
are not characters of The Clouds and Ulysses, respectively. We do this
by supposing, instead, that the objects known as "the Socrates of The
Clouds", and "the Dublin of Ulysses", are the relevant characters of these
stories. We could suppose that these latter objects were native to these
stories and use N-CHARACTERS to identify them. Such a procedure
could be broadened to identify all non-native fictional characters. For
example, we could say that the Sherlock Holmes of The Seven Per Cent
Solution is native to that work, even though Sherlock Holmes is not.
The problem with this procedure is that one is forced to say something
about the relationships between the real Socrates and the Socrates of The
Clouds, between the Sherlock Holmes native to the Conan Doyle novels
and the Sherlock Holmes native to The Seven Per Cent Solution, etc. This
is no easy task. Clearly, the notion of weak correlation or embedding
would not be of much help - the Socrates of The Clouds exemplifies-
according-to- The Clouds (and consequently, encodes) properties not
exemplified by the real Socrates. A full discussion of the host of problems
which arise here would take us too far afield. Much further investigation
is warranted before this procedure is to be adopted. Let us then turn to
the next group of data.
(8) Santa Claus does not exist.
(9) Santa Claus might have existed.
(10) Franz Kafka wrote about Gregor Samsa.
(11) Some Greeks worshipped Dionysus.
(12) Pro spero is a character of The Tempest.
(13) Raskolnikov is a fictional student.
96 CHAPTER IV

Now it seems to me that there are two important readings for (8):
(8)' ~ E!se
(8") ~E! !se
(8') is provable, once we have used N-CHARACTERS to identify Santa
Claus. But if we symbolize the English word "exists" as we have done in
(8') as "E!", then (8) turns out to be a necessary truth, since A-objects
necessarily fail to exist. This conflicts with (9), however. (8) seems to have a
reading on which it is not necessary. We have captured this reading of(8)
with (8"). (8") asserts that no existing object exemplifies all the properties
Santa Clause encodes (recaIl D 10 ' Chapter II, Section 3, and the remarks at
the very end of Chapter III, Section 4). This clearly is a contingent truth.
This second reading of (8) is important for our understanding of (9).
We cannot represent (9) as "<)E!se" for its negation is provable from
N-CHARACTERS and the assumption that Santa Claus is a native
character. We capture the truth embedded in (9) as (9'):
(9') <)E!!sc
Surely in some possible world, there exists an object which exemplifies all
the properties Santa Claus exemplifies in the legend (let us assume Santa
Claus does not exemplify incompatible properties in the legend). These
remarks about the proper translations of (8) and (9) and the consequences
thereof apply to all other data similar to (8) and (9) which involve other
names of native characters described by consistent stories.
We translate (10) and (11) using exemplification formulas because they
involve extranuclear properties which A-objects may exemplify.
(10') Wks
(11') (3y)(Gy & Wyd).

Being written about and being worshipped are extranuclear properties (or
so I am supposing). They were not ascribed to (exemplified by) Samsa
and Dionysus in the relevant stories.
Given our work above, (12) should be translated as:
(12') Char (Prospera, The Tempest).
However, (13) is a more subtle case. Being fictional is a notion we have
defined - it may not be a property ("[Ax Fictional (x)]" is ill-formed). But
being a student is a property that Raskolnikov encodes, since he is native
 THE APPLICATIONS OF THE MODAL THEORY 97

to Crime and Punishment and exemplifies that property in the novel.

Consequently, we may define:
D39 x is a fictional student ("F -student (x)") = df
(3s)(N ative(x,s) & LsSX),
where "S" denotes being a student. Then from the assumptions that
Raskolnikov is native to Crime and Punishment and that he is a student
according to that story, we have (13') as a consequence:
(13') F-student(Raskolnikov).
In fact, we can generalize and suppose there is a whole group of notions,
each one defined with respect to a given property G:
D40 x is a fictional G =df(:ls)(Native(x,s) & ~sGx).
So Holmes is a fictional detective, Achilles is a fictional Greek warrior,
etc., given the appropriate assumptions. 25
Finally, we discuss definite descriptions. Consider (14) and (15):
(14) The detective who lived at 221 Baker st. in the Conan Doyle
novels is more famous than any real detective.
(15) In Crime and Punishment, Porphyry arrested the student who
killed an old moneylender.
It would be inappropriate to read the description in (14) as "the object
which exemplifies detectivehood, exemplifies living at 221 Baker St., and
exemplifies being a character of the Conan Doyle novels", since this
description fails to denote. But we often use the description in (14) to refer
to Holmes. The proper way to translate it is as "the object which according
to the Conan Doyle novels exemplifies both detectivehood and living at
221 Baker St". Using "MFT" to abbreviate "more famous than", and
other obvious abbreviations, we may read (14) as:
(14)' (y)(Dy & E!y->MFT(IX)~CD(Dx & Lx)y).
This says that every existing detective y is such that the object which
according to the Conan Doyle novels exemplifies both detectivehood and
living at 221 Baker St. is more famous than y.
A similar reading must be given to the definite description in (15). The
following would be the WRONG symbolization of (15):
~cpAp(IX)(SX & (:ly)(OMLy & Kxy)).
98 CHAPTER IV

The definite description fails to denote anything, even though it is entirely

within the scope of the story operator. There may not be an object which
exemplifies being a student and which killed an old moneylender. Or there
may be two. But there is exactly one object which according to Crime
and Punishment exemplifies being a student and killing an old moneylender.
Consequently, (15) is properly read as (15)':
(15)' LcpAp (ZX)Lcp(SX & (:ly)(OMLy & Kxy)).
When we read (hear) definite descriptions in the context of a story, there
is an implicit understanding that the description denotes a character of
the story. This implicit understanding is captured by placing the appro-
priate L-operator immediately after the iota-operator of the description.
This guarantees that the description, should it denote, denotes a character
of the story. (These remarks should also cast light on a very common
kind of definite description used in natural language: "the person who
allegedly ... ", "the man who, according to recent sources, ... ".)
To see this, consider the above example (15'). If we assume that
Raskolnikov is the object which according to Crime and Punishment is a
student who killed an old moneylender, we can show that Raskolnikov
is a character of that story. So assume (16):
(16) r = (IX)LcpSX & (:ly)(OMLy & Kxy).
By DESCRIPTIONS, it follows that according to Crime and Punishment,
Raskolnikov is a student who killed an old money lender, i.e.,
(17) LcpSr & (:ly)(OMLy & Kry).
By Lcp-SUB, it follows that Raskolnikov exemplifies being a student who
killed an old moneylender, i.e.,
(18) Lcp[h Sx & (:ly)(OMLy & Kyx)]r.
So there is a property which Raskolnikov exemplifies according to Crime
and Punishment. By D 31 , Raskolnikov is a character of that story. So by
placing the story operator immediately after the iota operator in the
description, we guarantee that the object denoted, if there is one, is a
character of the story.
Finally, note that (16) is a true identity statement. From (15') and (16),
it follows that according to Crime and Punishment, Porphyry arrested
Raskolnikov, i.e.,
(19) LcpApr.
 THE APPLICATIONS OF THE MODAL THEORY 99

Consider next,
(20) Ponce de Leon searched for the fountain of youth.
(21) According to the myth, the fountain of youth is in Florida.
(22) The fountain of youth might exist.
Where "M" names the relevant myth, "8" denotes the searching relation,
"E' denotes the being located in relation, "Y" names the property of being
a fountain the waters of which confer everlasting youth, "p" denotes Ponce
de Leon, and "I" denotes Florida, then we read (20) and (21) as follows:

(20') Sp(lX)LMYX.
(21') LML(lX) LM Yxf.

These readings are straightforward, given our earlier discussion. Since (22)
is true, we want to be sure not to capture it as an attribution of possible
existence to an abstract object. Recall the discussion of (8) and (9). The
English word "exists" as it occurs in (22) is not to be translated by our
primitive notion of existence. Instead, (22) must be understood along the
lines of (9) as (22'):

(22') OE! !(lX)LM Yx

The above results should establish at least a PRIMA FACIE case for thinking
that stories and characters are abstract objects. The groundwork has been
laid for further investigations which might fill in more details. In many
ways, declarative discourse among persons is like storytelling. It might be
worthwhile to regard a given discourse of an individual as a story. All
of the names and descriptions represent characters in the story. This might
facilitate suspension of belief when something false or suspicious sounding
arises. Various "eyewitness" versions of what happened in a given situation
constitute different stories. For each consistent story, worlds can be
described in which there are existing objects which exemplify there all of
the properties the characters of the story exemplify according to the story.
It seems to me that there are clear possibilities for future research here.

5. MODALITY AND DESCRIPTIONS

We now examine another class of English sentences which seem to be

true A PRIORI. They have the form "The F l' F 2' ... , F n is G", where G is
100 CHAPTER IV

necessarily implied by one of the Fi and where there is not an object

which (uniquely) exemplifies F l ,F2, ... ,Fw Here are some examples:
(1) The set of all non-self-membered sets is a set.
(2) The even prime number greater than two is not odd.
(3) The existent golden mountain has a shape.
For considerations similar to those in Chapter II, Section 2, we translate
the English definite descriptions as A-object descriptions. Except in these
cases, we translate "the F l , F z, ... ,Fn" as "(IZ)(G)(ZG =- Fl ~G V F z ~G
v ... v Fn ~G)", where "F ~G" means that necessarily, everything exem-
plifying F exemplifies G.
Consequently, the English descriptions in (1)-(3) are represented as
follows, using obvious abbreviations:
(a) (zz)(G)(zG =- [AXSX & (Y)(YEX =- Sy & y¢y)] ~G)
(b) (zz)(G)(zG =- [AxNx & Ex & Px & x> 2] ~G)
(c) (zz)(F)(zF =- E! ~F v G ~F v M ~F).
In the metalanguage, we signify this reading of the definite article as
"the."/,, and we assimilate the reading of the definite article proposed in
Chapter II, Section 2 to this reading. Let us abbreviate (a)-(c) respectively
as (IZ)t/!l - (IZ)t/!3' By A-DESCRIPTIONS, it follows that any property
satisfying the formula on the right of the biconditional in t/! is encoded
by the object denoted by the entire description.
Take (a) for example. Since the property of being a set is necessarily
implied by the property of being a set of non-self-membered sets, it follows
that (IZ)t/!lS. Our representations of (1)-(3) turn out to be theorems:
(1') (IZ)t/!lS
(2') (IZ)t/!2[h ~ Ox]
(3') (IZ)t/!3S,
This reading of the English definite article has another important
application. Philosophers since Russell have been puzzled by the following
two arguments:
(I) (4) Necessarily, the teacher of Alexander is a teacher.
(5) Aristotle is the teacher of Alexander.

(6) Necessarily, Aristotle is a teacher.

 THE APPLICATIONS OF THE MODAL THEORY 101

(II) (7) Necessarily, nine is greater than seven.

(8) Nine is the number of planets.

(9) Necessarily, the number of planets is greater than seven.

We seem to have conflicting intuitions about each of these arguments.

On the one hand, they both appear to be valid, since they seem to be based
on a simple application of the rule of identity elimination. On the other
hand, in each argument, the premises appear to be true and the conclusions
false.
Philosophers have explained the conflict in one of two ways, depending
on whether the English descriptions are taken to be contextually defined
(in the traditional Russellian manner) or taken to be complex terms
constructed with a primitive operator "the". Let us look at the Russellian
explanation first.
If we ignore the fact that "Aristotle" and "Alexander" are supposed to
be abbreviated descriptions, then a classic Russellian explanation of the
problem starts with the supposition that (4), (5), (8), and (9) have a complex,
rather than simple, logical form. Contexts in which descriptions appear
are systematically eliminated in favor of contexts in which existential
and uniqueness clauses make explicit the information implicit in the
description. And the present situation is further complicated by the fact
that when there are modal operators around, there is both a way to
eliminate the description so that the existential and uniqueness clauses
appear before the operator (wide scope) and a way to eliminate the
description so that these clauses appear after the operator (narrow scope).
Consequently, (4) and (9) each get two readings, whereas (5) and (8) each
get one. If we let "a" denote Aristotle, let ¢4 = ITxr -, (i.e. "x taught
Alexander"), and let ¢s =1 N x{ u Iu is a planet}' (i.e. "x numbers the set
of planets"), then the Russellian readings of (4)-(9) are as follows:

(4.1) (wide scope)

(4.2) D(x)«3Y)(¢4~ & (U)(¢4~-+U=y) & y=x)-+Tx)
(narrow scope)
(5.1) (3X)(¢4 & (Y)(¢4~-+y=X) & a=x)

(6.1) DTa

(7.1) 09>7
(8.1) (3x)(<ps & (y)(<Ps~-+ y = x) & 9 = x)
102 CHAPTER IV

(9.1) (:3x)(¢5 & (y)(¢5~ ~ Y = x) & 0 X> 7) (wide scope)

(9.2) 0 (x)(CjY)(¢5~ & (V)(¢5~ ~ V= y) & Y =x) ~ x> 7).
(narrow scope)
So Russellians claim that there are really two arguments to consider, one
invalid, the other valid, when accounting for the conflict of intuitions
about Argument I. The invalid argument has (4.2) and (5.1) as premises,
which assert, respectively, that in every possible world, if there is a unique
teacher of Alexander there, then whoever it is is a teacher, and that there is a
unique teacher of Alexander who happens to be Aristotle. These are both
true, but they do not jointly imply the falsehood (6.1) that Aristotle was a
teacher in every possible world. The valid argument has (4.1) and (5.1) as
premises, and they jointly imply the falsehood (6.1). But there is no cause for
alarm because (4.1) is false, since the person who in fact taught Alexander
did not teach Alexander in every world.
Both of the arguments which need to be considered in the case of
argument (II) have true premises, since both (7.1) and (S.1) are true. But
(7.1) and (S.l) jointly imply the truth (9.1), and do not imply the falsehood
(9.2). From the facts that necessarily nine is greater than seven and that
there's a unique object which numbers the set of planets and which happens
to be nine, it does not follow that in every world, if there is a unique
object there which numbers the planets in that world, then it is greater
than seven.
The Russellian explanation of the apparent validity of Arguments (I)
and (II) clearly works. The only trouble with it is that it doesn't preserve
intuitions some of us may share about the logical form of (4), (5), (S), and
(9). Some of us may share the intuition that the logical form of the sentences
which follow the adverb "necessarily" in (4) and (9) is rather simple. These
sentences seem to be atomic sentences with a complex subject term. And
some of us may share the intuition that (5) and (S) are simple identity
statements, constructed out of a primitive name, a complex term, and the
"is" of identity. These intuitions are not preserved when (4), (5), (S), and
(9) are represented in the Russellian fashion.
Philosophers who take these intuitions seriously will locate the source
of trouble in the above arguments somewhere else. These philosophers
will take the English definite article "the" as a primitive, represent it with
the Greek letter iota, and construct complex terms like "(IX)¢" to represent
the English descriptions. Since it is taken as data that (4) has at least one
 THE APPLICATIONS OF THE MODAL THEORY 103

true reading and that (9) has at least one false reading, these descriptions
must be interpreted as non-rigid designators. There just doesn't seem to
be any alternative for preserving the truth values of (4) and (9) in
exemplification logic with primitive descriptions, since the use of rigid
descriptions to represent the English descriptions in (4) and (9) would
yield readings on which (4) was false and (9) was true. So (4)-(9) are
customarily translated as follows, where the descriptions are not rigid:
(4 a) o T(zx) 4>4
(5a) a = (zx) 4>4
(6 a) OTa
(7a) 09>7
(8a) 9 = (zx) 4>5
(9a) o (zx)4>5 > 7.
On these representations, (4) and (9) are not ambiguous and there is no
question of wide and narrow scope for the descriptions - all the des-
criptions are within the scope of the modal operator. And this fact, it is
claimed, is just what is causing the trouble. The sentences "T(ZX)4>4" and
"(zx)4>5 > 7" may denote different propositions from world to world
because the descriptions in them may denote different objects from world
to world. So if one of the terms in a contingently true identity statement
like (5a) or (8a) can change denotation from world to world, it is illegitimate
to use the rule of identity elimination to substitute one of the terms for
the other inside a modal context. Identity elimination is a rule which will
preserve truth in modal contexts only if either the identity statement itself
is a necessary truth or both of the terms in the identity statement are rigid
designators (in which case, the identity statement will again be necessary).
So the second standard kind of explanation about the tension we feel with
respect to Arguments (I) and (II) is that they are, in fact, invalid. A properly
stated rule of identity elimination makes the inferences illegitimate.
This latter explanation also clearly works. However, it fails to preserve
the intuition some of us may share that Arguments (I) and (II) are simple
valid arguments. And some of us may share the intution that there is a
way to resolve the conflict without having to place "inelegant" restrictions
on the rule of identity elimination. A natural suggestion to make has
generally run up against a difficulty. The natural suggestion is to represent
104 CHAPTER IV

the English descriptions in these arguments as rigid designators. Then

(4)-(9) get translated just as (4a)-(9a) above, except the descriptions in
(4a), (5a), (8a), and (9a) denote in a given world, whoever it is that uniquely
satisfies the description in the actual world. On this interpretation of the
description, (4a) and (5a) logically imply (6a), because if in every possible
world, the person who taught Alexander in the actual world is a teacher
at that possible world, and if Aristotle is the teacher of Alexander in the
actual world, then it must surely follow that Aristotle was a teacher in
every possible world. And in a similar manner, (7a) and (Sa) logically imply
(9a), on this reading of the description. The rule of identity elimination
preserves truth no matter what the context.
As we previously noted, however, the difficulty with this proposal is
that it does not square with our intuitions that (4) seems to be true and
(9) seems to be false. Construing the descriptions in (4a) and (9a) as rigid
leaves (4a) false and (9a) true. There is no reason to think that the object
which taught Alexander in the actual world (i.e. Aristotle) was a teacher
in every possible world. Nor is there reason to think that there is a world
where the object which in fact numbers the planets (i.e., the number nine)
is less than seven. So we have not accounted for the intuitive truth values
of (4) and (9). In fact, these results have lead many philosphers to conclude
both that we cannot use rigid descriptions to represent these English
descriptions and that, therefore, we need to resolve our conflict of intuitions
about Argument (I) and (II) in one of the two ways outlined above.
These conclusions are not warranted however. There is a solution which
both allows us to use rigid descriptions to preserve the intuitive truth
values of (4) and (9), and allows us to resolve the conflict of intuitions
over Arguments (I) and (II). The solution is based on suppositions for
which we have found considerable evidence. They are that the English
"is" sometimes should be read as "encodes" and the English "the"
sometimes should be read as "thed ". It is then straightforward to claim:
that (4) and (9) are ambiguous, that the reading on which (4) comes out
true is that necessarily thed teacher of Alexander encodes being a teacher,
and that the reading on which (9) comes out false is that necessarily, the d
number of planets encodes being greater than seven. To be precise, let us
translate these new readings for the English descriptions in (4) and (9) as
(d) and (e), respectively:
(d) (IZ)(G)(ZG == [AX<P4J =G)
(e) (IZ)(G)(zG == [AX<PsJ =G).
 THE APPLICATIONS OF THE MODAL THEORY 105

Now, let us abbreviate the descriptions in (d) and (e) as "(IZ)1jJ 4'1 and
"(IZ)l/!s", respectively. We then propose the following translations for
(4)-(9), where all the descriptions are rigid:

(4a) o T(IX)¢4
(4b) o (IZ)1jJ4 T
(Sa) a = E(IX)¢4
(6a) OTa
(7a) 09>7
(8a) 9 = E(IX)¢s
(9a) o (lX)¢s > 7
(9b) o (lZ)l/!s [h x> 7].
These representations preserve all of the following intuitions: (i) that
(4) seems to say something true Gust consider (4b) and the fact that the
property of being a teacher is implied by the property of being a teacher
of Alexander); (ii) that (9) seems to say something false Gust consider (9b)
and the fact that the property of being greater than seven is not necessarily
implied by the property of being something which numbers the set of
planets); (iii) that the sentences following the adverb "necessarily" in (4)
and (9) have a simple logical form Gust consider the fact that (4a), (4b),
(9a), and (9b) all involve atomic formulas); (iv) that Arguments (I) and (II)
are simple valid arguments based on the simple rule of identity elimination
Gust consider that identity elimination works unrestricted in all contexts,
and legitimately takes us from (4a) and (Sa) to (6a), and from (7a) and (8a)
to (9a)).
Of course the reader may not share these intuitions and consequently
may not be moved by these results. Or the reader may have further
intuitions about Arguments (I) and (II) which have not been preserved. If
either of these are the case, then we may at least claim to have shown
that we are not forced to accept the two traditional ways of explaining
certain conflicting intuitions we have about Arguments (I) and (II).
An obvious plan for further investigation is to try to find data which
involve descriptions and which force us to use non-rigid descriptions to
preserve the intuitive truth values. Rigid descriptions clearly prove useful
for understanding why (10) seems true:
106 CHAPTER IV

(10) The inventor of bifocals might not have invented bifocals.

A straightforward translation of (10) can capture it as an atomic sentence.
Consider (10'):
(10') [Ax O",IX](ly)Iy.
Since "(ly)Iy" has a denotation, we may use AE to prove (11) from (10'):
(11) 0", I(ly)Iy.
(11) would be false were the description non-rigid, since in no possible
world would the person that invented bifocals in that world fail to invent
bifocals. So we're forced to use a rigid description to capture the truth in
(10) (though if we are prepared to give up the intuition that it has a
simple logical form, we could use a Russellian elimination in which the
description gets wide scope to get a true reading). The question is though,
will rigid descriptions and the logic of encoding always suffice?
 CHAPTER V

THE TYPED THEORY OF ABSTRACT OBJECTS

The typed version of our theory commits us not only to abstract objects,
but also to abstract properties, abstract relations, abstract properties of
properties, abstract properties of relations, etc. We can use these entities
to model impossible relations, like the symmetrical, non-symmetrical
relation, and fictional relations, like simultaneity.! However, the primary
motivation for developing the typed theory is to account for the data
concerning the propositional attitudes and to model the fictional relations
of mathematics.
The verbs of propositional attitude (e.g., believes, knows, desires, hopes,
expects, discovers, etc.), often combine with the word "that" and an English
sentence to produce logically problematic predicates like "believes that
Cicero was a Roman", and "hopes that Kennedy is elected President."
Frege noticed that terms (simple, complex names) inside these pro-
positional attitude constructions exhibit rather strange behavior. In
particular, Frege noticed that from the fact that someone believes that
... 1"! ... , it doesn't follow that they believe that ... 1"2' .. , even when 1" 1 = 1" 2
(where ... 1"1 ... is any English sentence in which term 1" 1 occurs, and ... 1" 2...
is the result of replacing one occurrence of 1" 1 with 1" 2)' 2 For example, each
of the following triads of English sentences is consistent:
(1) S believes that Cicero was a Roman.
(2) S does not believe that Tully was a Roman.
(3) Cicero is Tully.
(4) S believes that Socrates was the teacher of Plato.
(5) S does not believe that the son of Phaenarete was the teacher
of Plato.
(6) Socrates is the son of Phaenarete.
(7) S believes that x is French fire engine blue.
(8) S does not believe that x is Crayola crayon blue.
(9) French fire engine blue just is Crayola crayon blue. 3
It seems that the law of identity elimination (= E) does not preserve truth
when applied to terms in propositional attitude contexts, and this

107
108 CHAPTER V

constitutes the problem of "the logically deviant behavior of terms in

intermediate contexts".
If in a given case, the law of identity elimination appears to fail,
philosophers call the belief (context) DE DICTO, and distinguish it from a
belief (context) DE RE, in which identity elimination preserves truth. When
S's belief is DE RE, it does follow from the facts that S believes that ... , 1'"
and '1 =,z, that S believes that .... 'z ....
To account for this phenomenon of DE DICTO propositional attitudes,
Frege theorized that there must be distinct entities, "senses", associated
with the terms, 1 and, 2' These entities lend the terms with which they
are associated information, or cognitive, value by serving somehow to
RE-present the object or relation denoted by the term. This "mode of
presentation" embodied by the sense of the term stores information about
the denotation of the term, assuming it to have one. And it is the sense
of the term which the term denotes when it is situated in a DE DICTO
context. Frege would argue that identity elimination is a perfectly good
rule of inference; it is just that English terms are ambiguous, and have
different denotations when they are in and out of DE DICTocontexts. Identity
elimination preserves truth when you substitute terms which have the
same denotation.
Using the theory we have so far, we could construe the senses
of English names and descriptions which denote objects as abstract objects.
An association of abstract objects with English terms would allow us to
picture how a given term had "information" or "cognitive" value. Abstract
objects could "RE-present" an object denoted by a term by encoding
properties the object exemplified. They could serve to store information
by encoding many such properties. Finally, they could serve as the
denotation of the term when the term is located inside DE DICTO contexts.
Such an association between terms denoting objects and abstract objects
is one of the most important features of the language developed in this
chapter. We use this language to translate data similar to (1)-(6) in
Section 1 of Chapter VI. However, (7)-(8)-(9) constitute an example of
the DE DICTO phenomenon with respect to English names which denote
relations. "French fire engine blue" and "Crayola crayon blue" are names
of certain properties - properties which we could suppose are identical.
In order to account for the logically deviant behavior of these names, we
associate with them abstract properties - properties which encode pro-
perties of properties. These abstract properties can lend property names
 THE TYPED THEORY OF ABSTRACT OBJECTS 109

their information value - they could store information about the properties
denoted by such names by encoding properties of them. And these abstract
properties can serve as the denotation of these names when the name is
located in a DE DICTO context.
Similarly with English names which denote relations - we utilize
abstract relations, relations which encode properties of relations, to serve
as their sense. A completely general account of the senses of names of
relations in the type hierarchy requires that we have abstract entities at
each type which encode properties of the entities of that type. This is by
far one of the most interesting applications of the typed version of our
theory.
In what follows, we shall use the word "object" in a new manner.
The things which we have been calling "objects" will now be called
"individuals". We shall now use the term "object" to discuss any kind of
entity whatsoever - existing and abstract individuals, existing and abstract
properties and relations, existing and abstract properties of properties
(relations), etc. Thus, we call the developments in the next few pages "the
typed theory of abstract objects", and we affectionately refer to it as
"metaphysical hyperspace".

l. THE LANGUAGE

We first recursively define the set of types. 4 For our purposes, we may
think of types as symbols which serve to simultaneously categorize both
the terms of the language and the entities denoted by those terms. The
set of types, then, is the smallest set, TYPE, which satisfies the following
conditions:
(1) "j"ETYPE.
(2) "p"ETYPE.
(3) Whenever tl, ... ,tnETYPE, then'(tl, ... ,tn)/p'ETYPE.
Intuitively, "i" is the type of individuals and "p" is the type of propositions.
"(t l , ... , tn)/P" is the type ofrelations whose arguments have types t l , ... , t"'
respectively. In what follows, we drop the quotation marks around the
type symbols. The properties (and the expressions which named them)
which we used in Chapters I-IV were of type i/p. The relations were of
type (i, ... , i)/p. But now we have an infinitely branching hierarchy.
110 CHAPTER V

A. PRIMITIVE TERMS

Officially, we use ai, a~, ... as names, and xi, x~, ... as variables for objects
of each type t. These are the only primitive terms of the language. However,
the following conventions shall hold. Whenever a, b, c, . .. and x, y, z, ...
appear without typescripts, they denote (range over) individuals (unless
their first occurrence in a formula has a typescript, in which case it shall
be understood, if they appear later in the formula without typescripts,
that the typescripts have been omitted for convenience). Also, we use
p(t 1 ..... tnJ/P, Q(t1 ..... tnJ/p, ... and F(t lo .... t nJ/P, G(t 1 ..... tnJ/ p, ... as names and variables
for objects of relational types (t 1 , ... , tn)/p. And we use pP, QP, ..• and
FP, GP, •.. as names and variables for objects of type p.
It will be convenient to distinguish certain names for special purposes.
We use E!t/p as the existence predicate for objects of type t. We use = E'
as the E-identity predicate for objects of type t. We use Tr P/p as the explicit
truth predicate for propositions. We use Ex«t, .....tnJ/P.t, ..... tnJ/p as the explicit
exemplification predicate, for all types tl'"'' tn' We use BY·PJ/p, B~·PJ/p, ...
to translate the verbs of propositional attitudes. Finally, we use R(t.t.iJ/p
as the representation predicate - some objects of type t will represent other
objects of type t for an individual of type i.
In addition to these terms, we utilize our usual list of logical and
grammatical symbols: connectives: ~,~; quantifier: V; lambda: A; iota:
I; box: 0 ; and parentheses and brackets: (,), [,]. We add to this list a
one-place sentential operator: that-.

B. FORMULAS AND TERMS S

We simultaneously define (propositional) formula and term of type t. The

definition has eight clauses and is rather complex. We reserve the extended
comments and the examples until after the definition.
(1) All primitive terms of type t are terms of type t.
(2) Atomic: If , is a term of type p, then, is a (propositional)
formula.
(3) Atomic exemplification: If p is a term of type (t l , ... ,tn)/P, and
'l''''''n are terms with types tl, ... ,t., respectively, then
p, 1"'" 'n
is a (propositional) formula.

In atomic exemplification formulas, we call p the initial term and, l' ... , 'n
 THE TYPED THEORY OF ABSTRACT OBJECTS III

are called argument terms. Primitive terms of type p -will also be called
initial terms. Initial variables are variables which are initial terms
(4) Atomic encoding: If p is a term of type t/p and T is a term of
type t, then Tp is a formula.
(5) Molecular, Quantified, and Modal: If ¢, t/J are any (proposi-
tional) formulas and rx is any variable of type t (which is not
an initial variable somewhere in ¢), then (~ ¢), (¢ ---+ t/J), (I;;Irx)¢,
and (O¢) are (propositional) formulas.
(6) Complex higher order terms: If ¢ is any propositional formula
and rx l , ... , rxn are any variables with types t l , ... , tm respectively,
such that none of the rxi'S are initial variables somewhere in ¢,
then [Arx l ••. rxn ¢] is a term of type (t 1 , ••• , tn)/p and both ¢
itself and that-¢ are terms of type p (it will sometimes be
convenient to regard ¢ as the degenerate A-expression [A¢]).
(7) Sense t~rms: If /(t is any primitive name of type t, and (J is any
primitive term of type i, ~ is a term of type t.
(8) (Sense) Descriptions: If ¢ is any (propositional) formula with
one free variable x of type t, then (zxt)¢ ((~) is a term
of type t.

In the usual manner, we define:

D1 xt is abstrad/ p ("A !t/p xt") = dJ [Ai 0 ~ E!t/Py]x t.
We shall use zt-variables to range over abstract objects of type t.
By inserting the parenthetical remarks in clauses 2, 3, 4, and 5, we obtain
a definition of propositional formula. Essentially, a formula ¢ is proposi-
tional iff ¢ has no encoding subformulas and none of the initial variables
appearing in ¢ are bound by a quantifier. Clearly, (:3G i/P)(X iG & ~ Gx)
and (Gi/p)(XiG---+ Gx) fail both of these restrictions. However, (VFi/P)Fx i,
(3F i/P)(G i/PXi & ~FXi), (I;;IFP)(FP v ~FP), and (3GP)(F i/PX i & ~GP) all fail
the second restriction. As we noted in earlier chapters, the second restriction
allows us to simplify the semantics. But now that we have a type theory,
there are other ways to express the propositions these latter four formulas
seem to express. For example, we could use the explicit exemplification
predicate of type (i/p,i)/p and suppose that (VFi/P)Fx i was simply an
abbreviation for (1;;1 Fi/P)ExFx i. This latter formula is propositional, and
says that all properties of individuals Fi/p are such that they bear the
112 CHAPTER V

exemplification relation of type (i/p, i)/p to Xi. Also, we could use the
explicit truth predicate and suppose that ('if P)(P v ~ P) abbreviated
('if P)(TrP v ~ TrP). The latter formula is propositional and says that
all propositions P are such that either P exemplifies the property of
being true or fails to exemplify this property. In general, we can always
reconstitute a propositional formula from a formula which fails just the
second restriction on propositional formulas. 6 This lessens the significance
of the second restriction.
In clause 6, there is an additional restriction placed on the formulas cP
which may appear behind a A: variables bound by the A must not appear
as an initial variable somewhere in cPo Techniques similar to the above
allow us to reconstruct a propositional formula satisfying this restriction
from one which fails it. For example, [},Fi/p Fx i], an expression which
fails this restriction., could abbreviate [AFi/p ExFx i], where the Ex
predicate is of the appropriate type. [AP P -+ P] could abbreviate
[APTrP -+ TrP]; the latter denotes the reflection of the conditional-
ization of the property of being true with itself. Abbreviational procedures
such as these lessen the significance of the restriction in clause 6. In fact,
these abbreviational techniques allow us to construe any A-expression
[AlX l ... lXncP] in which cP lacks encoding subformulas either as a well-formed
A-expression or as an abbreviation of a well-formed A-expression. But they
do not allow us to construe [AXi(:JFi /P)(xF & ~ Fx)] or [Axi(Gi/P)(xG-+
Gx)] as abbreviations of well-formed A-expressions.
Clause 7 of the above definition gives us a means for denoting the
abstract object an individual associates with a given English proper name
as its sense. We suppose that the sense of a name varies from person to
person (see Chapter VI, Section 1). For example "Socrates" and "Frege"
are names oftype i, so "Socrates Frege" is a sense term of type i. It shall denote
the abstract individual which serves as the sense of the name "Socrates"
with respect to Frege. "French fire engine blue" is a name of type i/p (since
it names a property of individuals). So "French fire engine blue Frege" is a
sense term of type i/ p and will denote the abstract property which serves
to represent the property of being French fire engine blue to Frege. This
abstract property encodes properties of type (i/ p)/p, i.e., properties of
i/p-properties.
Clause 8 of our definition simultaneously gives us both descriptions and
sense descriptions. Where "T" denotes the (i, i)/p-relation of teaching, and
"p" denotes Plato, (lx)Txp reads "the teacher of Plato". Where "C" denotes
 THE TYPED THEORY OF ABSTRACT OBJECTS 113

the (i/p)/p-property of being a color, "P" denotes the preference relation

of type (i, i/p, i/p)/p, and "m" denotes Mary, (zxi/p)(Cx & ~ (3y)
(C y & Pmyx)) might read: the color Mary prefers to all others (i.e. "Mary's
favorite color").
If the formula ¢ used in constructing the description is propositional,
then ~ is called a sense description. They will help us to model the
senses of English definite descriptions. UKJs/l shall end up denoting the
abstract object of type t which encodes just the property [hI ¢ & (y)
(¢~-'> y= EX)] ("being the unique ¢"). For example, (zxi)Txp shall denote
the abstract individual which encodes just the property of being the teacher
of Plato. When we concern ourselves specifically with the fact that the
English description "the teacher of Plato" exhibits logically deviant
behavior inside DE DICTO attitude contexts, we shall translate the English
as we normally would into the standard type theoretic language and then
underline it. By doing so, we will have formed an expression of our
language which denotes the sense of the English description.
Finally, we say that r is a term iff there is a type t such that r is a term
of type t.
2. THE SEMANTICS

A. INTERPRETATIONS

An interpretation, .1', of our type theoretic language is any octuple,

>,
<"fII, too' q;, extw' 2, ext.91, oen, ff which meets the conditions described
in this subsection. The first member of .1' is a non-empty class, "fII, called
the class of possible worlds. The second member of .1', too' is a member
of "fII and is called the actual world. The third member of .1', q;, is a
non-empty class called the domain of objects. q; is the union of a collection
of non-empty, indexed, classes, i.e., q; = UIETYPEq;!" Each class in the
collection, q;t, is called the domain of objects of type t. We call q;i the
domain of individuals, q; p the domain of propositions, q; 1/ p the domain of
properties of type t objects, q;(ll, ... ,tn)/p the domain of n-place relations among
objects with types t 1 , ... ,t., respectively. We use "~/" as a metalinguistic
variable ranging over the objects in q;1'
For convenience, we call the class of all objects with types not equal
to i the class of higher order objects and we use "~" to denote this domain.
So ~ = Ultiq;t. ~ is closed under all the logical functions specified in 2,
the fifth member of an interpretation. ~ may be subdivided into domains
114 CHAPTER V

of relational types Pll(tl, ... ,tn)/p and the domain of propositions Pllp. We
use ",zt" as a metalinguistic variable ranging over the higher order objects
of type t.
We also let "sI/' denote the class of abstract objects of type t.
d t = {c/I (w)((/¢ext",(ff(E !t/P)))}, where ext", and ff are the fourth and
eighth members of the interpretation, as defined below. We use "0/" as
metalinguistic variables ranging over the members of sit.
The fourth member of Jf, ext"" is a function defined on Pll x 1f! and
indexed to its second argument as follows:
(a) ext",: Pll(t" ... ,tn )/p x 1f! ~~(.@tl X .@t2 X .•• X .@tJ
(b) ext",: Pllp x 1f! ~ {T,F}.
Thus, the ext", function distributes an exemplification extension at each
world to all the higher order objects.
The fifth member of Jf, !I!, is a class of logical functions with members:
[l}!I!O!t"§j' O!tJVJf1/), ~@JV1/},k' PllIff§'!I!j,k' "f/sI~j,t', ~@JV'@, JVIff"§, and
JV Iff~. These functions are defined as follows:

(a) [l} !l!O!t"§j is a function from (UI ,sj,s x .@Ijinto

nPJl(tl, ... ,lj , ... ,ln)/P)
(Ul,sj,snPll(tl, ... ,tj_l,lj+l, ... ,ln)/p)uPllp subject to the following
conditions:

(b) O!tJVJf"f/j is a function from Ul5,j5, nPll(tl, ... ,tj, ... ,tn)/p into
(U 1 5,j,sn Pll(ll, ... ,tj_ 1, Ij+ 1, .. . ,tn)/P) U Pll P subject to the conditions:
(1) for n> 1, extu'cO!tJVJf"'fj(,z(tl, ... ,ln)/P)) =
{«/1, ... , ot j - 1 , olj+ " ... ,oln) I(If .(/j)
« ot 1 , ••• , otj-l, ot j , olj+ " ... , otn) Eext",(,z(tl , ... ,tn)/P))}
(2) for n = 1, ext,JO!tJVJf"f/1(,zll/P)) =
T iff (\fotl)(otl Eext,J,zll/P))
{
F otherwise.
(c) ~@JV"fj,k is a function from into
Ul,sj<k,snPll(tl, ... ,tj, ... ,Ik, ... ,ln)/P
U I5,j<k,sn Pll (t" ... ,Ij-l, Ik, Ij+ 1, ... ,tk-l, Ij,tk+ " ... ,In)/p subject to the
 THE TYPED THEORY OF ABSTRACT OBJECTS 115

following condition:
extu,(~(1}';v "fIj,iI[(11 ,... ,lnJ/P)) =
{ ( ()II , ••• ,QIj_1 ,.0Ik ,0Ij+1 , ... ,.0- tk-I ,.0Ij ,0Ik+l , •.. ,0In)1

< ,(/
II
, ••• , £}
Ij
, ••• ,,{1
Ik
, ••• ,,(}-
In)E ,ext w (h(tI, ... ,lnJ/P)}
'£- •

(d) fYllfffii !i!j,k is a function from Ul S;j<ks;nfYl(II, ... ,lj, ... ,lk ..... lnJ/P into
Ul S;j<ks;nfYl(II, ...• ljo ... ,lk_I,Ik+ 1, .... lnJ/p subject to the following
condition:
extto-(fYllff fii !i!.),k (1[(llo ... ,lnJ/P)) =
{ <,otl, ••. ,.otj, ... ,otk-l,otk+l, ... ,otn)1

.{/ , ... ,0(/~tj"." Olk '"0' oln )Eextto(1[(t"

( ~tl ... ,lnJ/P) and olj = Olk} •

(e) "fI d~j,I' is a function from (U1S;jS; n+lfYl(tI, ... ,lj .... ,tnJ/P)
U fYl pinto (U
1 s;js; n+ 1 fYl (t, ... . ,Ij_ 1,1' . lj,lj+ I, •. . ,tnJ/p) U fYlI'/p
subject to the conditions:
(1) ift=(tl, ... ,tj, ... ,tn)/p, then 1 ~j~n+ 1 and
ex t eo ("'/,_40;;
'
(hl))_{<
.,Y$ 'TO j, t' '(, -
I,
. 0 , .•. , .()
Ij_1
,,(},,o,
I' Ij
,0
Ij+1
, ••• , 0
In)1

<
011, ••• , oh, ... , otn )Eext",(I[t)}

(2) if t = P, then j = 1 and ext j "fI d~ 1,1'(I[P)) =

{,/'lextjI[P) = T}.

(f) ~(1}';vE!& is a function from fYl x fYl into fYl subject to the
following conditions:
(1) if t = (t 1, ... , tn)/p and t' = (t~, ... , t~)/p, then
extw(~(D';vE!&(I[I"l)) =
{ ( .0 I I , ••• ,.0- In ,,0 I, " •• ,.0
I;" ) 1< ,0
II, ••• ,.0In)
¢ext,jl[l) or <ol\, ... ,ol;")Eext,J:/)}

(2) if t = (t 1, ... , tn)/p and t' = P, then

ext", ('{?(D';vE!& (1[1, ,l")) =
{<o\ ... , oln) 1<0\ ... , oln)¢ex(u(l[l)
or ext",J/') = T}
(3) if t = P and t' = (tl'"'' tm)/p, then ext,J'{?(D%~ (1[1 J,)) =
{<
011 , ... , Ol~) lext,j1[1) = F or

<
011, •.• , olm )Eext",C,l)}

(4) if t = P and t' = p, then ex()'{?(D%~(I[t,jl')) =

{ T iff extw(l[t) = For extw(.,/') = T
F otherwise.
116 CHAPTER V

(g) JII rffqJ is a function from fll into fll subject to the conditions:
(1) if t = (t 1 , ••• , tn)/p, then extjJII rffqJ(-z~) =
{< ,,I,, ... , ;;/n) 1<ot" ... , otn)¢extj-zt)}
. {T iff ext (-zP) = F
(2) If t = p, then ext,JAf@"~( i P)) = ~
F otherwIse.

(h) JII rff~ is a function from fll into fll subject to the conditions:
(1) if t = (t 1 , ••• , tn)/p, then ext,jJII rff~(i)) =
{< <
ot" ... , otn) 1(£0')( 0\ ... , otn) Eext"Att))}
Tiff (£O')(ext ,(-zP) = T)
(2) ift=p, then extjJllrff~(-zP))= { h' W

F ot erWIse.

This completes the definitions of the logical functions. The sixth member
of ~,ext$ ' is a function defined on UtETYPEfllt/P' For a given type t, ext $
maps !!llt/p into 'PU~t). ext $ assigns each higher order property of t-objects
an encoding extension among these objects.
Let JIlt be the set of primitive names of type t of our language. Then,
the seventh member of § is the sense function, oen, which maps [0i x JIlt
into sit (the set of abstract objects of type t). For convenience, we index
the Oen function to its first argument. Thus, for a given individual 0, oenu
associates with a given name Kt of type t an abstract object of type t. We
call oenu(K) the sense of K with respect to 0. 7 Intuitively, if "Socrates" is a
name of type i, then OenY;,ege("Socrates") is the abstract individual which
serves as the sense of the name "Socrates" with respect to Frege. We shall
assign this object to the sense term "Socrates Frege ". And we shall make it
a logical truth that Socrates Frege represents Socrates to Frege. 8 We shall
sometimes index the sense function to the type of the name upon which
it is operating. For example, oenP:ege("French fire engine blue") is the
abstract i/p-property which serves as the sense of "French fire engine blue"
with respect to Frege.
The eighth member of § is a function, ff', defined on the primitive
names AND on the closed sense terms of the language. For each name Kt
of type t, ff'(K t )E[0t. For each closed sense term K! of type t, ff'(K!) =
oeny;(u) (Kt). Recall that sense terms can have only primitive terms as
subscript. So the closed sense terms will have only primitive names as
subscripts. 9
In addition, we place the following three restrictions on ff':
 THE TYPED THEORY OF ABSTRACT OBJECTS 117

(1) <
.ext j.?(Ex«t 1 ,"" tn)(P,t 1",.,tn)(P)) = { ,z(t " ... ,tn)(p, ,/1, ... , otn >\
<011, ... , otn) E.ext j,z(I" ... ,ln)(P)}.
So ff must assign to the explicit exemplification predicates relations with
the "appropriate" extensions.
(2) .extw(.?(Tr» = {,zP\.extw(,zP) = T}.
Here too, .? must assign to the explicit truth predicate a property of
propositions with the appropriate extension.
(3) .ext j.?(R(t,l,i)/p» = {< a,1, or, oi) \(3KO')(ff(KO') =,a} &
.?(K) = ot & F(a) = oi)}

Thus, "R(I,I,i)/p" denotes a three place relation which objects a,t, 0 1, and oi
bear to one another just in case there is some (closed) sense term KO' such
that a,t is the sense of K with respect to oi and 0 1 is the denotation of K.
We say that a,t represents ot with respect to oi.

B. ASSIGNMENTS AND DENOTA TlONS

For the most part, the definitions partitioning the A-expressions are similar
to those developed in Chapter III, Section 2, B. However, we need to type
the added place in the definition of vacuous expansion. We also need to
concern ourselves with argument variables (rather than the "object"
variables of Chapter III) throughout these definitions.
If J1 is an arbitrary A-expression, [Aa l ... an¢], J1 is classified as follows:
(1) If (3j) (1 ~j ~ nand aj does not occur free in ¢ and t' is the type of
aj andj is the least such number), then J1 is the j, t'-vacuous expansion of
[Aa l '" aj - l aj + I'" an¢].
(2) If J1 is not a j, t'-vaCUOllS expansion, then if (3j)(1:5,j:5, nand aj is
not the P free argument variable in ¢ and j is the least such number), then
where rlk is the P free argument variable in ¢, 11 is the j, kth -conversion of
[Arl l ... rl j - l rlkrl j + 1 ... rl k - 1 rlfXk+ 1 ... rl n¢].
(3) If 11 is neither of the above, then
(a) if ¢ = (~I/I), 11 is the negation of [Arl l ... rlnl/l]

(b) if ¢ = (1/1 ~ X), and t/J and X have no free argument variables
in common, then where rl 1 , ..• , rlm are the variables in 1/1 and
rlm+ 1"'" rln are the variables in X, 11 is the conditionalization
of [Arl 1 ... rlml/l] and [Aa m + 1'" rlnX].
118 CHAPTER V

(c) if ¢ = (V f3)ljJ, and f3 is the r

free argument variable in ljJ,
then f1 is the r-universalization of [;tal'" aj- I f3 apj+ I'" anljJ].
(d) if ¢ = (0 ljJ), then f1 is the necessitation of [A,a I ... anljJ ].
(4) If f.1 is none of the above, then if (3j) (1 sj s nand Cl j occurs
free in more than one place in ¢ andjis the least such number), then where:
(a) m is the number of free argument variables between the first
and second occurrences of aj ,
(b) ¢' is the result of replacing the second occurrence of a j with
a new variable f3 (with the same type as a),
(c) k=j + m + 1, then
f1 is the j, kth-reflection of [A,a l ... aj+mf3ak'" an ¢'].
(5) if f1 is none of the above, then if"i is the leftmost argument term
occurring in ¢, then where
(a) k is the number of free argument variables occurring before "i,

(b) ¢' is the result of replacing the first occurrence of r by a new

variable f3 (with the same type as r),
(c) j = k + 1, then
f.1 is the r-plugging of [A,Ct! ••• Ct k - I f3Ct j" • Ct n¢'] by r.
(6) If f1 is none of the above, then
(a) ¢ is atomic
(b) Ct l , ... , Ct n is the order in which these variables first occur in ¢,
(c) f.1 = [A,a l ... CtnpnCtI ••. Ct n], for some term pn, and
(d) f1 is called elementary .

.Y-assignments. If given an interpretation .Y of the language, an

.Y -assignment will be any function, f, defined on the primitive variables
of the language such that when Ct is a variable of type t, /(Ct)E£0 t .
Denotations. If given an interpretation .Y and an .Y-assignment f, we
recursively define the denotation of term r with respect to.Y and f("d",/(r)")
as follows:
(1) where K is any primitive name, d,,'/K)='?(K)
 THE TYPED THEORY OF ABSTRACT OBJECTS 119

(2) where K" is any closed sense term, d5)K)=~'¥(K)

(3) where IX is any primitive variable, dJ,/(IX) = /(IX)
(4) where ~~ is any open sense term of type t,
ocnd,f ,/("iKt )
d J ./ts:~) =
(5) where J1. is an elementary A-expression [AlXl , .. IXnP IX 1 •• • lXn]
d 5 )J1.)=dJ ,/(p)
(6) where J1. is the r-plugging of ~ by r,
d 5 ,/(J1.) = f!J 2o/f<;§j(dJ,/(~)' dJi',/( r))
(7) where J1. is the /h-universalization of ~, d J ,/(J1.)=
Olt JVJl' ij(d J,/(~»
(8) where J1. is the j, k1h-conversion of ~,
dJ,/{J1.)=~(9ffYj,k(dJ,/(~»

(9) where J1. is the j, k1h-reflection of ~, d J ,/{J1.) = fYllff:JP 2 j ,k(dJ ./~»

(10) where J1. is the j, t'-vacuous expansion of~,
d J ,/{J1.) = rd~j,t,(dJ,/(~»
(11) where J1. is the conditionalization of ~ and (,
dJ,/{J1.)=~(9ff~(dJ, / (~), d J ,f(0)

(12) where J1. is the negation of~, dJ,/{J1.)=fflff<;§(dJ,/(~»

(13) where J1. is the necessitation of~, dJ,/{J1.)=fflff~(dJ,/(~»
(14) where J1. is any propositional formula c/J, dJ,/(c/J) is defined as
follows:
(a) if c/J is a primitive term of type p, dJ,/(c/J) is already defined
(b) if c/J=pnrl ... rmdJ,/(c/J)=
f!J 20ltrg 1(f!J 20lt<;§ 2('" (8P20ltrg n(dJ,Apn), dJ,/rn», ... ),
d J ,Ar 2 », dJ,/r 1»
(c) if c/J=(~I/I), dJ ,/(c/J)=fflffrg(dJ,iI/l»
(d) if c/J = (1/1 -+ X),dJ,/(c/J) =~(9ff~(dJ, / (I/I),dJ,/(x»
(e) if c/J = (VlXt)l/I,dJ ,/(¢) =OUfffYl ([AlXtl/l])
(f) if c/J =(01/1), dJ,/(c/J)=fflffrg(dJ,/I/I»
120 CHAPTER V

(15) where that-cp is any complex propositional term,

dJ./(that-cp) =dJ./(cp)
(16) where (IXt)cp is any object description, dJj(IXt)cp) =
0.1 iff (3/')(/',,1 & I'(xt) =(/ & I'satisfies rPwith respect to

{ tOo & (I") (/'~ / & /" satisfies 1> with respect to tOo--+
1"= /'»
undefined, otherwise
(17) where (IT).sk is any sense description, dJ./«~) =
=
dJ./«lzt)(ptiP)(zF F = [Ax rP & (i)(cpi --+ Y = EX)]»lD.

C. SATISFACTION

Given an interpretation cI and an cI -assignment f, we may define Isatisfies

rP with respect to £0 as follows:
(1) If cp is any primitive term of type p, I satisfies cp with respect
to £0 iff £xt,)dJ./(rP» = T
(2) If rP = pnt 1 ... t m I sa tisfies rP with respect to £0 iff (301) ... (30 n)
(3t n)(01=dJ ./(t 1) & ... & 0n= dJjtn) & -in =dJ./(pn) &
< 01"'" on>E£xt w(tn».
(3) If cp = t p, I satisfies cp with respect to £0 iff (30 )(3t)
(o=dJ./(t) & -i=dJ./(p) & oE£xt,nt(t».
(4) If rP = (~ 1jJ), I satisfies IjJ with respect to £0 iff I fails to satisfy
IjJ with respect to w.
(5) If rP = (1jJ --+ X),! satisfies cp with respect to £0 iff I fails to satisfy IjJ
with respect to to or I satisfies X with respect to w.
(6) If rP = (If rl)ljJ, I satisfies rP with respect to tv iff (If 1')(1V --+ I'
satisfies !/J with respect to £0).
(7) Ifcp=(OIjJ),/ satisfies cp with respect to £0 iff (w'HI satisfies
1/1 with respect to £0').

D. TRUTH UNDER AN INTERPRETA TlON

rP is true under cI iff every cI -assignment I satisfies cp with respect to £00'

rP is false under cI iff no c1-assignment I satisfies cp with respect to woo
 THE TYPED THEORY OF ABSTRACT OBJECTS 121

3. THE LOGIC

A. LOGICAL AXIOMS

In order to state some of the logical axioms, we will need to utilize the
following three definitions, which are similar to their counterparts in the
elementary and modal versions of the theory:
D2 pip = alp =djD(xt)(xF == xG)
D3 F(tl, ... ,tn)/P=G(tl, ... ,tn)/P=din> 1)
(xt2) .. . (xtn)([Al' Fyx t2 ... xtn] = [Al' Gyxt2 ... xtn]) &
(xtl)( x t3) ... (xtn)( [Al2 Fxtl yxt3... xtn] = [Al2 Gx tl yxt3 ... xtn]) &
... & (xtl) '" (xtn-I)([Aln Fxtl ... x tn - 1 yJ= [,lin Gx tl ... xtn-Iy]).
D4 FP=GP=dj[Ayi F] = [Ayi GJ.
The logical axioms of our system are to be all of the modal closures of
the following schemata, with the exception of the object description
schemata, the unmodalized instances of which are to be axioms:

Propositional Schemata
LA 1: <p ---7 (tjJ ---7 (jJ)
LA2: (<p ---7 (tjJ ---7 X)) ---7 « <p ---7 tjJ) ---7 (<p ---7 X))
LA3 : ( ~ <p ---7 ~ tjJ) ---7 « ~ <p ---7 tjJ) ---7 <p ).

Quantificational Schemata
LA4: (a) (rx)<p ---7 <p~, where r contains no object descriptions and is
substitutable for rx 11
(b) where tjJ is any atomic formula, and r
(rx)<p---7(tjJp---7<p~),
both contains an object description and is substitutable for
both rx and p.
LA5: (rx)(<p---7tjJ)---7(<p---7(rx)tjJ), provided rx is not free in <p.

Modal Schemata
LA6: 0 <p ---7 cp
LA7: D(<p ---7 tjJ) ---7 (D cp ---7 0 tjJ)
LA8: O<P oO<P---7
122 CHAPTER V

LA9: O(rx)<p == (rx)O<p

LA1O: (xt) (pt/P)(OxF -> OxF).

A-Schemata

A-EQUIVALENCE: where 4> is any propositional formula with no object

descriptions, the following is an axiom:
(xtl) ... (Xtn)([Arx 1 .•• rxn<p ]x t [ .•• Xt" == <p~::::::~:tn).

A-IDENTITY: where P is any relation term and rx l , ... , rxm rx;, ... , rx~ are
distinct object variables not free in pP, the following is an axiom:
[Arx l ... rxnprx l ... rx n] = P & [Arx l ... rxnP P] = [Arx~ ... rx~pP].

Object Description Schemata

L-DESCRIPTIONS I : where IjJ is any atomic formula or conjunction of
atomic formulas, the following is an axiom:
ljJ~xt)'" ->(jl)(<p~ & 1jJ~).

L-DESCRIPTIONS z : where IjJ is any atomic formula, the following is

an axiom:

L-DESCRIPTIONS 3 : where IjJ is any atomic formula with rx l free and X

is any formula with rx z free, the following is an axiom:
1jJ~;t)'" -> ~ ((jl)(<p~ & X~2) & (jl)(<p~ & ~ X: 2 )).

In addition to the modal closures of the above schemata, and the

unmodalized description schemata, the modal closures of the following
five logical truths are also to be logical axioms:
LA11: (F(tl .....tn)/P)(xt [) . .. (xtn)(ExFxt[ ... xtn == Fxt [ ... xtn)
LA12: (P)(TrP == P).
LA11 and LA12 are logically true because of our restrictions on the $'
function of interpretations (Section 2, A). They tell us that the explicit
 THE TYPED THEORY OF ABSTRACT OBJECTS 123

exemplification and truth predicates work as they should.

LA13: that-¢= ¢
LA14: A !tIP~~ & R(t,t,i)/P KaKa , where ~~ is any sense name.
LA13 is logically true since in clause 15 of the definition of denotatio~.J'
that-¢ and ¢ are assigned the (semantically) same proposition. So
[A,yi that-¢] and [A,yi¢] will be encoded by the same individuals.
Recall that we may read the second conjunct ofLAl4 as: the sense ofthe
name "K" with respect to a represents K to a. LA14 is also logically true
because of the restrictions placed on $'.
LA15: where ¢ is any propositional formula with no object descrip-
tions, the following is an axiom:
(IXt)¢ = (lzt)(pt/P)(zF == F = [ht¢ & (/)(¢~ -? Y = E'X )])12.
LA15's validity is a consequence of clause 17 in the definition of
denotation.§, /. Our sense descriptions denote Platonic Forms of type t (i.e.,
abstract objects oftype t which encode a single tip-property) which encode
an individuating property (i.e., one which at most one object of type t can
exemplify).

B. RULES OF INFERENCE

We use the same two rules of inference that were used in the earlier
versions of the theory, --+ E and UI. DI is still a derived rule, subject to
the restriction discussed in Chapter IV, Section 3. We shall of course avail
ourselves of the usual derived rules of inference and proof techniques.
RELATIONS is derivable in the same way it was derived in the earlier
chapters and it is still subject to the restriction that the formulas ¢ used
must not contain any object descriptions. Consequently, a type theory of
relations falls right out of A-EQUIVALENCE and D 3 • It is not a theory
in which logically equivalent relations are identical. PROPOSITIONS is
also derivable and subject to the same restriction as RELATIONS.
PROPOSITIONS and D4 give us a theory of propositions.
We call the metaphysical system which consists of the interpreted typed
language (without the unusual complex terms or distinguished predicates),
together with LA1-LA10, the Aand object description schemata, and the
124 CHAPTER V

rules of inference, the typed object calculus. The addition of the unusual
complex terms and distinguished predicates, together with their semantics
and logic (especially LA 11- LA15), constitutes a special modification of
the typed object calculus which has been designed specifically to deal with
the data about propositional attitudes and mathematics.

4. THE PROPER AXIOMS

We assert that the modal closures of AXIOMS 1, 2, 3, 4, and 6 are all

true A PRIORI, as well as the unmoda1ized instances of AXIOM 4. We insert
a definition after the second axiom, in terms of which the third axiom is
stated:

AXIOM 1. ("Et-IDENTITY"): xt = E'i == OE F/px & OE F/Py &

o (pt/P)(Fx == Fy).

Since we have general identity defined for all objects other than individuals,
we need to say when two individuals are the same:
D5 Xi=yi=dfxi=Eii v (A !i/PX &A!i/ py & o (Fi/P)(xF == yF)).
The following axiom is therefore meant to govern any objects satisfying
D 2 , D 3 , D4 or D 5 •

AXIOM 3. ("IDENTITY"): 0: = /3 ~(4)(0:,0:) == 4>(0:,/3)), where 4>(0:,/3) is the

result of replacing some, but not necessarily all, free occurrences of 0: by
/3 in 4>(0:,0:), provided /3 is substitutable for 0: in the occurrences of 0: it
replaces.

AXIOM 4. ("A-OBJECTS t ,,): where 4> is any formula in which xt is not

free, the following is an axiom:
(3xt)(A !t/px & (pt/P)(xF == 4»).

AXIOM 5. ("DESCRIPTIONS"): where tf; is any atomic formula with a

free variable 0: of type t, the following is an axiom:
tf;~xt)'" == (3 !i)4>~ & (3i)(4>~ & tf;~).
 THE TYPED THEORY OF ABSTRACT OBJECTS 125

AXIOM 6. ("NECESSARY EXISTENCE"): for any type t, t =1= i, the

following is an axiom:
OE !t/pxt -+ OE !t/pxt.
Et-IDENTITY, NO-CODERt, IDENTITY, A-OBJECTSr, and DES-
CRIPTIONSt should be straightforward, given our familiarity with their
counterparts in Chapters I-IV. Note that abstract objects of type t might
encode abstract tip-properties, as well as (possibly) existing ones. 13 The
pip-quantifier in A-OBJECTS t ranges over all tip-properties.
We have added one extra axiom to the typed theory to preserve the
intuition that higher order objects are not contingent beings. Since higher
order objects either possibly exist or fail to possibly exist, it follows from
NECESSARY EXISTENCE that either they necessarily exist or they
necessarily fail to exist. Philosophers who do not share the intuition that
higher order objects are not contingent beings may not wish to embrace
this axiom.
I think that Et-IDENTITY, NO-CODERt, IDENTITY, A-OBJECTS t,
DESCRIPTIONSr, and NECESSARY EXISTENCE shall prove to be
consistent. 14 We have taken steps to prevent the offending instances of
property abstraction from being denoted. However, we have assumed that
it is safe to have abstract objects of type t encode abstract properties of
type tip. I do not think that this move will introduce paradoxes, but it
might. Should it do so, there are obvious ways to weaken the theory and
preserve some of the applications which follow (we would, however, lose
the very important model of mathematical entities). There is a great deal
of investigation which must be carried out before we can feel confident
that this particular version of the theory is consistent.
As usual, we add abbreviations for the appropriate English gerunds to
our primitive vocabulary. And we add abbreviations for English
proper names - names which are not necessarily associated with works
of fiction.ls Finally, we use the distinguished constants Bl (i.p)/p, B2 (i,p)/p, ...
to abbreviate the verbs of propositional attitudes such as believes, hopes,
knows, expects, etc. l 6
 CHAPTER VI

APPLICATIONS OF THE TYPED THEORY

1. MODELLING FREGE'S SENSES (I)

Frege's explanation, by way of ambiguity, of what appears to be the logically deviant

behavior of terms in intermediate contexts is so theoretically satisfying that if we have not
yet discovered or satisfactorily grasped the peculiar intermediate objects in question, then
we should simply continue looking.
DAVID KAPLANl

In this section, we translate and discuss the propositional attitude data

which involve English names and definite descriptions that denote
individuals. The data sentences are labelled (A)-(X), and because we are
supposing with Frege that certain English terms occurring in them are
ambiguous, there are several readings possible for each one. These readings
are provided immediately after the particular datum is presented, and a
discussion usually follows. In these discussions (in this section only), we
revert to using the word "object" to refer to individuals and the word
"property" to refer to properties of individuals (i.e., i/p-properties).
Also in these discussions, we shall modify somewhat the standard
Fregean metalinguistic and metaphysical terminology. On the strict
Fregean view, a term expresses its sense and denotes its denotation. And
it is also said that the sense of a term belongs to the denotation of the
term. Pictorially, these relationships are sometimes represented as follows:

expres~
term T
;es ~sense

belongs to
j
denote~
:s ~denotation

Now we shall talk as if terms do denote their denotations (this is made

precise by our definition of denotation",I' V, Section 2), but we shall not
suppose that terms "express" their senses. Instead, we shall talk about the
A-object which is associated with the term with respect to an individual.
Sometimes, we shall say that the A-object serves as the sense of the term
with respect to a given individual. We assume with Frege that the sense

126
 APPLICATIONS OF THE TYPED THEORY 127

of a term (especially proper names) varies from person to person. The

special sense terms (and their interpretation) that we added to our language
in Chapter V help us to represent this phenomenon and help to make the
above terminology precise.
For reasons which will soon become apparent, we shall not talk in
terms of the metaphysical "belonging to" relationship between senses and
denotations. Instead, we shall talk about the weak correlates which the
A-objects that serve as senses may have. Should the A-object have
one or more weak correlates, we do not suppose that any of these
objects necessarily serves as the denotation of the term in question.
Diagrammatically, we get:

A-object (sense)
is associated with may be a weak
(with respect to a n / ~ -- correlate of
individual) ~
terms'! object
denote~ /may be identical

object

A. S believes that Lauben is late

(.1) Bsthat-Ll (DE RE)
(.2) Bsthat-LL (DE DICTO).
Suppose John feels ill one morning and resolves to stop in at the first
physician's office he happens to pass on his way to work. He rounds a
corner and sees a sign on a door: 2
DR. GUSTAV LAUBEN
General Practioner
8 :00 A.M. - 4:00 P.M.
At this point, he has now become part of a causal chain of events involving
the name "Gustav Lauben".3 He associates an A-object with this name-
an A-object which serves as the sense of that name for him. We call
that A-object "LaubenJohn". Lauben JOhn lends the name "Lauben" its
cognitive significance or information value for John. It does so by encoding
properties which serve to RE-present to him the object that he supposes
128 CHAPTER VI

is denoted by the name. Lauben John may encode such properties as: being
a (the) doctor whose office is at 15 High St., being a general practitioner,
being the doctor whose signpost this is, being a doctor who works from
eight to four, etc.
Some other person, S, who first encounters the name "Lauben" under
different circumstances will not associate with this name the A-object
John's utilized. The A-object S utilizes will encode properties presented
to S as being characteristic of the object named "Lauben". Thus,
Laubens lends the name "Lauben" a cognitive significance for S which
is distinct from the cognitive significance this name has for John.
On the theory we have developed in Chapter V, it is axiomatic that
Laubens represents Lauben for S. This is one way of capturing Frege's
principle that the sense of a term account for its information value. Frege
also required, however, that the sense of a term determine at most one
object, and that this object, should there be one, serve as the denotation
of the term. It requires additional semantic complexity to capture these
Fregean principles, and were we interested in a more strict modelling of
Frege's ideas, we could modify our semantics.4 However, we've chosen
not to place these constraints on senses because: (a) the successful
explanation of the data on which we have chosen to work does not seem
to require that we have such constraints on senses, and (b) there are cases
which suggest that Frege's principles are too strong.
Suppose that the week before John went to Lauben's office, the Medical
Review Board stripped Lauben of his license to practice, his medical school
invalidated his degree, and he subsequently sold his office, never to return.
It is just that no one bothered to take the sign down. In this situation,
all of the properties which we have suggested might be encoded by
LaubenJOhn are not exemplified by Lauben. Lauben is not the weak
correlate of LaubenJohn ; in fact, no object is.5 Nevertheless, it seems
reasonable to suppose that if John hasn't learned about Lauben's recent
calamity, LaubenJohn , as described, still serves as the sense of"Lauben" and
lends it cognitive significance.
To see this more clearly, suppose John knocks at the office door and
no one answers. He notices that it is just after 8 :00 A.M. He believes that
Lauben is late (our datum sentence). His belief is not DE RE, since he
believes this without believing that the friend of Leo Peters is late (suppose
Lauben is Peter's unique friend). So his belief is DE DICTO.
Now even though Lauben is not the weak correlate of Lauben John ,
the latter A-object could still be instrumental in helping John to construct
a proposition which serves as the "object" of his DE DICTO belief We may
 APPLICATIONS OF THE TYPED THEORY 129

suppose that the propositional object of his DE DICTO belief is [1J !l!O/irg 1
(6einfl t <hie, !I! <hU6en,juli,J Had his belief been DE RE, the propositional
object of his belief would have been [1J !l!O/irg 1 (6einfl t <hie, !I! <hu6en). His DE
DICTO belief will be true just in case this latter proposition is true. 6 Since
Lauben no longer comes to his office, his DE DICTO belief is a false one.
(In the above paragraph, we have switched typestyles and described the
propositional objects of John's belief semantically. This allows us to
graphically display the structure of these objects. In what follows, we
frequently switch to the script typestyle when discussing either the
propositional objects of beliefs or their constituents. With this warning,
no confusion should arise.)
Consequently we seem to be able to describe the important facts ofthis
situation without having to suppose that Lauben John has a (unique) weak
correlate, and without having to suppose that one of its weak correlates
has to serve as the denotation of "Lauben". It is for this reason that we
have chosen not to further complicate our semantics in order to present
a more strict modelling of Frege's ideas. 7
We have, however, validated another one of Frege's principles in the
process - the A-object which serves as the sense of the term also serves
as the denotation of the term inside DE DICTO belief contexts. To make
this clearer, let us look at our two readings A.l and A2 in more detail.
Our DE RE reading is Bsthat-Ll (A.l). From A.l, we can prove (1)-(4):
(1) (3x)Bsthat-Lx
(2) [Ax Bxthat-Ll]s
(3) [Ax Bsthat-Lx] I
(4) [hyBxthat-Ly]sl.
On the assumption that Lauben exists, A.1 also implies (5):
(5) (3y)(E !y & Bsthat-Ly).
We have symbolized the DE DICTO reading of (A) as BSthat-Ll.
(A2). Thus, it is the sense of the name "Lauben" with respect to S which
serves as the denotation of the name inside the DE DICTO belief context.
From A2, we may prove (6)-(10):
(6) (3x)Bsthat-Lx
(7) (3x)(A !x & Bsthat-Lx) (LA 14)
(8) [Ax Bxthat-Ll.]s
130 CHAPTER VI

(9) [Ax BSthat-Lx]L,

(10) [AXY Bxthat-Ly]s1.
Quantification into the belief context works NORMALLY, as does A-
conversion.
An examination of another case should help - here is one inspired by
Quine's work. s
B. Ralph believes that Cicero was a Roman
(.1) Brthat-RC (DE RE)
(.2) Brthat-R£,. (DE DICTO).
C. Ralph does not believe that Tully was a Roman
(.1) '" Brthat-Rt (DE RE)
(.2) '" Brthat-Rt.,.(DE DICTO).
O. Cicero is Tully
(.1) C= t
(.2) C = Et.
The triad B-C-O (and other triads like it) constitutes a paradigm case
where the English proper name exhibits logically deviant behavior. From
BJ and 0.1 (or 0.2), it follows that Brthat-Rt. Identity elimination works
normally. But from B.2 and OJ, nothing follows. And there is no reason
to think that B.2, C.2, and (0) are jointly inconsistent. From B.2 and CJ,
we can conclude both that £r f- t and that R£r f- Rt. From B.2 and C.2, we
can conclude both that £r f-lr and that R£r f- Rlr. Thus, we follow Frege in
thinking that it is the ambiguity of the English proper name inside DE DICTO
contexts which accounts for its logically deviant behavior.
We now precisely define the conditions under which someone has a
true belief. Let us define the erasure of a formula </> ("</>*") as the formula
which results by deleting all the underlines and subscripts from terms
occurring in </>. So where </> = Ri,., </>* = Rt. We now define:
D6 S truly believes that </> ("TBsthat-</>")=dfBsthat-</> & </>*.

So from B.l or B.2 and the supposition that Cicero was a Roman,
it follows that Ralph has a true belief. From A.1 or A.2 and the supposition
that Lauben is not late, it follows that S does not have a true belief.
Note that in the case of DE DICTO readings, S can truly believe that </>
even when </> is false. In B.2, the propositional object of Ralph's belief is
 APPLICATIONS OF THE TYPED THEORY 131

a false proposition - by our AUXILIARY HYPOTHESIS, no A-object

exemplifies the (nuclear) property of being a Roman. This false proposition,
however, is just a neutral object which helps Ralph to represent f1jJ .!I!Oltl'fJ 1
(t!-e£'n? -a !?kom-an, ~£·ceM). Consequently, we must abandon a certain
principle some philosophers hold about true belief. The principle that S
truly believes that cP iff S believes that cP and cP is true must be given up,
not just because it is inconsistent with our treatment of belief, but also
because doing so allows us to construe the logic of propositional attitude
contexts as another application of the philosophical logic of encoding
properties. The usefulness of abandoning the old principle is a good reason
for doing so.
E. S believes that Lauben was mugged
(.1) Bsthat-Ml (RE)
(.2) Bsthat-ML (DICTO).
F. S' believes that Lauben was mugged
(.1) Bs'that-Ml(RE)
(.2) Bs'that-Ml., (DICTO).
Suppose John goes to a party in the evening of the day he knocked on
Lauben's door. Suppose also that Lauben is in good standing in the
medical community, but that he just did not go to work that day.
Leo Peters (Lauben is his unique friend, and John's unaware of their
relationship) is there and John overhears him say "Dr. Lauben was mugged
last night". The proposition John grasps when he hears this utterance is
f!J> fL'0Zt1'fJ 1 (t!-ein? mu??ed, fL' -aut!-en ,fo&n)' His belief is DE DICTO, because
he believes that Lauben was mugged without believing that the friend of
Leo Peters was mugged. "fL' -aut!-e;z ,fo&n" mayor may not be the (semantic)
name of the A-object he associated with "Lauben" that morning. As John
went through the day, he might have been involved in another context
in which the name was used. The new information he gathers might get
"encoded" by associating some distinct A-object which encodes all the
old and new properties he now uses to re-present Lauben to him via the
name "Lauben". For now, however, let us suppose that this name retains
its earlier cognitive significance.
Now is there any reason to believe that the proposition John grasped
when Peters uttered his sentence was the same proposition that Peters
was entertaining? Suppose Peters' belief were DE DICTO. It seems like there
would be "more intimate" properties encoded by Lauben Peters than are
132 CHAPTER VI

encoded by Lauben John ' We have supposed that Lauben is Peters' friend,
and there might be a very complex A-object which Peters associates with
"Lauben". Although the fact that Laubenpeters =1= Lauben John is not a
guarantee, it seems likely that the propositional object of Peters DE
DICTO belief may differ from the object of John's DE DICTO belief.
Despite the fact these propositions may differ, there may still be good
reason for thinking communication has taken place. A full discussion of
how the communicative process operated in this situation would take us
too far afield. We would have to discuss the intentions of the speaker to
refer to Lauben, determine whether the speaker succeeded in referring to
Lauben, and these might involve a discussion of the presuppositions of
the context of the utterance. Even if we had a reasonable understanding
of these features of the communicative process, it now seems in order
to consider two further features. And they are, the degree to which
Laubenpeters and Lauben John are "similar" A-objects and the kind of
correspondence there is between the properties these A-objects encode
and the properties Lauben exemplifies.
In the ideal case, Laubens and Laubens ' will be identical (or one
will be embedded in the other) and Lauben will be the unique weak
correlate of both of them. At the other extreme, Lauben sand Lauben S'
will have no properties in common and Lauben will be the unique weak
correlate of neither of them. Communication takes place to a greater or
lesser degree depending on whether the former or the latter of these two
extremes is more closely approximated. So despite the fact that "Lauben
was mugged" might be used by S to express one proposition and used by
S' to construct another proposition, communication between Sand S'
takes place to a greater degree if both Lauben sand Lauben S' encode,
for the most part, properties which Lauben exemplifies. In the cases
where Laubens and Laubens' have little in common (with Lauben),
communication is rather crude and not straightforward. Yet even in these
latter cases, it is important to note that the language is holding everything
together (as we might expect for DE DICTO beliefs). Laubens and Laubens '
would have in common the fact that they are both associated with the
name "Lauben". If Lauben was mugged, Sand S' have true beliefs.
G. S does not believe that the friend of Leo Peters was mugged
(.1) ~Bsthat-M(IX)Fxp (RE)

(.2) ~ Bsthat-M(lx)Fxp (DICTO).

 APPLICATIONS OF THE TYPED THEORY 133

Recall that we established that John's belief that Lauben was

mugged was DE DICTO by the fact that he did not also believe that the
friend of Leo Peters was mugged. But the English definite description
exhibits logically deviant behavior inside belief contexts as well. On the
DE RE reading of G, the proposition that he fails to believe is flJ !t'U!Lrg 1
(tJein{f -mu{f{fed, tlie Ilf~·end u/ !t'eu gpet eMJ), i.e., flJ!t'U!Lrgl (tJein{f
-mu{f{fed, !t' o,u6en). On the DE DICTO reading of (G), the proposition he
fails to believe is flJ!t'U!Lrgl(6ein{f -mu{f{fed, tlt e/iiend u/!t'eu flJetei:J).
The friend of Leo Peters is the abstract object which encodes just the
property of being tht: friend of Leo Peters (by LA15). That is,
(lX)FXP = (lZ) (G)(zG == G = [Ax Fxp & (y)(Fyp --+ y = EX)]).
The friend of Leo Peters serves as the sense of "the friend of Leo
Peters". It lends the English description cognitive significance and
information value. It also has at most one weak correlate, and in this
case, its unique weak correlate happens to be the denotation of the
description. Finally, the friend of Leo Peters serves as the denotation of the
description when the description is inside a DE DICTO context.
H. Ralph believes that the man in the brown hat is a spy

(.1) Brthat-S(lx)¢l (RE)

(.2) Brthat-S~l (DICTO).

l. Ralph does not believe that the mayor of the town is a spy
(.1) ,. . ., Brthat-S(lX)¢2 (RE)
(.2) ,. . ., Brthat-S~2 (DICTO).
J. Ralph believes that the mayor of the town is not a spy
(.1) Brthat-,....., S(lX)¢2 (RE)
(.2) Brthat-,....., S~2 (DICTO).
K. Ortcutt is both the man in the brown hat and the mayor of the town
(.1) 0 = (lX)¢l & 0 = (lX)¢2
(.2) 0 = E(lX)¢ 1 & 0 = E(lX)¢2·

If Ralph's belief in (H) is DE RE, the object of his belief is gp 2!t1/1rg 1(tJein{f
a, :Jjzy, tlie -mo,n in tlie tJltuwn lio,t), i.e., fJjJ 2!t1/1rg 1 (tJein{f a, :Jjzy, (!htcutt).
Given H.1 and (K), we may conclude (11):
134 CHAPTER VI

(11) Brthat-So.
Given 1.1 and K, we may conclude (12):
(12) ~ Brthat-SO.
So H.1 and 1.1 are inconsistent. Ortcutt himself is the constituent of the
propositional object of the DE RE belief - the descriptions inside the relevant
belief ascriptions "contribute" their denotation to the proposition. From
H.l and DESCRIPTIONS, we also get (13):
(13) (:3 !Y)(cP11 & Brthat-Sy).
However, let us suppose Ralph's belief is DEDICTO. The object of his belief
is f!J fi'llItrg 1 (6einfl a djzy, tAl! man t'n tAl! IiZ()tNl Aal). From H.2 and (K),
nothing follows. From H.2 and 1.2, it follows that the man in the brown hat
=1= the mayor of the town. If the mayor ofthe town is a spy, then it follows
from H.2 that Ralph has a true belief. (14) also follows from H.2, given LA 15
and DESCRIPTIONS:
(14) (:3 !x)(A !x & (P)(xP == P = [AX cP1 & (Y)(cP1i -;. y = EX)])
& Brthat-Sx).
Besides these, we have the usual consequences of H.2 based on existential
introduction and A-conversion:
(15) (:3x)Brthat-Sx
(16) [AX Bxthat-S(Q;.)dLl]l'
(17) [AX Brthat-S X ] (Q;.)dL 1
(18) [AXY Bxthat-Sy]r(Q;.)dL1
Note that H.l and J.1 ascribe contradictory beliefs to Ralph. Given (K),
H.l implies (11) and J.l implies (19):
(19) Brthat-~So.

From (11) and (19) we get (20):

(20) (:3pO)(Brthat-PO & Brthat- ~ pO).
However, (20) does not imply that Ralph believes a contradiction.
If H.2 and J.2 correctly describe Ralph's state of mind, then we cannot
prove that Ralph has inconsistent beliefs. From H.2 and J.2, we cannot
deduce that f.u{)s/21 =1= (Q;.)dL2' but we can prove this from the plausible
 APPLICATIONS OF THE TYPED THEORY 135

assumption that the property of being the man in the brown hat is distinct
from the property of being the mayor of the town. Since ~1 f i1M.2'
no substitutions into H.2 and J.2 would lead us to think that Ralph has
inconsistent beliefs. However, from 1.1 or J.2 and the fact that Ortcutt is
a spy, we can prove that Ralph has a false belief, where,
S falsely believes that ¢ ("F BSthat-¢") = df
BSthat-¢ & '" TBsthat-¢.
L. Ralph believes that the shortest spy is a spy
(.1) Brthat-S(lx)¢3 (RE)
(.2) Brthat-S~3 (mcTo)
If L.1 expresses what Ralph believes, then his belief would be of interest
to the FBI. (21) follows from L.1:
(21) [AX Brthat-Sx] (IX)<P3.
If Bond is the shortest spy, then (22) follows from L.1, and (23) follows
from (21) or (22):
(22) Brthat-Sb
(23) [Ax Brthat-Sx]b.
If we assume that Bond exists and that an existence claim is built into
<P3' then we can generalize on (21)-(23) to get:
(24) (3x)(E !x & Brthat-SX)
(25) (3y)(E!y & [AxBrthat-SX]Y).

All of this results because the propositional object of Ralph's belief has
an existing object, namely Bond, as a constituent.
None of these results follow if L.2 expresses what Ralph believes. There
is no way to use "exportation" on L.2 to produce (24) or (25). We can
only reap the "standard" inferences from L.2 based on existential and
lambda introduction:
(26) (3x)Brthat-Sx
(27) [AxBrthat-Sx]~3'

(28) (3y)(A!y & [AX Brthat-Sx]y).

136 CHAPTER VI

I take it that the FBI would not be interested by the fact that Ralph, like
most everyone, uses an abstract object to represent whoever it is that is
the shortest spy (in the absence of a DE RE belief).
M. Ralph believes someone is a spy
(.1) (3x )Brthat-SX
(.2) Brthat-(:lx)Sx.
M.l and M.2 disambiguate (M). M.1 is similar to (24) and we might prefer
to use the latter to read (M) properly. M.2 relates Ralph to a proposition
which fails to have object constituents. No legitimate exportation on M.2
will get us to M.l.
N. S believes that Newton met Leibniz
(.1) BSthat-Mnl (RE)

(.2) BSthat-MlJ) (DICTajRE)

(.3) Bsthat-Mnl, (RE/DICTa)

(.4) Bsthat-M!1.sL (DIcTa).

In order to determine which of the readings of (N) is the correct one, we

have to examine data triads to discover how the names "Newton" and
"Leibniz" are functioning.
O. Frege believes that Hesperus is Hesperus.
P. Frege does not believe that Phosphorus is Hesperus.
Q. Phosphorus is Hesperus.
There are various ways to represent the triad O-P-Q. The preferred
representation is as follows:
0.' Bjihat-h.J = Eh.
P.' '" Bjihat-l2J = Eh.
Q.' P=E h.
Suppose Frege as a young man is being taught the names of the stars.
Early one evening, his teacher points out Venus and says 'That is
Hesperus - it is the first visible star of the evening". Frege becomes, at
that moment, part of an historical, causal chain of events connecting him
with the name "Hesperus". So we suppose that Frege associates an abstract
 APPLICA TIONS OF THE TYPED THEORY 137

object sense with the name. Hesperus Frege may encode: being the star to
which my teacher is pointing, being the first visible star of evening, being
clearly visible, being situated in position p in the western sky at
5:30 P.M. Thursday, December 7, 1860, etc. "Hesperus" would have
a different cognitive value for someone who learned the name in different
circums tances.
Now suppose Frege's teacher points out Venus to Frege early the next
morning and says "That's Phosphorus - it is the last star visible in the
morning". The young Frege will associate some new, distinct A-object
with "Phosphorus". That's because the features of the learning situation
are radically different. The object pointed out to him is in a position of
the sky that appears unrelated to the position of the object pointed out
the evening before. The names introduced are distinct. There is no reason
for Frege to believe that the object pointed out to him then is identical
with the object pointed out to him the evening before.
So if Frege's teacher does not tell him that Phosphorus is Hesperus,
Frege could believe that Hesperus is Hesperus without believing that
Phosphorus is Hesperus. Although there are various ways to represent
this data as a consistent triad, we have chosen the reading on which Frege
believes f!J!!l'0/1'§ 1 (f!J!!l'0Zt'§ z (~dentttYE' Yf <XJ/te'tu{)), Yf <XJ/te'tu{)g;"flJ and
fails to believe f!J! 20/1'§ 1 (f!J!2iJ[t'§ zV deniitYH Yf <XJ/te'tu{)),
fjJ/iO{j /t/io'tU{) g;«,J.

R. John hopes that the strongest man in the world, whoever he is, beats
up the man who just insulted him.
Preferred reading:
(R') Hjthat-B(lx).sk4 (IX)cP5 (mcTO/RE).

On the preferred reading of (R), we interpret the first definite description

as occupying a DE DICTO position and suppose that it contributes its sense
to the proposition which is the object of John's hope.
S. Mary believes that the wife of Tully is the wife of Tully.
T. Mary doesn't believe that the wife of Cicero is the wife of Tully.
U. The wife of Cicero is the wife of Tully.
Preferred representation: 9
S.' Bmthat-(IX)Wxt m = E(IX)WXt (DICTO).
138 CHAPTER VI

T.' ~Bmthal-('X)WX-km=E(IX)Wxt (OICTO).

u.' (IX)WXC = (IX)WXt.

S-T-U is an interesting triad since it requires that we use the senses
of the names "Tully" and "Cicero" with respect to Mary to construct the
senses of the English descriptions "the wife of Tully" and "the wife of
Cicero". That's because the wife of Tully and the wife of Cicero are
identical, and so (S") and (T") are inconsistent.
S." Bmthat-(lx)Wxt=£(IX)Wxt (OICTO).

T." ~ Bmthat-(lx)WXC = £(IX)WXt.

The wife of Tully and the wife of Cicero are identical because Cicero is
Tully, and so being the wife of Tully just is being the wife of Cicero. Since
these properties are identical, the object which encodes just being the wife
of Tully is identical with the object which encodes just being the wife of
Cicero. So we can not use (S") and (T") to help us understand how S-T -U
is consistent because the proposition that the wife of Tully is identical£
with the wife of Tully is identical with the proposition that the wife
of Cicero is identical£ with the wife of Tully.
So we MUST use the senses of "Tully" and "Cicero" with respect to Mary
in order to suppose S-T-U is consistent. Thus, the wife ofThlli'm is a
constituent of the propositional object of Mary's belief in (S'). Though
the wife of TullYm could have at most one weak correlate, it fails to have
any. By the AUXILIARY HYPOTHESIS, A-objects fail to exemplify the
(nuclear) property of having a wife.!O
Our definition of true belief still works fine:
(29) T Bmthat-(zx)Wxlm = £(IX)WXt ==
Bmthat-(lx)WXlm=£(zX)Wxt & (zx)WXt=£(IX)WXt.

Given (S') and given that there is a unique wife of Tully, it follows that
Mary has a true belief.!! If the negation of (T') represented Mary's state
of mind, she would still have a true belief. If Mary believes that the wife
of Cicero was not the wife of Tully, and this was correctly represented as
Bmthat-(lx)WXC m +£(IX)WXt, then she would have a false belief.
We insert here a few general remarks about our treatment of definite
descriptions. G-L and R-U give us evidence for thinking that English
descriptions have both a sense and a denotation. The sense of the definite
description lends it cognitive value - a value to beings with represent-
 APPLICATIONS OF THE TYPED THEORY 139

ational capacities in that it enables them to recognize (or understand what

it might be like to recognize) objects which have (never) been presented
to them. If we are interested solely in describing the cognitive value of a
given English description, we always have available to us a SENSE-
description of our formal language. English descriptions do not come with
their property terms marked as to whether the property denoted is
exemplified or encoded by the object being described. And it might be
that some other description of our language is better suited in having the
intuitively right denotation of the English description. For example, we
might prefer to use a description which contains an encoding subfbrmula
to translate "the student who killed an old moneylender," where this
English description is meant to refer to Raskolnikov of Crime and
Punishment (Chapter IV, Section 4). But insofar as we are interested purely
in the phenomenon of the apparent deviant behaviour of this English
description inside DE DICTO contexts, it might just be that the student who
killed an old moneylender serves well enough as its cognitive value. This
depends on whether there is conclusive data which shows that the question
of getting the denotation right and the question of explaining apparent
deviant behavior are not independent. 12
V. Bill believes that there is a barber who shaves all and only those
who do not shave themselves.
W. Bill does not believe that the sun is shining and the sun is not
shining.
x. Necessarily, there is a barber who shaves all and only those who
do not shave themselves iff the sun is shining and the sun is not
shining.
(V') Bbthat-(3x)(Bx & (y)(Sxy == '" Syy))
(W') '" Bbthat-(Ss & '" Ss)
(X') O((3x)(Bx & (y)(Sxy == '" Syy)) == (Ss & '" Ss)).
Were we to identity propositions with their extensions (i.e. with functions
from possible worlds to truth values), all contradictions and necessary
falsehoods would be identified. That's because contradictions and
necessary falsehoods are false at all possible worlds, and there is a unique
function which maps all the worlds to the truth value: False. This model
of propositions has the unfortunate consequence that if we believe a
proposition P we believe all propositions Q which are necessary equivalent
140 CHAPTER VI

to P. In particular, if we believe some necessary falsehood then we believe

all contradictions, since they are the same "proposition". But surely,
V-W-X count as data and give us good evidence for thinking that
necessarily equivalent propositions may be distinct. These considerations
provide overriding reasons for choosing the style of semantics we have
employed. In fact, it follows from (V') and (W') that the propositions in
question are distinct, since (V') and (W') imply (30):
(30) (h)(Bx & (y)(Sxy == ~ Syy)) '" (Ss & ~ Ss).

2. MODELLING FREGE'S SENSES (II)

We now consider the propositional attitude data triads which involve

English terms that denote higher order objects.
A. John believes that Woodie is a woodchuck

(.1) Bjthat- Ww (DE RE)

(.2) Bjthat- WjW (DE DICTO)13.
B. John does not believe that Woodie is a groundhog
(.1) ~ Bjthat-Gw (DE RE)
(.2) ~ Bjthat-QjW (DE DICTO).

C. Being a woodchuck just is being a groundhog

(.1) G= W
A.1 and B.l are inconsistent, given c.l. Al asserts that John
believes the proposition f1J>!t?uurg 1 (tJe£'nfj a wood cAuc£, if/"ood£'e), where-
as B.I asserts that John does not believe f1J> !t?uurg 1 (tJeinfj a fj'tound-
AOfj, 111 oodie). But since (C) is a true identity statement which asserts
that the properties of being a woodchuck and being a groundhog are
identical, it's provable that these propositions are identical. So either A1
or B.1 must be false.
A2 and B.2 can both be true together, however. A.2 asserts that
John believes f1J>!t?uurg 1 (tJe£"nfj a woodcAuci JoAn' 1f!oodie). Being a
woodchuck John is the abstract property of individuals (i/p-property) which
serves as the sense of the name "being a woodchuck" with respect to John.
The deni!tAn function of our semantics (Chapter V, Section 2, A) assigns
"being a woodchuck" a member of st i/ P (that is, a member ofthe abstract
 APPLICA TIONS OF THE TYPED THEORY 141

objects of type i/p). So "being a woodchuck John " denotes oenJoA)"being

a woodchuck").
Being a woodchuck John can be plugged up with any individual - our
f!lJ !l?if!ti§ function is defined so that it operates on all i/p-properties.
Consequently, f!lJ!l?if!t'§ 1 ((jeiJ?? a woodcltud ,Iud,,' 'if! oodie) is a type p
object and can serve as the propositional object of someone's belief.
B.2 asserts that John does not believe f!lJ !l?if!t'§ 1((jeinfJ a fJMund-
~"olin' "If! oodie). Being a groundhogJohn is the abstract i/p-property
which serves as the sense of "being a groundhog" with respect to
John. If A.2 and B.2 are true, it follows both that f!lJ!l?if!t'§l((jeinfJ a
woodCnfU;& ,loAn' if! oodie) =1= :?J>!l?if!ti§ 1 (6einr; a fJiOundnofl /olin'
if! oo-d ie) and that 6ein? a wood cnud / olin =1= (jet'enfJ a fJ'to-uend no-fJ / olin' So
as Frege predicted, the senses of the property terms flanking the identity
sign in C are distinct.
This seems right - "woodchuck" and "groundhog" probably entered
John's vocabulary under different circumstances. Maybe on one occasion
he saw and was told he was seeing a woodchuck. He then utilized
an A-object which encoded properties he took to be characteristic
of the property of being a woodchuck. And maybe on another occasion,
someone described woodchucks to him improperly, in the process saying
only that he was describing an animal called a "groundhog". John
would not have known that in fact these properties are the same.
Sentence (C) would be informative to him.
D. John believes that the chair in front of the class is Crayola crayon
blue
(.1) Bjthat-CCB(IX')¢l (RE)
(.2) Bjthat-CCBilX')¢l (DIeTO)
E. John does not believe that the chair in front of the class is French
fire engine blue
(.1) ",Bjthat-FEB(IX')¢(RE)
(.2) ",Bjthat-FEB/1X')¢1 (DIeTO)
F. Crayola crayon blue just is French fire engine blue
(.1) FEB=CCB
D-E-F is analyzed analogously with A-B-C. We may suppose 'French
fire engine blue" and "Crayola crayon blue" to be names of the same
142 CHAPTER VI

shade of blue. So (F) is an informative identity statement about properties.

As a boy, John may have become directly acquainted with this property.
But the label on his Crayola crayons just read "blue", and he has never
seen that shade of blue labeled "French fire engine".
G. John believes that Bill has the property of being a student
(.1) Bjthat-HasSb (RE)
(.2) Bjthat-HasjSb (mcTo).
H. John does not believe that Bill exemplifies the property of being a
student
(.1) ~ Bjthat-ExSb (RE)
(.2) ~ Bjthat-ExjSb (mcTO).
I. Having a property just is exemplifying a property
(.1) Ex = Has
G-H-I might describe a student beginning in philosophy, unaware of
the technical sense philosophers have for the word "exemplifies". "Has"
and "exemplifies" both denote relations of type (i/p, i)/p. We suppose that
there are abstract objects of this type which serve as the senses of these
names with respect to John. G.2 asserts that John believes r!J> fElllt'§ 1
(r!J>fElllt'§2 (Aav{'n[lfulm' !?ddt), te£'n? a dtadent); whereas H.2 asserts
that John doesn't believe r!J> fElllt'§ 1 (f!J fElllt'§ 2 (exem/zt i/yin? ful",' !?ddt),
tet'n? a dtadent),
To handle our next triad, J-K-L, we need to add some functional
notation to our language:
where p is a term of type (t1' t 2)/p and r is a term of type
t 1 , then per) is a term of type t 2 ,

Let us interpret this notation as follows

dJf,/(p(r» = dJf.l(lc,:!Z)pra).
So per) is the object which is such that r bears p to it. Using this interpreted
notation, we might construe adverbs as names of relations of type
(i/p,i/p)/p, i.e., relations which relate two i/p-properties. For example,
"slowly" might denote a relation between the property of walking and the
property of walking slowly. In the language, "slowly" combines with "walk"
to form "slowly (walk)", which denotes the property of walking slowly.
 APPLICATIONS OF THE TYPED THEORY 143

This gives us a means of representing J-K-L consistently.

1. John believes Bill walked bravely ....
(.1) Bjthat-B(W)b (RE)
(.2) Bjthat-lHW)b (DICTO).
K. John does not believe that Bill walked courageously .. ..
(.1) '" Bjthat-C(W)b (RE)
(.2) '" Bjthat-(;iW)b (DICTO).
L. Walking bravely ... just is walking courageously . . .
(.1) B(W) = C(W)
Examples like G-L should demonstrate that our analysis for DE DICTO
belief is generalizable throughout the types. This treatment of beliefs about
higher order objects suggests a solution to the "paradox" of analysis.
Central to this puzzle are data triads similar to the ones we've been
discussing. Here is an example:
M. It is trivial that the concept brother is identical with the concept
brother.
N. It is not trivial that the concept male sibling is identical with the
concept brother.
o. The concept brother is identical with the concept male sibling.
Although there are various ways to state the puzzle precisely, all we need
to say is that the puzzle involves the question of how an identity statement
like (0) can be (an) informative (analysis). Philosophers who believe that
property terms denote sets and express properties, and who hold that
"brother" and "male sibling" express the same property are left with no
means of accounting for the informative nature of the identity statement
formed by flanking an identity sign with the property DENOTING terms "the
concept brother" and "the concept male sibling". What is to serve as the
senses of these expressions?
We suppose here that property analyses are sentences which say that
two apparently distinct properties are identical. We simply extend Frege's
view of their informative character by supposing that there are distinct
abstract properties which serve as the senses ofthe above property denoting
expressions. In order to represent M-N-O correctly, we need to note that
144 CHAPTER VI

strictly speaking, triviality is person relative - what is trivial for one person
may not be trivial for another. We assume that triviality is a relation
between persons and propositions. Consequently, we introduce "T" to be
a name of type (i,p)/p and we read "TXthat-¢" as: it is trivial for x that
¢. This forges an analogy with the other propositional attitudes. Terms
which follow the "it is trivial for x" prefix behave like terms in propositional
attitude contexts - sometimes they denote their senses.
Clearly, if we're restricting overselves to a discussion of a particular
individual S, then (P')-(Q')-(R') would be the proper way to capture the
triad P-Q-R:
P. It is trivial for S that the concept brother is identical with the concept
brother.
Q. It is not trivial for S that the concept male sibling is identical with
the concept male sibling.
R. The concept brother is identical with the concept male sibling.
P.' TSthat-Bs = EB
Q.' ~ TSthat-MSs=E B
R.' MS=B.
The proposition asserted to be trivial by (P) is g> fi'OU'§ 1 (Y' fi'ou'§ 2
(idenlilllE' IJ.ein{j a IJ.iollte,z), IJ.£ing a lJ,zotltei.). Its triviality derives
from the logical truth that being a brothers represents being a brother
to S. The proposition that's not trivial according to (Q') is g> fi'OU'§ 1
(Y'fi'OU'§2(idenlilllE,6ein{j a IJ.Mllte,z), IJ.ein9 a mate ailJl'infj.).
(P')-(Q')-(R') may be a good account of P-Q-R, but the original triad
was M-N-O. How are we to represent it? Well, since the English prefix
"it is trivial" as it occurs in (M) is not relativized to a particular individual,
it seems that (M) asserts that it is trivial for everyone that the concept
brother is identical with the concept brother. The relevant reading of (N)
seems to be: everyone is such that it is not trivial for them that the concept
male sibling is identical with the concept brother. If we recall that we
have allowed primitive variables of type i to serve as subscripts for sense
terms, then (M')-(N')-(O') seems to be the correct way to translate the
data triad:
M.' (x)TXthat-B x == EB
N.' (x) ~ TXthat-MS x = EB
Q.' MS=B.
 APPLICATIONS OF THE TYPED THEORY 145

An even closer representation of M-N-O would be one which uses the

iota-operator to capture the English definite article. Let us represent "the
concept brother" as (lyi/p)(y =B) and "the concept male sibling" as
(lyi/p)(y=MS). A consistent reading of M-N-O would be:
M." (x)Txthat-(li/p)(Y=lU= E(lyi/p)(y=B).
N." (x),,-, Txthat-(lyi/p)(y = MS) = E(zyi/p)(y = B)
0." (li/p)(y=MS)=(lyi/p)(y=B).

3. MODELLING IMPOSSIBLE AND FICTIONAL RELATIONS

A. IMPOSSIBLE RELATIONS

Sometimes we think about "impossible" individuals. These are not

individuals which are such that some contradiction is true. Rather, these
are individuals like theJ1 round square, which encode incompatible
properties. We can also think about "impossible" relations - the symmetri-
cal non-symmetrical relation is one. Here is an A PRIORI truth about this
relation:
(1) The symmetrical non-symmetrical relation is symmetrical.
We cannot analyze "the symmetrical non-symmetrical relation" as a
description involving exemplification formulas since it would fail to denote.
There are no higher order objects which exemplify both being symmetrical
and being non-symmetrical. (1) would be turned into a falsehood.
However, we may read the description as "the object which encodes
symmetricality and non-symmetricality", and then generate the A PRIORI
truth that this object encodes symmetricality. Let us suppose that the
properties in question are oftype ((i, i)/p)/p.14 We then have the following
instance of A-OBJECTS:
(3x(i,i)/P)(A !«i,i)/p)/px & (F«i,i)/p)/p)(x F:= F =S v F = S))
This axiom, plus the definition of identity among (i, i)/p-objects, justifies
our talking about THE abstract relation which encodes being syrrnnetrical
and being non-symmetrical. As in Chapter II, we define:
theJ1 symmetrical non-symmetrical relation =dJ
(lx(i ,i)/P)(A!x&(F)(xF:=F=S v F=S)),
dropping the obvious typescripts. It is then provable that theJ1 symmetrical
non-symmetrical relation encodes symmetricality. This theorem represents
(1).
146 CHAPTER VI

The analysis proposed for data about the d round square seems,
therefore, to generalize within type theory to the data about the d
symmetrical non-symmetrical relation.

B. FICTIONAL RELATIONS

When we see a sentence like "Einstein discovered that there is no such

thing as simultaneity", how are we to understand it? Did Einstein discover
that no two events ever exemplify the relation of simultaneity? Or did he
discover that the simultaneity relation does not exist?15 I'm not sure how
to decide the issue, but the latter reading seems a legitimate option. We
might therefore suppose that simultaneity is a fictional relation, and for
our present purposes, we could suppose that it is native to the Newtonian
(science) fiction. So let us identify it in a way analogous to our earlier work.
Let us suppose that events are special kinds of propositions, and that
they are type p objects. A relation among two events would be of type
(P,p)/p. Newtonian mechanics presupposes that simultaneity is a (possibly)
existing relation of type (P,p)/p. However, in our view, it must be an
abstract relation. Let "s" be a name of type (p,p)/p which denotes this
relation. We analyze (2) as (2)':
(2) Einstein discovered that simultaneity does not exist
(2)' Dethat- '" E ! !S16
But which abstract relation does "s" denote? Well, by typing the definitions
of character ("Char (xt, s)") and native ("Native (xt, st), we may construct
a typed N-CHARACTERS axiom such as: 17
(xt)(s)(Native (x, s) ...... x = (zzt)(ptIP)(zF == ~s(Fx)).
Let "n" denote the story of Newtonian mechanics. We then get the
following instance of N-CHARACTERS:
Native (s, n) ...... s = (zz)(zF == ~.Fs),
where "s" denotes simultaneity and is not a restricted variable. This may
prove to be an interesting way to identify other fictional relations of
disproven scientific theories.
Could someone write a story about non-scientific, fictional relations?
Could we dream about non-existent relations? If these are genuine
possibilities, we will have further da ta for the application of our type theory.
 APPLICATIONS OF THE TYPED THEORY 147

4. MODELLING MATHEMATICAL MYTHS AND ENTITIES

In this final section of applications, we conjecture somewhat about the

farthest reaches of metaphysical hyperspace. Thus far, we have found
applications for abstract individuals which encode possibly existing
properties, and abstract properties (relations) which encode properties·
of properties (relations). If we now recall that the theory even asserts that
there are abstract individuals which encode abstract properties, then it is
only natural to wonder whether we can find applications for these recondite
creatures. It also seems natural to suggest that mathematical entities just
are such abstract individuals. We devote the rest of this section to spelling
out this suggestion.
The first thing to do is to get clear on the data. We look for true
sentences of natural language whose truth seems to require that there be
mathematical objects to serve as the denotations of certain terms found
in them. The following sentences must surely count as such:

(1) In Peano number theory, zero is not a successor of any number.

(2) In Peano number theory, there is a prime number greater than
two.
(3) John wondered whether the number one hundred seventeen
thousand, four hundred and sixty seven is prime.
(4) In Zermelo-Fraenkel set theory, there is a set which has no
members.
(5) In Zermelo-Fraenkel set theory, for any given property and
set, there is a second set whose members are just those members
of the given set exemplifying the given property.
(6) In Zermelo-Fraenkel set theory, every transitive set has a
transitive power set.
A few comments are in order. The locutions "in Peano number theory"
and "in Zermelo-Fraenkel set theory" are NOT meant to be short for "in
the standard model of Peano number theory" or "in the standard model
of ZF". Rather, we pretheoretically understand these locutions as a kind
of "in the story" prefix and analogize the situation to that of fiction. We
will try to make this pretheoretical understanding of these locutions precise
by supposing that mathematicians author mathematical stories. The stories
148 CHAPTER VI

are usually communicated to other mathematicians by declaring a few

basic principles of the stories (the axioms of the particular mathematical
theory) with the understanding that all necessary consequences of the
principles are also to be part ofthe story. If the story is rich enough, then it is
of interest to other mathematicians to try to discover other truths according
to the story. We are supposing that it is a contingent fact that there are any
mathematical stories or storytellers.
Failure to appreciate the fact that sentences like (1)-(6) are the basic
data has resulted in a confused debate about the ontological status of
mathematical objects. The "platonists" who accept Tarski's theory of truth
both mistakenly suppose mathematical assertions unprefixed by the story
operator, like "there is a prime number greater than two," are literally
true, and also fail to distinguish the quantifier "there is" from "there exists".
Consequently, they conclude, for example, from the fact that the sentence
"there is a prime number greater then two" is literally true, that there
exists a (mathematical or metaphysical) object which satisfies the open
sentence "x is a number and x is prime and x is greater than two". This
conclusion seems unwarranted. On the other hand, the nominalists who
accept Tarski's theory of truth simply deny that such unprefixed sentences
are ever true and proceed to try to show that we need not make an
essential appeal to the truth of such sentences in any of our subsequent
(scientific) theorizing or problem solving. But such a strategy simply doesn't
account for all the data. For example, we get true sentences like (1), (2),
(4), (5), and (6) by prefixing the story operator to the theorems of
mathematics. And, of course, there are data like (3) (one cannot discard
data like (3) as being part of the problem of propositional attitude contexts).
Any serious attempt to account for the truth of the above data on a
compositional basis must attribute semantic significance to the number
words in these sentences in such a way that these words have the same
significance when they occur in sentences like (2) as they do when they
occur in sentences like (3).
Now that we've identified the data, we next identify the denotations of
all the names appearing in the data by utilizing and extending the
machinery developed in IV., Section 4 and VI., Section 3. Let us
concentrate initially on datum sentence (1).
First, we identify the story in question. Let us use "PNT" to abbreviate
"Peano's Number Theory". Given STORIES (IV., Section 4), and the
supposition that PNT is a story, we may assert:
PNT=(IZ)(F)(zF == (3FO)(kPNTFo &F = [JeyFO])).
 APPLICA TIONS OF THE TYPED THEORY 149

That is, Peano's number theory is that abstract individual which encodes
just the vacuous properties constructed out of propositions true according
to Peano's number theory. We therefore know, in principle, what the
denotation of "Peano's number theory" is as it occurs in (1). We have a
clearer idea of which propositions are true according to mathematical
stories than we do in the case ofliterature. For Peano's number theory, we
know that the conjunction of (7)-(11) is true according to the story:
(7) Zero is a number.
(8) Every number has a successor which is a number.
(9) No two distinct numbers have the same successor.
(10) Zero is not the successor of any number.
(11) Ifzero has some property, and a number's having that property
implies that its successor has that property, then every number
has that property.
In addition, a very strong principle governs the mathematical story
operator; where "8" ranges over mathematical stories :18
MATH-I:-SUB:(FO)(GO)(LsFO & FO => GO -> LsGO).
So, all the necessary consequences of the conjunction of (7)-(11) are also
true according to Peano's number theory.
What about the denotations of the other names in (1)? There are three
names to consider: zero ("0"), number ("N"), and successor ("S"). If we
assume that particular numbers are individuals which may exemplify
properties but are not themselves exemplified by anything, then "zero"
names an individual. Which individual? Well, for one thing, it is a native
character of the story. Given N-CHARACTERS, we may conclude:
0= (lz)(F)(zF == LPNTFO).
So zero encodes the following sorts of properties (which are properties it
exemplifies according to the story): being a number, not being the successor
of any number, being less than all other numbers, etc.
So far, this is a straightforward application of the theory we elaborated
earlier. But the interesting new twist comes as we try to identify the
denotations of "number" and "successor". We regard the property of being
a number as a native character of the mathematical story as well.
Consequently, it is an abstract, rather than existing, property. It is not
150 CHAPTER VI

the kind of property that could exist. Which abstract property? Consider
the following instance of typed N-CHARACTERS:
N = (IZi/p)(F(i/p)/P)(zF -=LpNTFN).

That is, being a number is that abstract property which encodes just the
properties it exemplifies according to the story. So it encodes the following
sorts of properties: being a property that the number zero exemplifies
([AFi/p FO]), being a property such that everything that exemplifies it has
a successor which also exemplifies it ([AFi/p(x)(Fx~(3y)(Fy&Syx))]),
being a property such that zero is not the successor of anything which
exemplifies it ([AFi/P(x)(Fx ~ '" SOx)]).
The successor relation is also an abstract relation which is a character
of the story:
S = (IX(i,i)/P)(F«(i,i)/p)/P)(zF -= LPNyFS).
Consequently, this relation encodes: the property of being a relation which
zero fails to bear to any other number ([AF(i,i)/p '" (3x)(Nx & FOx)]), the
property of being a relation such that for every number, there is a second
number which bears it to the first ([AF(i,i)/p(x)(Nx~(3y)(Ny&Fyx))]),
etc.
We have now identified, in principle, the denotations of the names
occurring in (1), and (7)-(11). We may translate these sentences into our
formal language, knowing what the significance of each term is, as follows:
(1)' LPNT'" (3x)(Nx & SOx)
(7)' NO
(8)' (x)(Nx~(3y)(Ny & Syx))
(9)' (x)(y)(N x & Ny & x =1= y ~ '" (3z)(N z & Szx & Szy))
(10)' '" (3x)(N x & SOx)
(11)' (F)(FO & (x)(y)(Nx & Ny & Syx & Fx~Fy)~(z)Fz).

In datum sentence (2), we have three terms, "even", "prime", and "greater
than", which are definable using the primitives of the theory. So we could
regard (2) as an abbreviation of a truth-of-Peano-number-theory in which
all the defined terms have been eliminated. But it may be preferable
to regard terms like "even", "prime", and "greater than", as denoting
complex abstract relations. There is an interesting field of investigation
here - the semantics is quite prepared for accommodating complex relations
 APPLICATIONS OF THE TYPED THEORY 151

which are constructed (using the logical functions) out of simpler abstract
relations. Also, the abstraction schema for relations remains neutral on
the question of whether the relations constructed or the constructing
relations must be possibly existing or abstract relations. So there seems
to be no reason why we can't suppose "E", "P", and ">" in the following
translation of (2) denote complex abstract properties and relations:
(2)' LpNT(:JX)(Nx & Ex & Px & x> 2).
Datum sentence (3) is now translatable, given this understanding of (2).
One may plug abstract individuals into abstract (complex) relations to
produce propositions. Propositions may serve as the objects of belief. It
may just be that the object of John's wonder in (3) is the proposition
which results by plugging the abstract individual 117, 467 into the abstract
relation of being prime. So we translate (3) as follows: 19
(3)' Wjwhether-P117,467.
Next, we consider (4) and (5). Given our discussion of Peano's number
theory, the identification of the denotations of the terms occurring in them
should be straightforward. Zermelo-Fraenkel set theory is a mathematical
story - it encodes just the vacuous properties constructed out of pro-
positions necessarily implied by the conjunction of the axioms of the
theory (as they are formulated in standard second order predicate logic).
The property of being a set ("S") is an abstract property which is a character
of the mathematical story. The membership relation ("E") is an abstract
relation which is also a character of the story. These characters may be
identified in a manner analogous to the identification of the abstract
property of being a number and the abstract successor relation. And, (4)
and (5) may be translated as follows:
(4)' LZF(:JX)(Sx & (y)y¢x)
(5)' LZF(F)(X)(SX --+(:Jy)(Sy & (Z)(ZEY == ZEX & Fz))).
(4)' and (5)' are both true since they say that the NULLSET and SUBSET
axioms are true according to ZF. Sentences like these guarantee that there
are lots of abstract individuals which exemplify-according-to-ZF the
property of being a set. It also follows that (4)' and (5)' are necessarily true.
Finally, consider (6). The property of being transitive is a property of
sets just as being prime is a property of numbers. It is therefore a complex,
abstract property. It is constructed, using the logical functions, out of two
other abstract relations, the membership relation and the subset relation
152 CHAPTER VI

(which is also complex). Consider the following two },-expressions:

~) Dx~ ~~~EX-ZE~]
(b) [hi (/)(YEX _ Y s; x)]
(a) yields a complex relation. When this such relation holds between objects
x and y, we write "x s; y". Of course, no two objects bear this relation to
one another, but lots of objects bear this relation to one another according
to ZF. We know further that since this relation is a native character of
ZF, that it is abstract and encodes just the properties it exemplifies
according to ZF.
(b) yields the complex property of being transitive ("T"). Again, no
objects exemplify this property, but many objects exemplify it according
to ZF. And since it too is a native character of the story, it may be
identified as that abstract property which encodes just the properties that
the property of being transitive exemplifies according to ZF. Now if we
let "P" abbreviate [hy (U)(UEX == uS; y)], we may then translate (6) as:
(6)' LZF(X)(Tx -(3y)(Pyx & Ty)).
The truth conditions for (6)' are compositional and should be precise.
By translating just these six pieces of data, we should have given the
reader a good idea of how we would answer questions that have faced
philosophers in the other, traditional approaches to this topic. We've
presented a view on the nature of mathematical truth, for example, simply
by identifying the data as we have. If the translations are successful, not
much more need be said on this question, since the metaphysical theory
provides us with lots of consequences of the translations. These consequen-
ces will be unacceptable only to those who already have some favorite
philosophy of mathematics. Another question which has turned out to be
quite a conundrum arises in connection with model theory. Given that
there are several ways to construct models of number theory in set theory,
which of the structures that do the modelling really are the numbers? We
close with a brief discussion on this topic.
The question in fact seems a little misguided. All that the model-theoretic
facts show is that mathematicians could have carried on without a separate
story about the numbers. But as long as the story of numbers can be
considered to be distinct from the story of sets, numbers and sets have
separate ontological status. The stories just have different characters.
Model-theoretic facts about the possible models of a theory have no
 APPLICA nONS OF THE TYPED THEORY 153

bearing on the question of what the mathematical entities ARE. This is a

question decided by a metaphysical theory. Indeed, no model-theoretic
facts about the interpretation of our metaphysical theory change any of
the definitions or theorems telling us about the nature of possible worlds,
monads, or fictional characters. Such facts only tell us whether the theory
is consistent, whether the logic is strong enough to deduce the consequences
of the theory, or whether the theory forces a categorical structure for its
interpretation, etc.
We have attempted to produce an explanation of what mathematical
entities and stories are entirely within the framework of metaphysics. We
have done so by forging a strong link between myth and mathematics.
Our account helps us to explain the semantic significance of number words
(in so far as they are used as nouns or names) as they occur in natural
language, as well as the semantic significance of other special words of
mathematics. And we have verified what mathematicians have claimed
all along, namely, that sets, numbers, etc., are abstract objects. And this
should come as no surprise to those who take metaphysics to be
ontologically prior to mathematics.
There is one last puzzling group of data: non-mathematical statements of
number which take the form "there are n F's." For example, "there are two
planets." A pure logicians' analysis of this example is: there are distinct
objects x and y which are both planets and all other planets are identical
with x or y. We know such a procedure can be generalized. But an
important alternative view is: number words in non-mathematical state-
ments of number denote "natural" numbers. And there is a "natural" way to
identify such numbers using the metaphysical and logical machinery of our
system, without appealing to any mathematical notions: let "zero" denote
the A-individual which encodes just the properties which fail to be
exemplified; let "one" denote the A-individual which encodes just the
properties which are uniquely exemplified; etc. Then, a metaphysical
analysis of "there are nine planets" would be: nine encodes the property of
being a planet. This "DE RE" reading is necessarily true if true, as opposed to
the pure logicians' contingent "DE DICTO" reading, which is compati ble with
the truth that it's possible that there are ten planets. The metaphysical
analysis could be an example of a necessary, A POSTERIORI truth. Both
readings appear to be legitimate options and reveal a subtle ambiguity in
the data. I commend these considerations to my readers.
 CONCLUSION

I do not plan to specifically argue that I have established the thesis of

the book, namely, that our research program is a progressive one. It is
for the reader to decide, on the basis of the material presented in
Chapters II, IV, and VI, whether the theory anchoring the program both
helps us to explain our data and, together with the auxiliary hypotheses,
predicts hitherto unknown, novel facts. Nor do I intend to argue here
that the theory generates more interesting consequences and can correctly
represent more interesting pieces of knowledge than any of its competitors.
Readers with a firm grasp on (the outstanding data facing) other research
programs must decide which provides the most natural, elegant, and
unified treatment of the variety of basic problems tackled here. Instead,
I would like to end with a few paragraphs in which I conjecture about the
even greater variety of future research possibilities which flower in the
foregoing formal garden. I will identify issues which may be of interest to
logicians, pure and applied metaphysicians, linguists, and pure and applied
epistemologists.
We may have provided logicians with a new "primary interpretation"
for second order logic. Since we have a precise, axiomatic theory of
relations, there is no longer any reason to suppose that the n-place relation
names and variables of second order languages denote (range over) sets.
Interpretations in which they denote relations seem to me metaphysically
prior and preferable to the traditional primary interpretations. Con-
sequently, standard incompleteness results for the second order predicate
calculus, which are based on interpretations in which the domain of n-place
relations just is the power set of the d h Cartesian product of the domain
of objects, do not apply to the elementary object calculus (or to any
second order predicate calculus interpreted in a similar manner). We
must look to the theory of relations to give us the basic facts about the
(size of the) realm of relations. This approach to the semantics of
second order languages should provide a wide range of metatheoretic
research possibilities (I have tried to formulate the model theoretic
semantics as elegantly as possible to facilitate future work on metatheoreti-
cal questions).

154
 CONCLUSION 155

For those who prefer pure metaphysics to model theory, the most
promising new line of research concerns time. There should be a way to
model instants of time in much the same way that we have modelled
possible worlds. If the basic tense operators are added to our language
(and interpreted in the normal way as quantifiers over an ordered set of
times, taken as primitive), the definition of a world should convert into a
definition of a time. That is because the diamond operator in the second
clause of the definition takes on new significance. Semantically, this clause
would tell us " ... and at some world and time, z encodes just the vacuous
properties constructed out of propositions true there and then". All of the
theorems about worlds should also convert into theorems about times.
The interesting project would be to then try to reconstruct the worlds by
finding the relevant A-objects which harness all the instants of time that
occur at a given world.
There are other research possibilities for the pure metaphysician, since
many traditional and modern issues can be reanalyzed from our new
perspective. For example, traditional ontological arguments could be
reanalyzed in a way which does justice to their logic, since the theory of
A-objects makes it possible to coherently reason about a thing without
prejudice as to whether that thing exists. Arguments may fail because the
exemplification-encodes ambiguity infects premises involving singular
predications. The theory however would not provide support for any A
PRIORI argument which concludes that THERE EXISTS an object which
exemplifies every property God exemplifies according to the story.
Having a precisely demarcated background ontology might also make
worthwhile a reexamination of such modern questions as the ontological
status of kinds, minds, works of art, etc. Could a case be made for
thinking that these types of entities are species of abstract objects, or must
they be assigned separate ontological status? Pure metaphysicians may
ask about our theory of identity. Is it general enough? Also, I have tried
to make suggestions along the way for extending the theory which may
be worth pursuing. The theory may be rich enough for others to create
new definitions, discover interesting consequences, add new auxiliary
hypotheses, etc.
-in aQ(hbon· to .traQ1i:lOn~u -quesircml; -lWWlh -i'&e;reYle.., 'wte&\~lllS' cyJ'Do
nymy, analyticity, and realism (all about which a cogent view may naturally
evolve from the theory), applied metaphysicians (for example, philosophers
of language) might be interested in the systems which result by adding
context dependent names to our syntax. Context dependent names would
156 CONCLUSION

receive both a denotation and a sense relative to each context of utterance.

It might be possible to render pronouns such as "I", "he", "she", "you",
and "it" as such context dependent names. This would help us explain
why we can't eliminate such pronouns in certain belief contexts through
substitution of co-referential proper names, since it might be the senses
of such expressions (relative to the given context) which are involved
rather than their denotations.
Also, the development of a Montague style fragment of English would
be of interest to philosophers of language and linguists. Since the data we
considered consisted of true English sentences, it becomes important to
show that our formal languages have enough resources to provide a general
semantic treatment of all such sentences (I have relied on a certain tradition
of translation in the applications). The resulting system would differ from
Montague's in many important respects. Provisions would have to be
made for the structural ambiguity we may have discovered in singular
predication. English names and descriptions would be ambiguous in belief
contexts, given our Fregean solution to the puzzles of belief. Belief contexts
would not involve the same scope ambiguities which Montague relies on
to differentiate DE RE and DE DICTO beliefs, though certain other scope
ambiguities will still be present. Since monads mirror their worlds and
consistent stories may be systematically related to worlds, it might prove
theoretically useful to dispense with existing objects in the immediate
interpretation of English. That is, it might simplify the rules of inter-
pretation. The idea is that proper names of existing individuals could
denote their monads and names and (in)definite descriptions of discourse
objects could denote characters of stories. At present, however, this may
just be a wild conjecture and a great deal of research must be carried out
before we could determine whether such a project is going to pan out.
However, I do not think that we'll encounter the same problem of finite
representability which faces Montague's systems, since the terms of our
language denote entities which are NOT functions defined on infinite
domains.
This brings us to the last area for future investigation we will
discuss - epistemology. Since thinking of, dreaming about, searching for,
and worshipping objects all may involve a complex relationship to
A-objects, pure epistemologists may consider postulating a basic kind of
acquaintance relationship which might serve to ground these intentional
relations. The idea is that we analyze worshipping Zeus, searching for the
fountain of youth, thinking about Hamlet, etc., in terms of acquaintance
 CONCLUSION 157

with these objects plus different intellectual (possibly propositional)

attitudes we adopt toward them. This relationship of acquaintance is not
a causal one, though we may come to bear this relation to A-objects, in
part, through causal interactions with copies of novels, storytellers, etc.
(The causal theory of names seems to me to fail. A-objects do not have
spatial location, and so no "dubbing", in the customary sense, of an
A-object ever takes place. But I hope we have provided a good enough
case for thinking that some proper names denote A-objects).
Nor is this acquaintance relationship the same kind of intellectual
acquaintance we have with properties. We must certainly be acquainted
with properties like being red, being round, being sharp, etc. We may all
agree that no matter what world we are placed in, we could recognize
whether or not an object in that world was red, round, sharp, etc. Yet
supposing that we are acquainted with properties seems to me to be a
natural explanation of this fact. These distinct acquaintance relationships
may serve as the basis for appending an epistemology to our metaphysics.
I think that a reasonable epistemology can be found, as long as we do
not suppose that we can gather knowledge about abstract objects in so
far as they are objects. The only proper knowledge that we can have with
respect to abstract objects is A PRIORI. But this does not rule out the fact
that we can gather knowledge about them in so far as they are fictional
characters, for example.
Applied epistemologists who investigate procedural models of semantic
competence and performance might avail themselves of abstract objects
as well. In building a program which models the acquisition of proper
names, maybe we should insert a subroutine which collates the available
assertions involving the name being introduced and associates with the
name the A-object which encodes just the properties denoted by the
predicates of such assertions. Such A-objects might serve as constituents
of propositions to be constructed when processing future assertions
involving the name. I am not sure how the mind works, but in storytelling,
the assimilation of each new assertion involving the name or co-referential
pronoun may involve a switch to a new abstract object which codes up
the properties jointly involved in the new and previous assertions. A full
procedural representation of discourse might therefore involve entire
sequences of abstract objects.
Of course, what is needed now is a group assault. It is my hope that
these suggestions are worthy of such a group enterprise.
 APPENDIX A

MODELLING THE THEORY ITSELF

This appendix will be divided into three parts. In part A, we discuss the
paradoxes of encoding and their joint solution. In part B, we describe an
extensional model of the monadic portion of the elementary theory,
suggested by Dana Scott. In part C, we describe Scott's model of the
monadic portion of the modal theory.

PART A

Just as in set theory, unrestricted abstraction schemata lead to paradox.

However, in the theory of abstract objects, it is the joint operation of two
unrestricted schemata which proves to be inconsistent, A-OBJECTS and
A-EQUIVALENCE. A-EQUIVALENCE has been restricted indirectly,
and RELA nONS has been restricted directly, so as to avoid these
paradoxes. There are two paradoxes to consider, one by Romance Clark
and the other by Alan McMichael, and they both stem from a common
source. It seems to me important to sketch the proofs so that the reader may
see how they arise from the unrestricted versions of our axioms.
Suppose we dropped the two major restrictions on A-formation and
RELA nONS (i.e., the restrictions imposed by the definition of pro-
positional formula). We could then form the following two A-expressions:
[Ax (:3F)(xF & '" Fx)] ("encoding a property that is not exemplified"), and
[Ax (F)(xF ~ Fx)] ("exemplifying every property that's encoded"). Alter-
natively, we could produce instances of RELA nONS as follows:

(3F)(x)(Fx == (3G)(xG & '" Gx))

(3F)(x)(Fx == (G)(xG ~ Gx)).

But then consider the following argument ("Clark's paradox"), first
reported in Rapaport [1976J, p. 225:
Consider the abstract object ao which encodes just [AX (3F)
(xF & '" Fx)], and suppose it exemplifies [Ax (F)(xF ~ Fx)]. By AE, it

158
 MODELLING THE THEORY ITSELF 159

follows that (F) (aoF ~ Fao), so ao must exemplify [AX (3F)(xF & "-' Fx)]
as well as encode it. Again, by AE, (3F)(a oF & "-' Fao), i.e., "-' (F)
(aoF~Fao). But then ao must fail to exemplify [A.x(F)(xF~Fx)] (by
A-EQUIVALENCE and == E), contrary to hypothesis.
So suppose ao fails to exemplify [Ax(F)(xF~Fx)]. Then "-'(F)
(aoF ~ Fao), i.e. (3F)(a oF & "-' Fao). Call this property" R" and note also
that by AI, ao exemplifies [Ax (3 F)(x F& "-' Fx)]. Since ao encodes just one
property, R must be [Ax (3F)(xF & "-' Fx)]. But by definition of R, ao fails
to exemplify R, i.e., "-' [Ax (3F)(xF & "-' Fx)]a o, contradiction. 181
A second contradiction would also be provable because we could form
the following A-expression: [AY y=x] ("being identical to x"), where this
abbreviates a much longer A-expression with encoding subformulas and
relation quantifiers. Again, by RELA nONS, we would know that there
is such a property. But then consider the following argument ("McMichael's
paradox"), first reported in a footnote to our [1979b]:
By A-OBJECTS, we have that (3x)(A!x & (F){xF == (3u)(F = [Ayy = u]
& "-'uF))). Call this object al and consider the property [AY y=a l ].
Assume that al encodes [AY y=alJ. By definition of al' we know
(3U)([Ayy=a l ]=[AYY=U] & ,,-,u[Ayy=a l ]). Call this object a z• So,
[AY y=a l ] = [AY y=a z] & ,,-,az[AY y=alJ. By =1, we know a l =a u and
by AI, we know [AY y=al]a l . Since [AY y=a l ] = [AY y=a z], it follows by
=E that [AY y=aZ]a l · So by AE, a l =a z. But then, ,,-,al[AY y=a l ]
(from the definition of az and = E), contrary to hypothesis.
So suppose that "-'al[AY y=alJ. By definition of al' ,,-,(3U)([AY y=
al]=[AYY=U] & ,,-,u[Ayy=a l ]). That is, (U)([Ayy=al]=[AYY=U]~
u[Ayy=a l ])· But since [Ayy=al]=[Ayy=a 1 ], it follows that al[Ayy=
al]. Contradiction. ~
It is doubtful that the source of these paradoxes lies with the presence
of relation quantifiers in A-expressions. Logicians have not found any
special trouble with the second order predicate calculus, in which one
finds relations defined with quantification over other relations. For
example, here's a standard instance of the relations schema of the second
order predicate calculus:
(3F)(x)(Fx == (VG)Gx).
This asserts that there is a property of "having all properties". This property
would be denoted using "[A.x(VG)Gx]". Properties such as these do not
seem to cause any special consistency problems. The only reason for adding
the "no relation quantifiers" restriction on A-expressions is that given the
160 APPENDIX A

style of semantics we have employed, it is rather complicated to interpret

such expressions without the resources of type theory (in type theory, we
suppose that "[Ax (VG)Gx]" abbreviates "[Ax (VG)ExGx]", where "Ex" is a
predicate which denotes the exemplification relation between a property
and an object which exemplifies it. We then interpret this latter A-expression
as OU% Ji"f/ 1(d.1,/(Ex)) i.e., as the first universalization of this exemplifi-
cation relation).
Consequently, the elimination of encoding subformulas from the
abstraction schema for relations seems to be the most theoretically
satisfying way of avoiding the paradoxes. McMichael first suggested this
move to me while I was writing [1979a], though at the time it turned out
to be insufficient. That was because the language which was being used,
had the logical notion of identity as a primitive. When McMichael
discovered his paradox while we were writing [1979b], we realized that
we would have to place extra restrictions on A-EQUIVALENCE and
RELA nONS. We had to banish primitive identity formulas from these
schemata as well as encoding formulas. However, early in 1980, I discovered
that identity for properties and relations could be plausibly defined in
terms of encoding formulas. The key to this discovery was D3 (Chapter I,
the definition for relation identity), which was forged during a search for
a complete theory of relations to accompany the theory of abstract objects.
As long as one uses the defined notion of relation identity and eliminates
the primitive logical notion from the language, one cannot generate a
relation of identity. [Axyx = y] is ill-formed and the definiens in D3 cannot
be used as the formula ¢ in RELA nONS. So it turns out that the
paradoxes stem from a common source, since the elimination of encoding
formulas alone from the abstraction schema for relations suffices to prevent
both paradoxes.

PART B

Dana Scott has suggested the following extensional model of the monadic
portion of the elementary theory. By an "extensional" model, we mean
one in which properties and relations with the same exemplification
extensions are identical. Although the theory does not require this
identification, doing so facilitates model construction a great deal without
calling the consistency results into question. However, as a project for
future research, we should look for models of the theory in which such
an identification is not made.
Since we are simplifying matters by considering only the monadic
 MODELLING THE THEORY ITSELF 161

portion of the theory, we shall construct a model in which the following

two axioms are true:
(:3F)(x)(Fx == CPo), where F is not free in CPo and CPo has no
encoding formulas
(h)(F)(xF == cp), where x is not free in cPo
We need not concern ourselves with the existence predicate or with
NO-CODER since in the following interpretation, only "abstract" objects
will do any encoding.
The interpretation is constructed in ZF plus individuals. Let us call our
set of individuals "lC"', so that it corresponds with terminology defined at
the end of Section 2, Chapter I. We will use "e" as variables ranging over
the members of Iff. Now let "r!li" be the set which is the union of two copies
of the power set of Iff, where the members of the first copy are coded with
a plus (or a one) and the members of the second copy are coded with a
minus (or a zero). That is,

The members of ill will be our properties, and we use "i" as variables
ranging over them. Also, we use "d" as variables ranging over the members
of the power set of Iff. So, for every i, there is a unique d such that either
r is identical with <+ , d) or i is identical with <- ,d). Also let us say
that the absolute value of i is just the set d such that i = <+ ,d) or
i = <-, d). We write "Iii" to designate the absolute value of i. Finally,
let ".91" be the power set of r!li. These will be our abstract objects, and we
use "0," as variables ranging over these objects.
Now in terms of these sets, we can specify the interpretation which we
suspect is a model and say what exemplification and encoding amount
to. Recall that an interpretation of the theory will be the form <!?J, r!li, ext~,
ext d, $'). So let !?J be the union of Iff and .91, and as usual, we use "0"
to range over the members of !?J. Let ill be just the r!li defined immediately
above. And in terms of this picture, we now state the conditions under
which an object exemplifies and/or encodes a property:
(1) oEext~(i)iff either
(i) (:3e)(o=e & (:3d)((i=< +,d) v i=< -,d» & eEd», or
(ii) (:30,)(0=0, & (:3d)(i=< +,d»)
(2) oEextd(i) iff either
(i) (:3e)(o=e & i =1= i), or
(ii) (:30,)(0=0, & iEo,).
162 APPENDIX A

So clause (li) ensures that an existing object exemplifies a property t just

in case it is an element of Itl. Clause (Iii) ensures that an abstract object
exemplifies a property t iff t is a plus-marked set of e'S. Since this condition
is vacuous with respect to the abstract objects, either all abstract objects
exemplify a given property or none do. Clause (2i) ensures that no existing
objects encode properties, since the condition i t i is never satisfied. Clause
(2ii) ensures that an abstract object a encodes a property t just in case t
is an element of a.
Now we need to prove the following claim: that both (3F)(x)(Fx cPo) =
and (3x)(F)(xF = cP) are true under this interpretation. The claim can be
demonstrated quite easily once the following fact and lemma are seen to
be true. The fact is that "(x)cP" is true iff both (e)cP' and (a)cP' are both
true, where cP' is the semantic translation of cPo Clearly, universal claims
of the object language are true just in case they hold with respect to all
existing objects and with respect to all abstract objects. The lemma we
need can be stated as follows:

Invariance Lemma: Given the above interpretation, let an assignment to

the variables, I, be fixed. Then, where cPo is an any formula with no
encoding formulas where x is free,
(3/')(3a)(/'x/&I'(x)=a &1' satisfies cPo)=
(/')(a,)(/', 1& I '(x) = a -> I' satisfies cPo)·

If we allow ourselves to talk derivatively about objects satisfying formulas

(instead of strictly talking in terms of assignments to the variables satisfying
formulas), then our lemma may be read quite simply: an abstract object
satisfies a formula cPo with no encoding formulas iff all abstract objects
do. This is a consequence of clause (Iii) above, which places an vacuous
condition on an abstract objects' exemplifying a property. Since the details
of the proof are somewhat messy, we save the proof of this lemma until
the end. For now, let's suppose that it's true.
=
Then consider the first axiom: (3F)(x)(Fx cPo), where cPo has no
encoding formulas and no free F's. Semantically, we have to show:
= =
(3i)(e)(eEext.@(i) cP~) & (a)(aEext.@(i) cP~», in view of the above fact
(again, cP~ is the semantic version of cPJ So let's describe a way to choose
i so that the conjunction of universal claims is true. Take an arbitrary
cP~ and let D be the set of existing objects satisfying cP;. Clearly there must
be such an D, since we have all the possible subsets of ~ to choose from
(we are appealing here to the fact the following instance of the SUBSET
 MODELLING THE THEORY ITSELF 163

axiom of ZF governs our interpretation: (3Y)(Z)(ZEY == zE0" & ¢~)). Now

take an arbitrary abstract object, say as. If as satisfies ¢~, then choose
-i to be <+ ,d). Clearly, if an arbitrary e is an element of ext8i!(z), then
eEd, by the above clauses for exemplification. So e satisfies ¢~. And if e
satisfies ¢~, then e is an element of d, and so eEext8i!(-i). So,
(e)(.eEext8i!(-i) == ¢~). Since as satisfies ¢~, all abstract objects do. And since
-i is <+,d),(a)(aEext8i!(-i)). So (a)(aEext8i!(-i)==¢~). Consequently, our
conjunction of universal claims is true.
On the other hand, if as fails to satisfy ¢~, choose -i to be <- ,d ).
Again, (e)(eEext8i!(i) == ¢~), since if an arbitrary .eE.ext8i!(i), then eEd and
e satisfies ¢~ (and vice versa). Since as fails to satisfy ¢~, no abstract
object satisfies ¢~. So every abstract object a satisfying ¢~ is such that
aEext8i!(i), by failure of the antecedents. Since i is <-,d), no abstract
object a is an element of ext 8i!(-i). So every abstract object a which is an
element of e xt8i! (i) satisfies ¢~, again by antecedent failure. So,
(a)(aEext8i!(i) == ¢~). So our procedure for choosing -i always guarantees
that the conjunction of universal claims is true. Hence our first axiom is
true in this interpretation.
Consider, now, (:J x)(F) (xF == ¢), where ¢ has no free x's. Semantically,
we have to show: (:Je)(i)(eEextd(-i) == ¢') v (:Ja)(i)(aEextd(i) == ¢').
Clearly, the right disjunct is the true one, and this is easily verified. For
an arbitrary <p', there is an element of sf which has as members all and
only the properties -i satisfying <p'. That's because sf contains all the
subsets of fJI, so pick the one whose elements are just the properties
satisfying <p' (again, we are appealing to the fact that the following instance
of SUBSET is true in the above interpretation: (:JY)(Z)(ZEY == ZEfJI & <p')).
It remains, then, to prove the Invariance Lemma:
Proof. By induction on the complexity of <Po.

(1) Since I is fixed, let f( G) be -i. (----» Assume the

antecedent and suppose It (x) = al and It
satisfies <Po, where luI So, 0,1 Eext8i!(-i), by the
definition of satisfaction. So, (:Jd)(i=(+,d»).
Now assume 12(X)=a2 and 12xl We need to
show (:Jd)(i=( +,d»). We already have it. So,
V')(a)(/~xl & I'(x) = a----> /' satisfies <Po}· (~)
trivial.
(2) Again, assume the antecedent and suppose
( ----> )
that A(X)=al and Axl and A satisfies <Po· So
164 APPENDIX A

II fails to satisfy t/I. Now the inductive hypothesis

is a biconditional and the existence of II shows
that the right side of the biconditional is false.
So the left side must be also, that is, ~ (3 I')
(a)(I'x/ &I'(x)=a & I' satisfies t/I). So, (I')
(a)(f'x/ & I'(x) = a--+ I'failsto satisfylfr). Now
assume Iz(x) =az and Iz x/ We want to show
that Iz satisfies cPo· But Iz fails to satisfy t/I, since
it is an I' which satisfies the antecedent of the
universal claim. So, Iz satisfies cPo. (<-) triviaL
(3) cP 0 = (t/I-+ )(). Exercise
(4) (--+) Assume the antecedent and suppose that
II(X)=a1 and luI and II satisfies cPo. So, every
I~&A is such that I~ satisfies t/I. Now assume
?l(x)=aZ and ?ul We want to show that?l
satisfies cPo. So, we want to show that every ?~a?1
is such that ?~ satisfies t/I. So, let us show, for
an arbitrary ?~, say ?j., that ?t satisfies t/I.
Pick an I~, say It, such that It(o:) = ?t(o:)·
So, If satisfies t/I. Hence, (3/~)(I~(x)=a1
and I~x/and I~ satisfies t/I). We may therefore
invoke the inductive hypothesis to get: (A)(a')
(A(x) = a' & Ax/--+ A satisfies t/I). Since 9~~91'
9t(x)=az. Now if we can show that 9'1';;1,
we can suppose that ?f is one of the A's and
conclude that 9t satisfies t/I. Since 9ul and
?f~l' it follows that Ax.• / But 91'(0:)= 11'(0:)
and Ifa/ SO 9f.J (<-) exercise. ~
PART C

I am indebted here again to Dana Scott for suggesting the following

interpretation which serves as a model of the modal version of the monadic
portion of the theory. The proof runs almost exactly like the consistency
proof of the elementary theory, and so we simply sketch the proof below.
We shall describe the interpretation in enough detail for the reader to
easily fill in the rest. First, we need to describe a few sets and then we
will identify the elements of the interpretation in terms of these sets.
Let 11/ and Iff be any two non-empty sets. 11/ will be the set of possible
 MODELLING THE THEORY ITSELF 165

worlds and C will be the set we'll use to help construct our possibly
existing individuals. The set of possibly existing individuals ("&>C") is
simply the set of all partial functions from if/ into C (i.e., &>C = C· 1V). We
let "fi." be variables ranging over the elements of &>C. These functions
have sometimes been known as "individual concepts". A given function
takes each possible world and either maps it to an element of C or is
undefined for that world. We use partial functions to represent the real
set of possibly existing individuals because the members of this latter set
may fail to exist in various worlds. Also, note that in intended inter-
pretations, the partial functions would be constant when defined.
Next, we need to construct our properties. To do this, first consider the
set which is the union of two copies of the power set of [1}C, where the
members of the first copy are coded with a plus and the members of the
second copy are coded with a minus. That is, first consider the set $:

Now we define fll to be the set of all functions from if/ into $, i.e., fll = $11'.
We take here the set of all total functions from if/ to $ because our logic
is two-valued. Now let us define the absolute value of property i at world
w("lil",") to be the set ° such that i(W) = <+ ,0) or i(W) = <- ,0). Here,
"0" is a variable ranging over the members of the power set of [1}~' Finally
let d be the power set of f1ll.
We then define the interpretation <if/,!!fi, fll,ext ""ext,sd, g;) as follows:
if/=if/
!!fi=f1J>~ud
fll = fll.
And we define exemplification and encoding as follows:
(1) oEext",(i) iff either
(i) (3fi.)(0 = fi. & (30)((i(W) = <+ ,0) V i(W) =
< -,0») & fi.Eo)), or
(ii) (3.a)(0=.a & (30)(i(W)=< +,0»))
(2) oEext,sd(i) iff either
(i) (3fi.)(0=fi. & qb'i)
(ii) (3.a)(0=.a & iEa.).
If we recall here that satisfaction of formulas is defined relative to possible
worlds, then these definitions have certain consequences. Let I be an
166 APPENDIX A

arbitrary assignment to the variables, and suppose that I(F) = ,z and that
I(x) = o. Then clause 1 above guarantees that I satisfies Fx with respect
to world £0 iff one of the following two conditions holds: (i) 0 is some
possibly existing object fz and fz is an element of l,zl~, or (ii) 0 is some
abstract object a and the value of ,z at £0 is a plus marked set of possibly
existing individual's. Since this latter condition is a vacuous one, if some
abstract object satisfies an atomic exemplification formula at some world,
all abstract objects satisfy the formula at all possible worlds. The Invariance
Lemma will guarantee that this happen for all formulas cPo which have
no encoding formulas. Clause 2 has the result that I satisfies xF with
respect to world £0 iff one of the following condition holds: (i) 0 is
some possibly existing object fz and ,z is not self-identical (so this never
happens), or (ii) 0 is some abstract object a and ,z is an element of a.
Clause (2i) guarantees that the following is true in the interpretation:
<>E!x---+O ~Cjp)xF. Clause 2 in general ensures that the following is
true: (x)(F) (<>xF ---+ 0 xF). That is because the satisfaction condition for
encoding formulas ends up being independent of the possible worlds.
It should be clear that a proof similar to the one constructed in part
B can be carried out. We want to show that the following two axioms
are true in the above interpretation:
(:3 F) 0 (x)(Fx == cPo), where cPo has no free F's and no encoding
formulas
o (:3x)(F)(xF == cP), where cP has no free x's.
Semantically, we have to show:
(:3,z)(£O)((fz)(fzEextj,z) == cP~) & (a)(aEext,),z) == cP~))
(£O)((:3fz)(,z)(fzEext"i,z) == cP') V (:3a)(aEext.,;(,z) == cP'))
To show that the first is true, we utilize a choice procedure for ,z just like
the one constructed in the elementary case. For any given world, the fact
that we have got all the (coded) members of the power set of [lJ>$ to serve
as extensions for our properties makes the left hand conjunct true, while
the Invariance Lemma will ensure that our choice of ,z makes the right
hand conjunct true. To show that the second axiom is true, just choose
a to be the set of properties ,z which satisfy cP' with respect to the given
world £0. There must be such a set since we've taken the power set of f1l
to be the set of abstract objects. [gJ (sketch)
 APPENDIX B

MODELLING NOTIONS

Throughout this work, we have talked about notions. Syntactic notions

like term, occurrence, the erasure of a formula ¢, etc., are not all that
mysterious - they seem to be BONA FIDE relations among linguistic objects.
However, there is a group of metaphysical notions which are rather
puzzling. These are all the notions constructed out of the primitive notion
of encoding. These notions fall into two groups: defined notions (such as
correlation, Form, Monad, World, complete, maximal, etc.) and paradoxi-
cal notions (such as exemplifying a property that is not also encoded,
exemplifying every property that is encoded, and identity). The former
group of notions may not be relations (since the formulas ¢ which would
"express' them violate restrictions on A-formation and RELATIONS) and
it is provable that the latter group could not be relations - if they were,
some contradiction would be true. The reason these notions are puzzling
is because as ontologists, we like to avoid uncategorizable entities
("ontological danglers"). So if these notions are not relations, what are
they? Do they have independent ontological status? Or is talk about
"notions" just a convenient reification, disguising metalinguistic talk about
objects which satisfy definitions?
Even if the answer to the last question is yes, it might be worthwhile
to look for a reification procedure whereby we do find some appropriate
object in our ontology to code up, or go proxy for our notions. An easy,
though risky way to do this would be to add an axiom which asserts that
there is a PRIMITIVE encoding relation and a relation behind everyone of
our DEFINED notions (and leave the paradoxical notions to dangle). Until
we have an easy way of confirming the consistency of the theories which
result, such a procedure seems suspicious and unsystematic.
There may be a better way, however; one which allows us to find proxies
for the primitive notion of encoding, the defined notions AND the
contradictory notions. The trick is to build the notional formula in question
into the defining formula for an abstract relation. Consider the following
two instances of A-OBJECTS:

167
168 APPENDIX B

(1) (:JZ(i,i/p)/P)(F«i,i/p)/p)/P)(zF::= (:Jxi)(:JHi/p)(xH &

F = [,W(i,i/P)/PGxH]))
(2) (:Jz(i,i)/P)(F«i,i)/p)/P)(zF::= (:J Xi) (:Jyi)((Hi/p) (yH ::= Hx) &
F = [,W(i,i)/PGXY])),

(1) says there is an abstract (i, i/p}/p-relation (between individuals and

properties of individuals), z, which encodes a property of such relations,
F, iff F is the property of: being a relation which relates an object Xi
with a property Hi/p it encodes, (2) says that there is an abstract relation
among individuals, z, which encodes a property of such relations, F, iff F
is the property of: being a relation which relates an object yi with its
correlate Xi, These two abstract relations are unique. It seems reasonable
to suppose that they could represent the primitive notion of encoding and
the defined notion of correlation, respectively.
Note that by abbreviating a crucial subformula, (Hi/P)(yH::= Hx), in (2),
we could rewrite (2) as:

(2)' (:Jz)(F)(zF::=(:Jx)(:Jy)(Cor(x,y) & F=[AGGXY]))

(3) and (4) are further examples:
(3) (:Jzi/P)(F(i/p)/P)(zF::= (:Jxi)(World(xi) & F = [AGi/PGX]))
(4) (:Jzi/P)(F(i/p)/P)(zF::= (:Jxi)(:JHi/p)(xH & ~ Hx) &
F= [AGi/PGX])).

These abstract properties could serve to represent the notions of being a

world and being a non-self-correlate, respectively, Necessarily, the latter
fails to have a weak correlate - if it did, some contradiction would be true.
It is important to see that this modelling of a contradictory notion
doesn't reintroduce Clark's Paradox, Let us use [AX (:JH)(xH & ~Hx)]
to denote the abstract property that (4) gives us. That is, we are using
what previously had been an ill-formed A-expression to name a unique
abstract object. But we cannot allow this A-expression to be used in
instances of A-EQUIVALENCE. That is because we have permitted
abstract objects of type t to encode abstract properties of type tip - so
we know that there would be an abstract object of type i which encodes
[AX (:JH)(xH & ~ Hx)J (the abstract property given by (4)). This would
be the first move in Clark's Paradox, The second would be to suppose it
exemplified [h (F) (xF ~ Fx)J (an abstract property constructed using an
axiom like (4)), But we can stop the paradox from developing to completion
 MODELLING NOTIONS 169

by not allowing A-conversion on the (formerly ill-formed) A-expressions

we have just used to name our abstract properties. Metaphysically
speaking, this means that we are not supposing that the abstract properties
we have chosen to represent the notions of being a non-self-correlate and
being a weak correlate of one's self have in their respective exemplification
extensions just the objects which are non-self-correlates or which are
weak-self-correlates. And in general, we do not suppose that the abstract
relations which represent our defined and contradictory notions have in
their exemplification extensions all and only the n-tuples of objects which
satisfy the defining formulas of these relational notions. This prevents the
paradoxes from being reintroduced.
A question then immediately arises. IT the proxy abstract relations do
not have the "appropriate" exemplification extensions, why is our modell-
ing procedure worthwhile? Well, the answer is that it is useful. IT we
generalize our procedure, we can handle data which we couldn't handle
before. First, here's our generalization:
where ¢ is any non-propositional formula, and 1X 1 , ... , IXn
are any variables of types t l' ... ,tm respectively, then:
[A1X 1... IXn ¢ J = abbr (IZ{t" ,.. ,tn)/P) (F{{t, ..... tn)/p)/P) (zF ==
(31X1) ... (3IXn)(¢ & F=[AG{t" ... ,tn)/PGIX 1 ···lXnJ))·
This gives us a way to easily name the abstract object which goes proxy
for the defined or contradictory notion "expressed" by ¢. This procedure
allow us to represent the following data triad:
A. S believes that Dostoyevsky wrote about the student who killed an
old moneylender according to Crime and Punishment.
B. S does not believe that Dostoyevsky wrote about Raskolnikov.
C. Raskolnikov is the student who killed the old moneylender according
to Crime and Punishment.
To get the denotation for the English definite description correct, it must
be symbolized as (IX)~cp(Sx & (3y)(OMLy & Kxy)). This is the way we
did things in Chapter IV, Section 4. But we cannot underline our represent-
ing description and turn it into a sense description because it is not
constructed out of a propositional formula. "(IX)~cp¢" is not a well-defined
sense description. So we did not have a definite description in our formal
language which had the right denotation and which, when underlined,
represented the sense of the English definite description.
170 APPENDIX B

But now we can do this. We can suppose that "(IXY'L cp ¢" was defined as
follows:
(ZX)LCpp = dJ(zz)(F)(zF == F = [Ax LCpcP &
(Y)(LCpcP~--+ Y =EX)J).

The A-expression used in this definition abbreviates a definite description

of an abstract property. (ZX)LcpcP is the abstract object which encodes just
this abstract property. Consequently, we may represent A-B-C as follows:
(A') Bsthat-Wd(IX)LcpcP (mcTo)
(B') ~Bsthat-Wd!.s (mcTo)

(C') r = (IX)LcpcP,
This should give the reader a good idea how to handle the data in
footnote 12, Chapter VI, Section 1. But what about the following data:
D. S believes that the person who killed an old moneylender according
to Crime and Punishment was a student.
E. S does not believe that Raskolnikov was a student.
F. Raskolnikov is the person who killed an old moneylender according
to Crime and Punishment.
To handle data like this, we would need to incorporate the new logical
function described in note 11, Chapter VI, Section 1. g %;?lJ 2'0Jt<;'§( t, 0-) is
the proposition that 0- encodes ,z. It would be denoted by "rp", where r
denotes 0- and p denotes t. We then get:
(D') Bsthat-(IX) LCp(PX & (:ly)(OMLy & Kxy)) S
(E') ~ Bsthat-LsS

(F') r = (ZX)LCp(PX & (:ly)(OMLy & Kxy)).

But there is still a problem about making this more general. Consider
G and H:
G. S believes that Porphyry arrested the student who killed an old
moneylender.
H. S does not believe that Porphyry arrested Raskolnikov.
To represent G and H correctly, and in a way which suggests a completely
 MODELLING NOTIONS 171

general treatment, we recall that "LCp¢" abbreviates "CP[Ay¢]". So can

we represent G and H as:
(G') Bsthat-CP[Ay Ap (IX)Lcp(SX & (3u)(OMLu & Kxu))]
(H') '" Bsthat-CP[Ay Apr.s].
The general solution is to use rff JV&12!uttcg on the vacuous property
encoded by Crime and Punishment to get the proposition that Crime and
Punishment encodes the vacuous property. This vacuous property is
denoted by a A-expression, [Ay Ap(IX)Lcp(SX & (3u)(OMLu & Kxu))]; but
in trying to capture the sense of the English description involved, we have
to underline the translating description, thereby denoting an abstract
object which encodes an abstract property.
 NOTES

INTRODUCTION

I Imre Lakatos, [1973], p. 4. I am indebted here to Robert Nola for discussions about
Lakatos' work.
2 In Chapters I -IV, we will use the word "object" to mean "individual". Objects are to be

distinguished from relations. However, in Chapters V and VI, the notion of an "object"
broadens - individuals and relations (properties, propositions) are all considered to be
"objects". Consequently, the abstract objects of Chapters I - IV are just abstract individuals,
whereas the abstract objects of Chapters V and VI include abstract individuals and abstract
relations.
3 Strictly speaking, data should not contain technical terms. But these hypotheses of Plato
and Leibniz do. So the sense of "data" being used in this case is simply "something to be
explained" .
4 We shall not attribute this theory to Russell, though we shall call this view "Russellian"

because so many philosophers seem to make the attribution. Russell maintained ([1918])
that the ordinary things we speak of as existing (you, my desk, sub-atomic particles) are
"logical fictions" (see p. 253, 270, 271). These things are not "ultimate simples out of which
the world is built" (p. 270). He adds facts to the list of things that there are, and he believes
that many unreal things like phantoms, hallucinations, and their constituents are in fact real
(pp. 257, 274-276). (I am indebted here to Mark Aronszajn for pointing out some of these
details.)
5 See the work of S. Kripke [1963], R. Montague [1974], and M. Cresswell [1973].

6 See the work of N. Goodman [1951], and H. Field [1980]. We take Quine to be one of

the less radical physicalists, but we may not be right in doing so. See Goodman-Quine
[1947] and Quine [1948].
7 See the bibliographical references to the works of Mally, Parsons, Routley, and Castaneda.

8 Readers who are unfamiliar with abstraction schemata should convince themselves that
example (c) does guarantee us an object which exemplifies both properties in question, and
no others, despite the fact that a disjunctive condition is involved. A perusal of the following
proof should do the trick:
(i) =
(3x)(F)(Fx F = R v F = S) (NOr)
(ii) (F)(Fa l = F=R v F=S) EE, (i) "al" arbitrary
(iii) Show: Ra l &Sa l & (F)(-(F=R v F=S)--> -Fa l )

(iv)
(v)
(vi)
fII
ShOW: Ral
Ra l =R=R v R=S
R=R
VE, (ii)
=1

172
 NOTES 173

(vii)
(viii)
(ix)
lRa l
R=RvR=S

Show: Sal
vi, (vi)
== E, (v), (vii)

(x) sal ==S=R v S=S UE, (ii)

(xi) [ S=S =1
(xii) R=SvS=S vI, (xi)
(xiii) Sal == E, (x), (xii)
(xiv) Show: (F)( ~ (F = R v F=S) -4 ~ Fa l )
(xv) Show: ~ (P=R v P=S)-4 ~ Pal "P" arbitrary
(xvi) [ ~(P=RvP=S) Assumption
(xvii) [ Pal == P=R v P=S UE, (ii)
(xviii) ~Pal == E, (xvi), (xvii)

9 See J. N. Findlay [1933], p. 176. Findlay refers to Meinong [1915], pp. 175-177.
10 There are a host of good papers by Parsons on the subject: [1974], [1975], [1978],
[1979b], and [1979c]. However, the most important statement of his theory is in his book
Nonexistent Objects [1980].
II Parsons [1980], Chapter IV.
12 Parsons [1980], Chapters VII, VIII.
13 Findlay describes Mally's theory of determinates in his [1933], pp.110-112 and

pp. 183-184. He cites Mally [1912], pp.64, 76. Mally's notion of satisfaction is to be
understood as our notion of exemplification.
14 Rapaport [1978], pp. 153-180. For some reason, Rapaport attributes the theory he is
working on to Meinong. He calls it a reconstruction of Meinong's theory. His Meinongian
objects clearly seem to be Mally's determinations. Although Mally uses the word "deter-
miniert", he also uses "Konstitutiven" ([1912], p. 64). Compare Rapaport's use of "being
constituted by".
IS In what follows, we use the word "abstract" purely as a piece of technical terminology.

Also, we take the words "existing", "actual", and "real" to be synonymous.

We could make a long list of entities which, at some time or another, philosophers have
supposed to be abstract. It is NOT to be presupposed that the set of abstract objects which
we will investigate is (intended to be) identical with the set of objects which some other
philosopher pretheoretically intuits to be abstract. Many philosophers have firm intuitions
to the effect that certain objects are abstract. However, these intuitions are rarely supported
by presenting precise conditions which tell us when there are abstract objects or which tell
us when any two abstract objects are identical.
There is both a prerogative and an intellectual obligation to specify how one plans to use
the word "abstract". This has been done informally with principles (I) and (II) which follow
in the text, and will be done formally in Chapter I (Sections I, 4). The notion we end up
with may not correspond exactly with that of others, but at least it should be clear. And in
the course of our investigations, we shall discover that certain objects that other philosophers
have taken to be abstract are identifiable among our abstract objects.
16 For a long time, I thought I had been the first to formulate (I) and (II). But I have
subsequently discovered that embedded in Rapaport's dissertation, we find principles roughly
similar to these in which Rapaport commits himself to sets. On page 190, we find the
174 NOTES

following principle, where "S" ranges over sets of properties, "F" ranges over properties, "0"
ranges over Meinongian objects, and "F c 0" means that the Meinongian object 0 is constituted
(determined) by F:
(T7a) (S)(3F)(FES --->(30)(F)(F co iff FES)).

And on page 184, we find:

OJ =0 2 iff (F)(Feo j =Fe0 2 )·

CHAPTER

j Every formula is a subformula of itself. If ¢ = (~l/I), (l/I & X), or (3rx)l/I, then l/I is asubformula
of ¢. If l/I is a subformula of X and X is a subformula of ¢, then l/I is a subformula of ¢.
2 See Alonzo Church [1941].
3 For convenience, we will read "E!y" as "y exists" and "[Axx=Eb]" as "being E-identical
to b", instead of using the more cumbersome readings.
4 These definitions are all standard.
S Compare Parsons [1980], IV, Section 3.

6 Some of these logical functions can be traced back to Moses Schonfinkel [1924]. Quine

developed his predicate operators Der, Inv, inv, Conj, Neg, etc., in his [1960], citing the
work of the combinatorial logician H. Curry. I borrowed i!l' !l'1I'!J from the 1978 manuscript
of Parsons [1980] and used it in the interpretation of the monadic theory of abstract objects
developed in my [1978] and [1979a] (Quine had no need for i!l' !l'1I'!J since his project was
to explain away singular terms). McMichael developed the other logical functions using
Quine's operators as prototypes. I learned these algebraic techniques from Alan, and these
functions were first used in our [1979b]. Several months later, we discovered that George
Bealer was working with logical functions similar to these.
Since [1979b], I have made some minor improvements on these logical functions. i!l' !l'1I'!J,
11%.fY,«/(!J%i', and iYltffff!l' have been turned into families of logical functions by indexing
them to the number of the place in the relation on which they operate. This allows us to
sharpen up the definitions so that we do not end up generating an infinite number of empty
relations like the plugging of a two-place relation in its 300 th -place. Also, in Chapter III,
% tff't! has been constructed and the extensions of all of the functions have been constrained
at all possible worlds. Finally, in Chapter V, the functions have been redefined so that they
operate throughout the branching type hierarchy.
7 I am indebted to T. Parsons for pointing this problem out to me. It was not until the

second draft that something was finally done about it.

8 One possibility I have yet to explore is a reference made by Quine in a footnote in [1960].
He says Bernays had developed a system which included axioms. So maybe there is such a
theory. Also, Tarski's cylindrical algebra or a polyadic algebra might be relevant here.
9 I was motivated to construct these definitions after reading Bealer [1981] (in manuscript),
Chapter 3. The definition he had constructed to partition his complex terms seemed too
complicated. I then realized that by indexing the place numbers to i!l' !l'1I'!J, and 11%..11',
and by ordering the rules for classification, a much simpler procedure for partitioning the
.ie-expression could be devised. Thanks goes to Alan McMichael for his valuable help in
working out many of the details in the following definition.
 NOTES 175

10 I'd like to thank M. Jubien for pointing out a flaw in an earlier version of the definition

of .I-assignment.
11 See Alfred Tarski [1931] and [1944].
12 With this definition of satisfaction, we may define what it is for a relation to be expressible

and define the important concept of logical consequence (model-theoretic, or semantic

consequence) :
Relation in of .I is expressible =dJ(3cf» (rjJ has n free object variables
VI'"'' Vn and (/)(1 satisfies rjJ '" d 1.,([AV 1". vncf>]) ~ -/n».o/ is a logical conse-
quence of cf> =di.l)(/) (/ satisfies ¢ ~ I satisfies 1/1).
13 I follow Eliot Mendelson [1964], p. 57. Also see Robert Rogers [1971], pp. 87-88.
14 ).-EQUIVALENCE first appeared with the formal interpretation it presently receives in
[1979b] (which was coauthored with Alan McMichael). It replaced the PROPERTIES
schema of my [1979a]. Since those early papers I have eliminated primitive identity formulas
from the lanl!l!a!!!; and drafted a definition of o[oopsitional formula. This allows a more
elegant formulation of A-EQUIVALENCE.
15 The axioms which follow represent the culmination of the process ofaxiomatization
which first began in my [1978] and [1979a]. The axioms found here are basically the same
as the ones found in these two early papers, the only difference being that the earlier versions
were couched in odd looking languages with primitive identity and which sorted terms
denoting (ranging over) A-objects from terms denoting (ranging over) existing objects. The
most important axiom, A-OBJECTS, was visualized after reading Parsons, Findlay, and
Rapaport.
16 The reader might wonder here why we have not just defined x = E Y instead of taking

= E as primitive. The reason is as follows: We shall want to be able to form A-expressions

like [AxY X= EY]. Hadwe defined x =EY as "E!x & E!y & (F)(Fx '" Fy)", [Axy x = EY] would
be ill-formed, due to the presence of the relation quantifier. Recall that we do not allow
relation quantifiers into A-expressions in order to simplify the semantics. The slight loss of
elegance which results by having to add a non-logical axiom governing x = EY is minor
compared to the complexity which would result from having to add the technical apparatus
required to interpret ),-expressions with relation quantifiers. We make an essential use of
),-expressions in which E-identity appears in Chapter VI, when we model the senses of
definite descriptions as objects which encode just the property [Ax ¢ & (Y)(4>~ ~ y = EX)].
17 In the standard second order predicate calculus, where identity is defined (x = Y = dJ
(F)(Fx '" Fy)), a restricted version of this proper axiom would be a logical theorem. If we
were given that for object terms 01 and 02'01 =02' we could show that for a formula 4> with
one free object variable v, cf>~1 ~ 4>~2. Here's how:
Since 01=02, (F)(Fol ",F0 2). In the second order predicate calculus, EVERY formula 4>
with one free object variable can be turned into a property denoting expression [.:I.v 4>], Thus
we may instantiate the universal F quantifier to get [.:tv 4>]0 , == [.:tv 4>]° 2, But by .:I.-abstraction,
[.:I.v4>]o, '" 4>~1 and [.:I.vcf>]02 '" 4>~2. So 4>~1 ~4>~2.
Note that no such proof could be carried out in the object calculus since NOT every
formula 4> with one free variable V can be turned into a property denoting expression [.:I.v 4>].
Therefore, our identity schema is necessary, since it is not derivable. Our identity schema
has even greater significance since it governs identities between relations terms as well. Most
treatments of the standard second order predicate calculus fail to discuss adding primitive
176 NOTES

identity formulas for relation terms, since this would automatically yield the intuitively false
consequence that (F")(Gn)(XI)'" (Xn)(F"x l ... Xn == GnX I ... Xn..... F n= Gn).
18 We define:

a set!/' of f-properties is expressible =dl(34)) (4) has exactly one free

P-variable and (1)(1 satisfies 4> iff d J,I(FI)E !/')).

19 Contrast Parsons [1980J, Chapter IV, Section 2; Rapaport [1976J, p. 190 T7a; Castaneda
[1974], pp. 15-21, C*.I-.7, and *C.1-7; Routley [1979], p. 263.
20 I have adapted these terms from Rapaport [1978].
21 We use abbreviations like "Notion (x,y)" to remind the reader that these formulas do

not abbreviate formulas which can appear behind A's.

22 Nor can you: (1) model existing objects as individuals, (2) model A-objects as sets of
nuclear properties, (3) model properties as extranuclear properties (where extranuclear
properties are conceived as sets of sets of nuclear properties, that is, as sets of A-objects),
(4) map down the extranuclear properties in an obvious way so that they become correlated
with nuclear, "watered down" versions, and (5) define "x encodes F' as "the nuclear, watered
down version of F is an element of x". The reason is that distinct extranuclear properties
must sometimes get mapped down to the same, nuclear, watered down version. However,
in the theory of abstract objects, if P 0/= Q, then the object which encodes just P is distinct
from the object which encodes just Q. But the above model does not reflect this fact, since
these two objects would be identified should the nuclear versions of P and Q be identical.
23 This result is a fortunate one, for otherwise the theory would be inconsistent. Roughly,
the problem would have been as follows. A-OBJECTS effectively yields a one-one function
from the power set of the set of properties to the set of abstract objects. Now if we were
able to produce a distinct property by plugging up R, for each distinct A-object, we would
have a one-one function from the set of abstract objects into a subset of the set of properties.
The composition of these two functions would have been in violation of Cantor's theorem,
since we would have had a one-one function from the power set of the set of properties into
a subset of the set of properties. So our theory rules that the properties produced by plugging
are not necessarily distinct.
24 By "standard", I mean that the verbs are not propositional attitude verbs and that,
intuitively, they do not denote higher order properties.
25 For more on this distinction, see the cited works of Meinong, Mally, Findlay, Parsons,
and Routley. Note that we differ with these authors on the property of existence. These
authors suppose that it is extranuclear. But for us, A-objects fail to exemplify this property
by definition. In Chapter III, it will turn out that they necessarily fail to exemplify this
property.
26 We could add axioms which govern nuclearity, for example, Nuclear (F) & Nuclear (G) .....

Nuclear ([AxFx & Gx]); and, Nuclear(F) ..... -Nuclear(Ux-Fx]); etc. See Parsons
[1979b], pp. 658-660.

CHAPTER II

I The material in parts B, C of the Appendix to this chapter is essential for seeing that the

axiom we are about to propose is not a logical axiom. Originally, I was misled into thinking
 NOTES 177

that this was a logical axiom by failing to remember that identity does not work in the
object language like it does in the semantics. To see that DESCRIPTIONS is not a logical
axiom, consider the Appendix (B,C) and a schematic instance:
l/!~X)~ == (3 !y)<p~ & (3y)(¢~ & l/!;J.
By the convention employed in Chapter I, Section 4, this abbreviates,
l/!~X)~ == (3y)(<p~ & (z)(<P~ ..... z= y)) & (3y)(<p~ & l/!~).
By D 4 , this abbreviates,

l/!~x).p == (3y)(<p~ & (z)(<P~ ..... (z = EY v (A !z & A!y & (F)(zF == yF))))) &
(3y)(<p~ & l/!~).
Instances of this schema should not be logically true - nothing has been said in our semantics
about the primitive relation denoted by "= e". They fail to be valid in the (..... ) direction
because of interpretations in which both "(IX)¢" denotes an existing object and" = E" denotes
an irreflexive relation. They fail to be valid in the ( <- ) direction because of interpretations
in which there are distinct objects which bear the relation denoted by "= E" to each other
but which both satisfy ¢ and l/!.
2 The results in this section were detailed principally in two early papers [1979c] and
[197ge]. I would like to thank Cynthia Freeland for her assistance in locating the relevant
passages in Plato's works.
3 See also Phaedo, 100c7-e2, lOla.

4 An orthodox theorist might suggest that Plato discovered that existential introduction

(EI) on predicate terms was a valid rule of inference. This would turn Plato into a language
theorist, whereas on our view, he was doing metaphysics.
5 Contrast Parsons [1980], Chapter VIII, Section 5; also Castaneda [1974].
6 We are justified in using UE on = I (which is a proper theorem, Chapter I, Section 4) to
get <l>p = <l>p because we can prove that <l>p has a denotation. In the Appendix to Chapter II,
we note that if we can show that some atomic formula containing <l>p is true, then <l>p must
have a denotation. Theorem 3 provides us with a true atomic formula in which <l>p occurs.
We may therefore appeal to LA4b to instantiate <l>p into universal claims. LA4b says:
(Cl)¢ ..... (l/!p ..... ¢~), where l/! is any atomic formula and r contains a description (for details,
see the Appendix to this chapter, parts B., c., and D.)
7 See Timaeus, 52c.
8 See Timaeus, 51e, 52a. Also see The Republic, 518ff.

9 See Vlastos' [1954]; and Strang [1971].

10 Besides using "is" to mean "exemplifies", we also use "is" sometimes to mean "is identical
to". And, there will be another defined use of "is", as in "x is a Form of G", "x is a possible
world", etc.
11 For those who prefer to think syntactically, let ¢ =r A!x & (F)(xF == X).' Let l/! =r vG'
Then l/!~X)~=r(lx)(A!x&(F)(xF==X))G.' We then deduced the right side of DES-
CRIPTIONS using UNIQUENESS and the fact that G satisfies X.
12 As usual, with restricted variables :

(i) (3z)¢~ abbreviates (3x)(A!x & ¢)

(ii) (z)¢~ abbreviates (x)(A!x ..... ¢).
178 NOTES

13 One suggestion for understanding the ontological status of mathematical objects is to

say explicitly which objects exist when formulating the relevant set of axioms. So, for example,
we formulate axioms for set theory as follows:
NULL (3x)(E!x & Sx & (y)(y¢x))
UNIONS: (x)(Sx ..... (3y)(E!y & Sy & (W)(WEY == (3U)(UEX & WEU))))
POWER: (x)(Sx ..... (3y)(E!y & Sy & (W)(WE Y == W S x)))
SUBSET: (F)(x)(Sx ..... (3y)(E!y & Sy & (Z)(ZEY == ZEX & Fz))).
It should be clear how to then formulate INFINITE, REGULARITY, and REPLACE-
MENT. On this kind of formulation, it is provable that there DOES NOT EXIST an object
which exemplifies being a set of all non-self-membered sets, though A-OBJECTS guarantees
that some objects encode this property.
The problem with this suggestion is that it undermines our natural understanding of
existence, namely, having a spatio-temporallocation.
14 See Meinong [1904], p. 86 (Section 4 of "Uber Gegenstandstheorie").

15 When we talk about the various senses of an ambiguous property name, we mean the
various properties it denotes. We are not referring to its "Fregean" sense.
16 Necessary beings exist in every possible world or fail to exist in every world - they do
not go in and out of existence from world to world. In the next chapter, we redefine A-objects
as objects which necessarily fail to exist.
1 7 Some philosophers may hesitate because they prefer to reserve the term "Platonic
existence" to describe properties, relations, and propositions. But we have seen that a certain
class of A-objects behave like the Forms and this is how we justify calling the kind of
existence A-objects exemplify "Platonic". Those who now hesitate probably used "Platonic"
in connection with properties, etc., in the first place because of the orthodox view that Plato's
Forms just are properties.
Those philosophers who still wish to preserve "Platonic existence" for properties, etc.,
would at least agree that on this usage, the term denotes a (higher order) property of
properties. But that would not have bearing on the important question we are now
facing - whether it's plausible to think of the negation of the first level property of existence
as some SPECIAL kind of existence.
18 Those philosophers who believe both that properties exist and that sets exist may wonder

why we can not dispense with A-objects by modelling them as sets of properties. For the
reasons why we can not do this, see the discussion at the end of Section 4, Chapter I.
19 That is, given cp~, we can produce ~(a)~¢ without derived rules as follows: By LA4,
we get (a)~cp .... ~cp~. By the theorem of propositional logic (¢ .... ~1/1) .... (1/1 .... ~cp), we get
cp~ ..... ~ (a) ~ cp.
20 We also have to modify existential introduction slightly. From ¢~, we may immediately

infer (a)cp only if'[ contains no descriptions. Otherwise, we first need to know that some
atomic formula 1/1 containing '[ is true.
21 I am indebted to Richard Grandy for noting that the following axiom had to allow for
conjunctions of atomic formulas and for suggesting examples of logical truths which would
not be derivable using it alone.
22 Recall note 20.
 NOTES 179

23 We have signalled that we are using the semantically primitive notion of identity

by switching type styles.

CHAPTER III

1 Now that we are in a modal theory, we have to face the question of whether our descriptions

will vary in denotation from world to world or be "rigid designators". On the first alternative,
they would denote at a world lo, the unique object satisfying I/> with respect to to (if there
is one). On the second alternative, they would denote at a world lb , the unique object
satisfying I/> at loo (the base world). We could have two kinds of descriptions in our
language - rigid and non-rigid descriptions. However, we shall employ just one type of
description, and suppose that all our descriptions are rigid designators.
We do this for two reasons. One is that we will not need non-rigid descriptions in any
of our applications. Instead, we shall try to show that rigid descriptions have interesting,
heretofore undiscovered, applications. Secondly, by having just rigid descriptions in the
language, we can simplify the definition of denotation" .1' Since all of the terms of the
language will be rigid designators, we need not define the denotati0f!,., 1 of term T WITH RESPECTTO
world w. Were we to allow descriptions which might change denotations from world to world,
we would have to define "d,?/ ((IX)I/>,w)". This would force us to revise the entire definition of
denotation so that it becomes a binary function.
2 The technique here is due to Saul Kripke [1963] , pp. 83-94.

3 Note that one can consistently maintain that necessarily true propositions (i.e., propositions

.0 such that for all woridslO, ex! =(. 0)= T) need not be identical. For example, the proposition
that if Carter is President then Carter is President (i.e. <;C(lJ.AI.@ (.9'2'''lt~ 1 (6einj' .9'''e~id ent,
%Mte.),.9'2'''lt~ 1 (6ei~.9''tedident, <;CMt e't))) need not be identical with the proposition that if
Nixon is President then Nixon is President (i.e. <;C(lJ.AI.@ (.9'2'''lt~ 1 (6ei:;m .9'iedi d ent, AIixon),
.9'Sf"lt~ 1 (6ei~ .9'iedident, .AIi x on))).
Given our statement of the axioms of set theory as in note 13, II, Section 3, we need not
believe that there is only one mathematical proposition.
4 In the definition which follows, and in the definition of denotation, recall that we will

often regard propositional formulas I/> as degenerate A-expressions [Av 1 .. · vnl/>] when n = O.
That is, [A 1/>] = dfl/>· So, in effect, the variables 1", C and, range over propositional formulas
in so far as they are considered as A-expressions. For example, according to clause I, [Ax Rab]
is classified as the 1st vacuous expansion of [A Rab], which according to the above convention,
is just Rab.
5 Recall that if a definite description fails to denote, the denotation failure is inherited by

all the complex terms in which the description appears. Propositional formulas are complex
zero-place terms, and hence, they fail to denote propositions if they contain non-denoting
descriptions. Note that their satisfaction conditions would still be well-defined.
6 Had we chosen to interpret our descriptions non-rigidly, this restriction would be

unnecessary. However, it is an interesting fact about the logic of rigid descriptions that the
rule of necessitation, 01, must be restricted. This prevents the following derivation of a
logical theorem which is not true in all interpretations :
(a) F(IX)GX Assumption
180 NOTES

(b) (3y)Gy L-DESCRIPTIONS ,

(c) F(zx)Gx --> (3y)Gy CP,(a)-(b)
(d) o (F(zx) Gx --> (3 y)Gy) OI,(c)
(e) OF(zx)Gx--> 0(3y)Gy. LA 7, (d)

(e) is not true in an interpretation in which:

(i) there is a unique object in Wo exemplifying G,
(ii) this object exemplifies F in every world, and
(iii) there are worlds in which there are no objects which exemplify G.
Unrestricted OI seems to be the source of the trouble.
I am indebted to Ed Gettier for pointing out this problem; suggestions from Richard
Grandy and Max Cresswell have led me to eliminate OI as a primitive rule and add the
modal closures of all the axioms except those governing descriptions. Since we are both
allowing open formulas to be assertible and using UI as a primitive rwe, I found that you
need to add the Barcan formula to successfully derive the restricted version of 01 (see the
following footnote).
7 To see that this restricted version of 01 is derivable, first suppose that we have added
the modal closures of all of the axioms. Then, the unrestricted rule of 01 could be derived
as follows: Suppose f-- 4>. We want to show that f-- 04>. If the proof of 4> is one line, then 4>
is an axiom. So f-- 0 ¢. If the proof of ¢ is more than one line, then either ¢ was derived
from 1/1 and 1/1 --> ¢ which appeared on two earlier lines or 1> = ((1.)1/1 and was derived from
an earlier line on which 1/1 appeared. If the former, then the inductive hypothesis is that
f-- 01/1 and f-- 0(1/1 --> 4» and so by appealing to LA 7, f-- 04>. If the latter, then the inductive
hypothesis is f-- 01/1, so by VI f-- ((1.)01/1. So by the Barcan formula (LA9), f- 0((1.)1/1,
i.e., f- 04>.
Sirice we have unmodalized axioms floating around (the instances of L-DESCRIPTIONS),
the base case in the above proof fails. But by adding the restriction to 01 as we have done
in the text, we ensure that the base case and the inductive hupothesis never fail. So our
restricted version of 01 is a good rule.
8 Here is the proof. Let R be anyone-place property. By A-OBJECTS, we have (3x)(A!x &

(F)(xF ==(3u)(F= [.icy Ru] & ~uF))). Call this object as and suppose ~as[.icyRas]. Then, by
definition of as, (u)([.icyRa s] = [.icyRu] -->u[.icyRa s]). So as[.icyRa s], contrary to hypothesis.
So suppose as [.icy Ra s ]. By definition of as, for some object, say a6 , [.icy Ra 6 ] = [.icy Ra s]
and ~ a6 [.icy Ra s ]. By definition of proposition identity (Section 3, Ds), Ra s = Ra 6 • But since
as encodes [.icy Ra s] and a6 does not, as 1= a6 ·
9 We can distinguish the strong possibility of Socrates' blueprint from the strong possibility
of its blueprint as follows:
z is stronglY£ possible=dfO(3x)(E!x & Blue(z,x)),
Socrates' blueprint is stronglYE possible, whereas its blueprint is not.
10 The material in this chapter was first sket~hed in my paper [1979d]. The decision to

suppose that there were objects which necessarily failed to exist was an agonizing one. It
seemed so inelegant at the time, and I tried to construct the objects needed out of the
nonexistent objects of the elementary theory. But once I realized that necessarily nonexistent
 NOTES 181

objects were simply not the kind of objects that could exist, everything seemed to feel a
little better. With hindsight, I see that it could not have been any other way.

CHAPTER IV

1 See F. P. Ramsey, [1927].

2 The material in this section was first sketched during the writing of the first draft in Fall
1979.
3 I am indebted to Blake Barley for noting this simplification.
4 I am also motivated here by the fact that my audience will not consist principally of

model-theoretic logicians or mathematicians. The section entitled "Semantics" in Chapter III

has been the most technical section so far, and I want readers who have skipped those
sections to be able to see that my claims are in fact consequences of the axioms. (As far as
metaphysical insights go, they will not have missed much by skipping that section).
5 The proof is left as an exercise.
6 See L. Wittgenstein, [1921].
7 See D. Lewis, [1968].

8 The material in this section was first developed in my [1979f], written for an independent

study on Leibniz. I would like to thank Mark Kulstad for his help in locating certain
passages in Leibniz.
9 The best attempt I know of to make this view precise in orthodox theory is Benson Mates
[1968].
1 ° Parsons was the first to attempt a precise modelling of monads in an object theoretical
framework. See his [1978], [1980]. Parsons' results are proven as metatheorems, with the
notion of possible world as primitive. Nevertheless, two of his metatheorems served as the
inspiration and prototypes for the results which follow. Castaneda claims to have suggested
similar results along these lines in [1974], p.24. The reader is encouraged to evaluate his
suggestion.
11 Contrast Parsons [1978], p. 147, R z, and [1980], Chapter VIII, Section 3, Metatheorem
1. Should counterpart theorists reject the analysis of their work presented near the end of
Section 2, I have an alternative explanation of why they have come to hold their views.
Maybe they have confused strong correlates with their monads. Theorem 8 indicates that
it is the monads, not their correlates, which are "world-bound". Such a confusion would
put counterpart theorists in good company, for Leibniz may have confused the two as well.
See the discussion at the end of this section.
12 Compare Leibniz [1686b], Section 9, and [1714], Section 56.
13 Contrast Parsons [1978], p. 147, R 3 .

14 Contrast Parsons [1978], p. 147, R 4 , and [1980], VIII, Section 3, Metatheorem 2.

15 We make no attempt to understand Leibniz's analysis of universal affirmative statements

of the form "every A is B". For Leibniz, the quantifier is virtually meaningless. This feature
of his logic is not one we wish to preserve.
16 B. Partee notes that nothing has been said to distinguish stories from essays. The intentions

of the author may be relevant here.

17 Suppose Story (zo)' (..... ) Assume zaP, for an arbitrary property P. Since zaP, Vac(P). So

for some proposition QO,P=[.l.yQ°]. So zo[AyQO]&P=[.l.yQ°]. By EI, (:3FO)(LzoFO&

P= [.l.yFO]); that is, </J~. So by A-DESCRIPTIONS, (lz')(z'F == </J)P.
(<--)Assume (lz')(z'F == </J)P. Then </Jr By reversing the reasoning, zoP.
182 NOTES

Consequently, Zo and (lz')(F)(z'F == q,) encode exactly the same properties, so they are
identical.
18 Of course, it may be a matter for literary debate as to which propositions are true
according to a given story. And the construction of principles which help us to decide the
conditions under which a given proposition is true according to a story poses an interesting
philosophical problem, one however which is of more pressing concern to a philosopher of
fiction than to a metaphysician. The sentences inscribed by the author in the manuscript
(or uttered in a storytelling) are not the only sentences which denote propositions true
according to the story. By far, the majority of propositions true according to the story are
not explicitly stated. Most are the result of an extrapolation process which facilitates
communication between the author and his audience. The principles governing the
extrapolation process are rather mysterious (see Parsons, [1980], Chapter VII). However,
we need not concern ourselves with such mysteries, since the place to begin investigation is
with the authorship relation - a relation we take as primitive. To the extent that this relation
is unclear, so, too, will our proposal be. However, it should be said that this really reflects
a genuine unclarity in our pretheoretical conceptions of the relevant stories.
19 I believe that this is an important result. It seems to me that much of the potential fiction

has for affecting us is bound up in our being able to project ourselves into unreal circumstances
which involve objects with which we are already familiar.
20The Seven Per Cent Solution, a novel by Nicolas Meyer, is supposedly about the secret
life of Sherlock Holmes. In this nove~ Holmes is a cocaine addict. The novel is meant to be
consistent with the Conan Doyle novels.
21 Contrast Parsons [1980], Chapter III, Section 2.

22 By N-CHARACTERS and A-DESCRIPTIONS.

23 L-SUB could be justified as a special instance of a more general principle governing

stories which involves the notion of relevant entailment (",,:"). When we add a proposition to
our "maximal account" of the story, we should add all the propositions relevantly entailed by
this proposition. The following general principle captures this intuition:

We can derive L-SUB from this principle if we suppose, as we surely must, that a consequence
of the correct axioms for the predicate logic of relevant entailment will be that
¢ ": [h ¢~]o, where 0 is any object term occurring in ¢.

24 D. Lewis poses the following rhetorical question in his [1978], p. 37: "Is there not some

perfectly good sense in which Holmes, like Nixon, IS [his emphasis] a real-life person of
flesh and blood?" We agree that there is. The sense of "is" in question is "encodes".
25 If we want to represent "Holmes is a famous fictional detective," we suppose that being
famous ("F") is an extranuclear property, and that this is a property Holmes exemplifies.
Consequently, we get: Fh & F-detective(h).
However, see Appendix B for a possible method of construing [).xFx & F-detective(x)]
as denoting an ABSTRACT property.

CHAPTER V

1 There may even by a way to model our metaphysical notions like Form, Monad, World,
etc., as abstract properties. Recall that we can not guarantee that they are real properties
 NOTES 183

because their definitions involve encoding formulas. So maybe they are abstract. See
Appendix B for the attempt to model them as abstract properties.
2 Frege [1892J, pp. 56-78.

3 I would like to thank Mark Aronszajn for contributing this example. It was a result of a
discussion with Mark that I discovered that the type theory could be used to model the
senses of expressions denoting higher order objects. It occurred to me that abstract properties
could be the constituents of propositions soon after Mark challenged me to take this type
of data more seriously.
4 See Russell and Whitehead, [1910]. Also Church [1940J, pp. 56-58. Our type hierarchy
shall bear no resemblance to Church [1951]. In this paper, Church builds a language which,
for any given term r of type IJ(m there will be found a term r " which denotes the sense of To
Instead of using just two base types (like i and p, as we shall use), Church introduces two
infinite lists of base types: 0 0 ,0 1,0 2"" and 1o,1 1, 12, .... The objects in the domain of type
0 0 are the truth values and the objects in 10 are individuals. Then, where IJ(n is the type of

a given domain of objects, IJ(n+ 1 is the type of the domain of objects which, intuitively, are
the "concepts" of the objects of type an' For example, the objects of type 0 1 are the concepts
of truth values (i.e., propositions); objects of type 11 are individual concepts; etc. These
concepts of type IJ(n+ 1 serve as the senses of terms of type IJ(n'
In contrast to Church's system, we shall not guarantee that for every term in our formal
language, there will be (constructible) a term which denotes its sense. We will concern
ourselves only with the senses of terms of natural language. Consequently, we will not need
an infinite hierarchy of senses for each given term. We shall get by with just two base types.
I am not sure that there is DATA which requires us to suppose that there are senses of terms
of a formal language.
S I am indebted to Barbara Partee here, whose comments on the syntax of the type theory

in [1979h] helped me to be more careful in the final formulation.

6 The procedure should be clear. Suppose (3IJ(')4> is an expression which only violates the
second restriction on propositional formulas (i.e. IJ(' appears as an initial variable somewhere
in 4». Then let t/Jl, ... ,t/Jn be the atomic (exemplification) formulas in which IJ(' is initial. If
t/Ji' 1 :s; i:s; n, is a primitive proposition term P, replace t/Ji by TrFP. If t/Ji is an atomic
exemplification formula a'r 1 ' " Tm replace t/Ji by Exa'T 1 ' " Tn' The formula (1/ which results
should be propositional and should "capture the intent" of 4>.
7 If we wanted to follow Frege a little more closely, we would define the set of senses

of type t, !I'" as follows:

!I', = dfn{o JoEd, & (w)(30 1(t'lp)(oE.ext ,<1(t) .....
0' Eext j.)) ..... (3 !o')Vlp)(OE ""(of ..... 0' E txt Ji)))}.

Thus, the objects in !I', are the abstract objects of type t which have at most one weak
correlate. This preserves Frege's intuition that senses DETERMINE at most one object. For
example, any A-object of type t which encoded an "individuating" property of type tip would
be in !I'" where
t'lp is an individuating property = df

(w)(3 0')(0 E extw(i)) ..... (3 !o')(oEext w(t))).

Should one decide that Frege's constraint on senses is essential, one would have to redefine
oenso that it mapped E'C i X .H, into !I',.
8 Note that if K' is not in the vocabulary of 0 or if 0 is an abstract individual without
184 NOTES

representational capabilities, then we may suppose that aen)K') is the null object of type t.
9 Should it become necessary, we could expand this device by allowing any complex term

of type i to serve as subscripts on sense terms. This complication need not be developed for
our purposes in Chapter VI.
10 "z'" ranges over abstract objects of type t. Consequently, the assignment to "(IX')c/>" is
the abstract object of type t which encodes just the tip-property of being the ,p.
II SO r must have the same type as ex.

12 Again, recall "z"' ranges over abstract objects of type t.

13 I believe that we can do this consistently. However, the theory could be weakened so

that A-objects of type t encode only (possibly) existing objects of type tip. One might think
that it is an encouraging sign that the semantics is all set for abstract tip-properties to have
encoding extensions that are non-empty.
14 I have not yet found an appropriate generalization of the extensional models of the
earlier theories which are described in Appendix A.
15 We sharply distinguish between those English terms which simply lack denotation from

those which denote non-existent or abstract objects (see Parsons [1979c]). If there are English
proper names which simply lack a denotation, then we need to revise our specification that
ff map all the primitive names to a denotation. We will then need to modify LA4a and
LA4b accordingly.
16 The system proposed in this Chapter was first sketched in my [1979h].

CHAPTER VI

I D. Kaplan, [1968].
2 I have borrowed this name, and a few others, from Frege's late essay [1918]. The discussion
here does not presuppose familiarity with that work, however.
3 See K. Donnellan [1974], pp. 3-32, and [1972], p. 377. Also see S. Kripke [1972], p. 302.
4 Placing such constraints adds the following complexities. As in note 7, Chapter V, Section 2,

we would first define the set of senses of type t, f:f'" as the abstract objects of type t which
have at most one weak correlate. Then we would have to define, for each object 0' in !iJ"
the set of senses of which 0 is the unique weak correlate ("f:f',(o)"). Then we would require
that the "en~, function assign to a given name K an object drawn from f:f',(ff(K)), i.e., an
object drawn from the set of senses of which ff(K) is the unique weak correlate. The assigned
object would serve as the sense of K with respect to 0'. So where ff(O")=o',ff~n)="en~,(K).
The object determines ff(K).
Although this succeeds in modelling Frege's ideas, I doubt that language works this way.
S We could even imagine a situation in which some other object besides Lauben was the

weak correlate of LaubenJohn'

6 It follows from the fact that John believes the former and not the latter that these
propositions are distinct. This DOES NOT follow from the fact that Lauben =F LaubenJohn'
Recall the note in Chapter III, Section 4 where we showed that some propositions with
distinct constituents might by identical.
7 Cases in Kripke [1972] and Donnellan [1972], [1974] would be relevant here. However,

it is slightly tricky to transpose their arguments designed to refute the RusseIIian view that
names are disguised descriptions into arguments designed to refute the Fregean view that
senses determine a unique object as the referent of the term.
S Quine [1956].
 NOTES 185

9 Note that the English "is the wife of Tully" could be represented either as "[h Wxt & (y)
(Wyt-> Y=EX)]" or as "=(IX)Wxt".
10 That is, any weak correlate of the wife of n!!!y m would be the wife of an abstract object. By

the AUXILIARY HYPOTHESIS, this never happens.

11 Mary's belief here is trivial because the wife ofDillYm represents the wife of Tully to Mary.
However, you might think that the propositional object of Mary's belief when she believes
that the wife of Tully is the wife of Tully is an A PRIORI truth. But consider the A PRIORI
truth that the wife of Tully encodes the property of being the wife of Tully. In our formal
language, we express this as:

Being an encoding formula, this sentence does not denote a proposition. We could, however,
develop a new logical function If % {l} .Pilltrg ("encoding plug"), which maps a property ~ and
an object 0 to the proposition, rff %{l} .Pilltrg(~, 0), which is such that extjrff %{l} .Pilltrg(t, 0)) =
T iff oEext.•l~). This would give us propositions for atomic encoding formulas to denote -
propositions which, if true, would be A PRIORI. Then we could represent the object of Mary's
belief (above) as an A PRIORI proposition.
We could also represent "Meinong believed that the round square is round" as a relation
between Meinong and the proposition rff%{l}.Pilltrg(lfeinfl Mund, tlied 'tound "'1t/a'te).
And we could represent "S believes that Holmes is clever" as a relation between S and the
proposition rff"¥{l} .2?iJ!irg(lfeinfl ctevet, Yf of med) or rff"¥{l} .2?iJ!irg ("f/,q/«j ({l} .2?iJ!irgj (lfein?
ct eve~, Yf ot »zeo)), tlie «jonan r!})oyt e novet 0). There are other possibilities here.
12 There could be a problem here. Is the following triad conclusive data showing that these
questions are not independent?
(i) S believes that Dostoyevsky wrote about the student who killed an old
moneylender (according to Crime and Punishment).
(ii) S does not believe Dostoyevsky wrote about the student who was arrested by
Porphyry (in Crime and Punishment\.
(iii) The student who killed an old moneylender (in Crime and Punishment) is the
student arrested by Porphyry (in Crime and Punishment).
The following representation gets the denotation of the descriptions correct, but does not
account for the (apparent?) consistency of the triad:

(i)' Bsthat- Wd(IX)Lcp(SX & (3y)(OMLy & Kxy»

(ii)' ~ Bsthat- W d(tx)Lcp(Sx & Apx)
(iii)' (tX)Lcp(SX & (3y)(OMLy & Kxy)) = (tX)Lcp(SX & Apx).
But we can not underline these descriptions to produce sense-descriptions. Should the reader
consider this to be a problem, consult Appendix B for a solution.
Also, Parsons has suggested that there is another kind of sentence which causes trouble:
"Some biblical prophets are real, some are unreal, and some I am unsure about".
To handle this sentence, let "B" denote the Bible, let "K" abbreviate the verb "knows",
and let "P" denote being a prophet. Then consider:
(3x)(Char(x,B) &2. BPX & E!x) & (3x)(Char(x,B) & 'i:.BPx & A Ix) &
(3x)(Char(x,B) & 2. BPX & - KSthat-E!x & ~KSrhar-A Ix).
186 NOTES

13 Of course, there are other DE DICTO readings:

Bjthat-W!!'j
Bjthat-!!:J!!'j"
The same goes for sentence (B) and many of the other sentences which follow. We are now
presenting the preferred readings.
14 There is typical ambiguity here. It should be unobjectionable. See Parsons [1979aJ.

15 Some philosophers may prefer to say that Einstein discovered that there were an infinite
number of simultaneity relations ~ one for each frame of reference. So we have to restate
o~r datum sentence as "Einstein discovered that there is no such thing as absolute
simultaneity".
16 "E!!" is of type «p, p)/p)/p and "E! !x" is defined as: (3y(P.P)/,,)(E!«P,P)/p)/py & (F«P'p)/P)/P)

(xF ..... Fy)). However, it is now necessary to both add "E! !" as a simple primitive predicate
of type tip and add the following axiom: E! !,/px' == (3y')(E!'/Py & (F'/P)(xF ..... Fy)). The reason
is that as "E! !x" is defined, it is not a propositional formula, so "E! !s" is not of type p and
does not denote a proposition. Consequently, our representation of the datum sentence is
ill-formed unless something is done.
17 In this axiom, "s" is a restricted variable ranging over stories.

18 Let's suppose that we have restricted our non-logical vocabulary to just the predicates

and names used in the story in question. That way, the following principle would not imply
that either the sun is shining or it is not the case that the sun is shining is true according
to mathematical stories.
19 I leave it to the reader to determine what the denotation of "117467" is.
 BIBLIOGRAPHY

Bealer, George: 1981, Quality and Concept, London: Oxford University Press.
Castaneda, Hector-Neri: 1974, 'Thinking and the Structure of the World', Philosophia 4,
3-40.
Church, Alonzo: 1940, 'A Formulation of the Simple Theory of Types', Journal of Symbolic
Logic 5, 56-68.
Church, Alonzo: 1941, Calculi of A-Conversion, Princeton: Princeton University Press.
Church, Alonzo: 1951, 'A Formulation of the Logic of Sense and Denotation', Structure,
Method, and Meaning, New York: The Liberal Arts Press.
Cresswell, Max: 1973, Logics and Languages, London: Methuen.
Donnellan, Keith: 1972, 'Proper Names and Identifying Descriptions', Synthese 40, 356-380.
Donnellan, Keith: 1974, 'Speaking of Nothing', The Philosophical Review 83, 3-32.
Field, Hartry: 1980, Science Without Numbers: A Defense of Nominalism, Princeton:
Princeton University Press.
Findlay, J. N.: 1933, Meinong's Theory of Objects and Values, London: Oxford University
Press, (1963).
Frege, Gottlob: 1892, 'On Sense and Reference', translated in Geach and Black (eds.),
Translations of Frege's Philosophical Writings, Oxford: Blackwell, (1970), 56-78.
Frege, Gottlob: 1918, The Thought: A Logical Inquiry', translated A. M. and M. Quinton,
Mind 65, (1956), 289-311.
Goodman, Nelson: 1947, 'Steps Toward a Constructive Nominalism', (with W. V. O. Quine),
Journal of Symbolic Logic 12, 105-122.
Goodman, Nelson: 1951, The Structure of Appearance, Cambridge: Harvard University
Press.
Kaplan, David: 1968, 'Quantifying In', Synthese 19, 178-214.
Kripke, Saul: 1963, 'Semantical Considerations on Modal Logic', Acta Philosophica Fennica
16,83-94.
Kripke, Saul: 1972, 'Naming and Necessity', Synthese 40, 253-354.
Lakatos, Imre: 1973, 'Science and Pseudoscience', reprinted in The Methodology of Scientific
Research Programmes, Cambridge: Cambridge University Press, (1978).
Leibniz, Gottfried Wilhelm: 1679, 'Elements of a Calculus', reprinted in G. H. R. Parkinson
(ed.), Leibniz's Logical Papers, Oxford: Clarendon Press, (1966).
Leibniz, Gottfried Wilhelm: 1686a, 'General Inquiries about the Analysis of Concepts and of
Truths', in Leibniz's Logical Papers, ibid.
Leibniz, Gottfried Wilhelm: 1686b, The Leibniz-Arnaud Correspondence, H. T. Mason (ed.),
Manchester: Manchester University Press, (1967).
Leibniz, Gottfried Wilhelm: 1686c, 'The Discourse on Metaphysics', reprinted in L. Loemaker
(ed.), Leibniz Philosophical Papers and Letters, Dordrecht: D. Reidel, Holland, (1969),
pp. 303-330.
Leibniz, Gottfried Wilhelm: 1714, The Monadology, reprinted in Leibniz Philosophical
Writings, London: 1. M. Dent, (1973), pp. 179-195.

187
188 BIBLIOGRAPHY

LemmDn, E. 1.: 1965, Beginning Logic, London: Thomas Nelson.

Lewis, David: 1968, 'Counterpart Theory and Quantified Modal Logic', Journal of
Philosophy 65, 113-126.
Lewis, David: 1978, 'Truth in Fiction', American Philosophical Quarterly 15, 37-46.
Mally, Ernst: 1912, Gegenstandstheoretische Grundlagen der Logik und Logistik, Leipzig:
Barth.
Mates, Benson: 1968, 'Leibniz on Possible Worlds', in Rootselaar and Stahl (eds.), Logic,
Methodology and Philosophy of Science, Amsterdam: North Holland.
Meinong, Alexius: 1904, 'Uber Gegenstandtheorie', translated by Lev~ Terrell, and Chisholm,
Realism and the Background of Phenomenology, Glencoe: The Free Press, (1960), pp. 76-
117.
Meinong, Alexius: 1915 Uber Moglichkeit und Wahrscheinlichkeit, Leipzig: Barth.
Mendelson, Eliot: 1964, An Introduction to Mathematical Logic, New York: D. Van Nostrand.
Montague, Richard: 1974, Formal Philosophy, New Haven: Yale University Press.
Parsons, Terence: 1974, 'Prolegomenon to Meinongian Semantics', Journal of Philosophy
71,561-580.
Parsons, Terence: 1975, 'A Meinongian Analysis of Fictional Objects', Grazer Philosophische
Studien 1, 73-86.
Parsons, Terence: 1978, 'Nuclear and Extranuclear Properties, Meinong and Leibniz', Nous
12,127-151.
Parsons, Terence: 1979a, 'Type Theory and Ordinary Language', in S. Davis and M. Mithun
(eds.), Linguistics, Philosophy, and Montague Grammar, Austin: University of Texas Press.
Parsons, Terence: 1979b, 'The Methodology of Nonexistence', Journal of Philosophy 76,
649-661.
Parsons, Terence: 1979c, 'Referring to Nonexistent Objects', Theory and Decision 11,95-110.
Parsons, Terence: 1980, Nonexistent Objects, New Haven: Yale University Press.
Prior, Arthur N.: 1956, 'Modality and Quantification in S5', Journal of Symbolic Logic 21,
60-62.
Quine, W. V. 0.: 1948, 'On What There Is', From a Logical Point of View, Cambrige: Harvard
University Press, (1953).
Quine, W. V. 0.: 1956, 'Quantifiers and Propositional Attitudes', The Ways of Paradox, New
York: Random House, (1966), pp. 183-194.
Quine, W. V. 0.: 1960, 'Variables Explained Away', Selected Logical Papers, New York:
Random House, (1966), pp. 227-235.
Ramsey, F. P.: 1927, 'Facts and Propositions', The Foundations of Mathematics, New York:
Humanities Press, (1950), pp. 138-155.
Rapaport, William: 1976, Intentionality and the Structure of Existence, Ann Arbor: University
Microfilms, ('* 77-1930).
Rapaport, William: 1978, 'Meinongian Theories and a Russellian Paradox', Nous 12,
153-180.
Rogers, Robert: 1971, Mathematical Logic and Formalized Theories, Amsterdam: North
Holland.
RoutIey, Richard: 1979, Exploring Meinong's Jungle and Beyond, Departmental Monograph
'* 3, Philosophy Department, Research School of Social Sciences, Australian National
University, Canberra.
Russell, Bertrand: 1910, Principia Mathematica (with A. N. Whitehead), Cambridge:
Cambridge University Press.
 BIBLIOGRAPHY 189

Russell, Bertrand: 1918, 'The Philosophy of Logical Atomism', reprinted in Robert Marsh
(ed.), Logic and Knowledge, New York: Macmillan, (1972), pp. 175-282.
Schonfinkel, Moses: 1924, 'On the Building Blocks of Mathematical Logic', in Heijenoort(ed.),
From Frege to Godel, Cambridge: Harvard University Press, (1967).
Strang, Colin: 1971, 'Plato and the Third Man', Plato: A Collection of Critical Essays, Vlastos
(ed.), New York: Doubleday, pp. 184-200.
Tarski, Alfred: 1931, 'The Concept of Truth in Formalized Languages', reprinted in J. H.
Woodger (trans!.), Logic, Semantics, and Metamathematics, Oxford: Clarendon Press,
pp. 152-197.
Tarski, Alfred: 1944, 'The Semantic Conception of Truth', Philosophy and Phenomenological
Research 431-575.
Vlastos, Gregory: 1954, 'The Third Man Argument', The Philosophical Review 63.
Wittgenstein, Ludwig: 1921, Tractatus Logica-Philosophicus, London: Routledge and Kegan
Pau!'
Zalta, Edward N.: 1978, 'An Unorthodox Alternative', unpublished.
- 1979a, 'Alternative Meinongian Semantics', unpublished.
- 1979b, 'An Alternative Theory of Nonexistent Objects', (with Alan McMichael), Journal
of Philosophical Logic 9, (1980) 297-313.
- 1979c, 'Plato and Predication', unpublished.
- 1979d, 'A Theory of Objects', unpublished.
- 197ge, 'A Platonic Theory of Forms and an Application to The Sophist', unpublished.
- 1979f, 'Monads and Meinong', unpublished.
- 1979g, 'A Semantics for Propositional Attitudes', unpublished.
- 1979h, 'Meinongian Type Theory', unpublished.
 INDEX

'Abstract' Barcan formula 70, 180

elementary definition 12, 18 Being 50-52
modal definition 60 Blueprints 35 (definiton)
typed definition 111 Box Introduction (Rule of Necessitation)
use of the word 173 71, 75, 123, 180
Abstraction schemata 8-14, 172
Characters of a story 92 (definition)
see also A-OBJECTS, A-EQUIVA-
Fictional 93
LENCE, RELATIONS, PROPOSI-
Native 93
TIONS
Clark's paradox 158
Abstracts
Cognitive value of a name 108-109
see A-predicates
Complete objects 87 (definition)
'According to the story.. .' 91ff, 147-148 Complex relations
'Actual' 173
of the metaphysics 31-32, 71-72, 123
Actual world
of the semantics 20-23, 61-63, 114-116
data 2
Compossibility 86 (definition)
definition 81
Concepts (Leibnizian) 87-90
semantic 61, 113
Conditionalization (primitive logical
A-DESCRIPTIONS (theorem schema) 48
notion: if-then) 7, 17
Ambiguity of 'is' 44, 47-49, 83, 90,94-95,
~(()%9& 22, 63,115
177
Conditionalization of I; and ( 25-27, 64,
A-object descriptions 48
117
A-OBJECTS (proper axiom)
Consistency of the theory 158-166
elementary version 34
Consistent objects 80 (definition)
modal version 73
Conversion
typed version 124
~(()%"f/"i.j 21, 62, 114
Appearance at worlds 85-86
i, fh-conversion of I; 25-27, 64, 117
Argument terms 111
Correlates 35 (definition)
Arrow Elimination (Modus Ponens) 30, 71,
Counterpa'rt theory 83, 181
123
Assignments to variables Dependance (of one formula on another)
discussion 23-24 30 (definition), 71
elementary definition 26 De rei de dicto
modal definition 65 data 127-145
typed definition 118 distinction 108
Authorship 91ff Derivability (see Provability)
AUXILIARY HYPOTHESIS 37-39, 76 Descriptions 40-41, 53-58, 74-75, 97-106,
Axioms 112-113, 132-135, 137-139
Logical 28-30, 68-70,121-123 DESCRIPTIONS (proper axiom) 40, 58,
Proper 32-34, 73-74, 124-125 74, 124

190
 IND£X 191

Domain of objects 20, 61, 113 Individuals 109, 126, 172

Domain of relations 20, 61, 113 Inference rules 30-31, 71, 123
Initial terms 110-111
E-IDENTITY (proper axiom) 33, 36, 56, Interpretations 20ff, 61-64ff, 113-117
73,74,124
E-Identity (see IdentitYE) A-EQUIVALENCE 28, 31 , 45, 55-58,
Embedding 86 (definition) 69-70, 78, 122, 158-160, 175
Encode (primitive metaphysical notion) A-IDENTITY 29- 30, 69-71,122
12ff A-Introduction, Elimination (rules of in-
Encoding extension 22, 63, 70,116 ference) 31, 71
Erasure of a formula 130 (definition) A-predicates 16, 18, 23-27, 60-61 , 111
Exemplification (primitive metaphysical L-DESCRIPTIONS (logical axioms) 57,
notion) 7ff 69- 70, 122
Exemplification extension 20, 61, 114 Logical axioms
Exist at (a world) 83 (definition) elementary theory 28-30
Existence (primitive theoretical relation) modal theory 68-71
12, 50-52, 63, 73, 83 typed theory 121-123
Explanation 1-4 Logical consequence 175 (definition)
Expressible relations 175 (definition) Logical fictions 172
Expressible sets of properties 176 Logical functions 20-22, 61-63, 114-116,
(definition) 174
Logical theorems 30 (definition), 71
False (under an interpretation) 27, 68 Logical truth 27 (definition), 68
Fictional relations 107, 146
First principles 1, 12-13 Mathematics 147-153
Forms 2-3,41-47,41 (definition) Mathematical myths 147-150
Fountain of youth 3, 99 Mathematical objects 147-153
Mathematical truth 147-148, 152
Generalization (see Universal Intro- Maximal objects 78 (definition)
duction) Methodology 1- 2
Gold mountain 9, 11, 13, 35, 47-50, 100 Mirroring (a world) 85 (definition)
Modal logic of encoding 70
Higher order objects 113 (definition) Modal schemata 69
Model-theoretic consequence (see logical
IDENTITY (proper axiom) 34, 36, 51, 56, consequence)
73, 74, 124 Modus ponens (see Arrow Elimination)
Identity 57 Monads 85 (definition), 84- 90
for objects 13, 33, 73
for properties 13, 29, 69, 121 Naive Object Theory 6-10
for propositions 72, 121 Names 15,17,59,157
for relations 29, 69, 121 N-CHARACTERS (proper axiom) 93,
IdentitYE 12, 32-33, 73,124, 175 95- 96, 149- 150
Identity Elimination (= E) 34 Necessary Equivalence 60 (definition), 179
Identity Introduction ( = J) 33 NECESSARY EXISTENCE (proper
Impossible objects 145 axiom) 125
Impossible relations 145 Necessary truth 2, 69, 83- 84
192 INDEX

Necessitation (primitive logical notion) 59, Proper theorems 30 (definition)

77 Properties 7, 12
%,g'<'6 63,116 Propositional attitudes 4, 107-109,
necessitation of ~ 64, 116, 118 126-145
rule of necessitation (see Box Intro- Propositional formulas 17-18, 59-61,
duction) 110-111, 183
Negation (primitive logical notion) 7 Propositions 61, 72
%,g'~ 22, 63, 116 PROPOSITIONS (logical theorem
negation of ~ 25-27, 64, 117 schema) 72, 123
NO-CODER (proper axiom) 33, 36, 73-74, Provability 30 (definition) 71
124
Notions 167-171 Reflection
Nuclear/extranuclear distinction 10-11, ~,g'ff!t'i,j 22, 62, 115
39, 176 i,j"-reflection of ~ 25-27, 65, 118
Numbers 148-151 Relations (primitive metaphysical notion)
Object (primitive metaphysical notion) 7, 7,12
12 RELATIONS (logical theorem schema)
'Object' 1, 109, 126, 172 31-32,71-73, 123, 158-160
Object calculus Representation 108-109, 117
elementary 32 Research programs 1, 5
modal 73 Rigid designation 103-106, 179
typed 123-124 Round square 3, 6, 8, 11, 13, 34, 47-49
Occurrence (of one proposition before Russellian framework 5, 9, 41, 101-102
another) 92-93
One Over the Many Principle 41-42 Satisfaction
Originate (in a story) 93 (definition) elementary definition 27
modal definition 67
Paradoxes 158-160 typed definition 120
Paradox of analysis 143 Semantic consequence (see logical
Participation 42 (definition), 41ff consequence)
Partitioning-expressions 24-26, 64-65, Senses 4,108-109,111-113,116,126,183
117-118 Sense descriptions 112-113
Platonic Being 43, 52 Sense terms 111-112, 116, 127
Platonic Existence 43, 52 Set theory 151-152, 178
Plugging Sophist 44-47
fJ!t'il!t~i 21, 62, 114 Stories 91 (definition)
11h-plugging of ~ by 0 25-27, 65, I 18 STORIES (proper theorem) 91-92, 148
Possible Objects Story Substitution (L-SUBSTITUTION)
possibly existing objects 60 (definition) 94
strongly possible objects 75 (definition) Substitutability 19
weakly possible objects 76 (definition) Substitution of identicals (see Identity Eli-
Possible worlds 61, 78-84, 79 (definition), mination)
113
Primitive notions 7, 12, 16, 32, 77 Terms 19,61, 113
Proof 30 (definition), 71 Theorems of Logic (see Logical theorems)
Proof-theoretic consequence (see Prov- Third Man Argument 43-44
ability) True Belief 130 (definition)
 INDEX 193

Truth 77 (definition) Universalization primitive logical

Truth at (a world) 81 (definition) notion: every) 7, 17
Truth under an interpretation 0/1%§1/, 21, 62, 114
elementary definition 27 l"h-universalization of ~ 25-27, 64, 178
modal definition 68
typed definition 120 Vacuous expansion
Types 109 (definition), 183 VAe, 62, 115
ith-vacuous expansion of ~ 64, 117
Uniqueness 34-35, 40 Vacuous properties 77 (defintion)
UNIQUENESS (proper theorem schema) Validity (see Logical truth)
34-35, 42, 48
Universal Elimination 31 Weak correlation 76
Universal Introduction 30, 71 Well-formed expression 19