Bare Possibilia

Author(s): Timothy Williamson

Source: Erkenntnis (1975-), Vol. 48, No. 2/3, Analytical Ontology (1998), pp. 257-273
Published by: Springer
Stable URL: http://www.jstor.org/stable/20012844 .
Accessed: 10/03/2011 20:33

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .
http://www.jstor.org/action/showPublisher?publisherCode=springer. .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

Springer is collaborating with JSTOR to digitize, preserve and extend access to Erkenntnis (1975-).

http://www.jstor.org
 TIMOTHY WILLIAMSON

BARE POSSIBILIA

0.

The theorems of the simplest and strongest sensible quantified modal logic
include the Barcan Formula and its converse. Both formulas face strong
intuitive objections. This paper develops a theory of possibilia to meet
those objections.

1.

The Barcan Formula is really a schema:

BF VxDa D Uixot

The variable x may occur free in a; read Vx and D as

'everything x [with?
out restriction] is such that' and 'it is necessary that' respectively. 0 and
3x are metalinguistic abbreviations of -O-? and -^Vx->. The modalities
will be understood here as metaphysical rather than logical or physical.
The interpretation of the quantifiers is discussed below. The consequences
of BF are more easily grasped when it is stated in the
logically equivalent
form 03*a D 3jcOo?; if there could have been something that was such
and-such, then there is something that could have been such-and-such.1
The Converse Barcan Formula is this schema:

BFC Uixa D VjcDcx

Its consequences too are more easily grasped when it is stated in the
logi?
cally equivalent form 3*0? D (>3xu\ if there is something that could have
been such-and-such, then there could have been something that was such
and-such. Together, BF and BFC imply that commutes with
'everything'
'necessarily' and 'something' with 'possibly'.
Given that there could have been a golden mountain, BF does not im?
ply that there is a golden mountain, which may nevertheless not exist. It
implies only that there is something that could have been a golden moun?
tain: a possible golden mountain in the sense that it possibly is a golden

Erkenntnis 48: 257-273, 1998.

jJL
T^ ? 1998Kluwer Academic Publishers. Printed in theNetherlands.
258 TIMOTHYWILLIAMSON

mountain, not in the sense that it is a golden mountain and possibly exists.
BF allows one to insist that absolutely all mountains are to be found in

ordinary space and time in the ordinary way.

Nevertheless, BF appears to succumb to compelling counterexamples.
Read a as 'Wittgenstein fathered jc'; although Wittgenstein died childless,
it is metaphysically possible that he fathered someone (()3xot). It follows
by BF that there is something that he could have fathered. But what is it?
On the plausible assumption that one's parentage is essential to one, no ac?
tual person could have been fathered by Wittgenstein. A non-person seems
an even less likely candidate
for something that he could have fathered;
only the most crudely of
reductionist
view persons would imply that he
could have fathered some actual set of atoms. Apparently, nothing is such
that Wittgenstein could have fathered it (-^3x()a). To vary the example,
let 'these' rigidly designate everything that there is, and read a as 'x is
not one of these'. Apparently, there could have been something that is not
one of these; indeed, there could have been more things than there actually
are. It follows by BF that there is something that could have not been one
of these. But whether it is one of these
is not itself a contingent matter;
if it could have one of these, it is not one of these. Thus there
not been
is something that is not one of these; but that is a contradiction, for by

assumption 'these' designates everything that there is.2

BFC also appears to succumb to counterexamples. Read a as 'nothing
is jc' (?'3y x = y). Apparently, the river Inn is such that if no part of
the earth's surface had ever been covered by water, nothing would have
been it, so something is such that possibly nothing is it (3x()-^3y x =

y); indeed, there could have been fewer things than there actually are. It
follows by BFC that possibly something is such that nothing is it (()3x-^3y
x = y). But that is impossible, -
for necessarily everything is something
itself (DVx3y x = y).

2.

BF and BFC are often discussed in relation to the actualist thesis that

everything exists (everything actually exists). The anti-actualist is sup?

posed to be in a much better position than the actualist to resist the putative
The former can say that there is a non-existent possible
counterexamples.
child of Wittgenstein, that the sense in which there could have been more
more exis
things than there actually are is only that there could have been
tents than there actually are, and that the sense in which the Inn could have
been nothing is only that itmight not have existed. In other words, if pos
sibilia are what could exist, then the anti-actualist can defend BF and BFC
 BARE POSSIBILIA 259

by positing non-existent possibilia and quantifying over existent and non?

existent possibilia. Given actualism, there are no non-existent possibilia to
be quantified over, and BF and BFC look much less attractive.3
Both actualism and anti-actualism are obscure doctrines, for the crucial
term 'exist'is ambiguous. It has a narrow sense, applicable only to a lim?
ited set of categories. In this sense, it is a solecism to say that events exist;

they occur. But the actualist does not want to deny that there are events
merely because they do not exist in this sense. Nor would it help to define
'exist' as 'have spatio-temporal location', for the actualist does not want
to deny that there are numbers merely because they do not have spatio
temporal location. In some sense there exist prime numbers between 60
and 80, but is that to say anything more than that there are prime numbers
between 60 and 80? Thus actualism and anti-actualism are in danger of

boiling down to the platitude that everything is and its negation. But when
actualists assume without argument that existents might not have existed,
they may be reverting to the narrower sense of 'exist'. A ban on the word
'exist' in philosophy would be salutary.4
Actualism and anti-actualism do not become any clearer if they are
redefined as the thesis that everything is actual and its negation. If 'actual'
means actually existing, we are back to 'exist'. If 'actual' means actu?
ally being, we are back to the platitudinous interpretation of actualism. If
'actual' means something else, what is it?
For these reasons, the ensuing discussion avoids reference to actualism
and anti-actualism, although some of the issues raised may be relevant to
those obscure doctrines. The putative counterexamples to BF and BFC
were stated without use of 'exist' or its affiliates, and can be evaluated
accordingly. The case against BF and BFC assumes that there is nothing
that Wittgenstein could have fathered, that there could have been more
things than there actually are, and that possibly nothing is the Inn. These
assumptions will be called into question. Before that, however, the logical
case for BF and BFC will be stated and examined.

3.

The simplest and strongest sensible quantified modal logic with identity is
LPC = S5.5 Its language is the result of adding D to a standard language for
the lower predicate calculus with identity but without individual constants
or function terms (LPC=). LPC = S5 can be axiomatized thus:

PC Every truth-functional tautology is an axiom.

 260 TIMOTHYWILLIAMSON

VI Every formula of the form Vxa D ajj/x] is an axiom, where x and y

are any variables, y is free for x in a and a [y/x] results from
replacing
each free occurrence of x in a by y.
=1 x = x is an axiom, where x is any variable.
=2 x = y D (a D ?) is an axiom, where x and y are any variables and ?
differs from a at most in having y free at some places where a has x
free.

K Every formula of the form D(a D ?) D (Da D D?) is an axiom.

T Every formula of the form Da D a is an axiom.
E Every formula of the form ()a D DOa is an axiom.
MP If a and a D ? are theorems then so is ?.
V2 If a D ? is a theorem and x is not free in a then a D Vx? is a theorem.
N If a is a theorem then so is Da.

PC and MP form the non-modal propositional basis of this axiomatization;

the addition of VI and V2 axiomatizes non-modal first-order quantified
logic, to which
=1 and =2 add the logic of identity; the addition of K, T, E
and N to the propositional basis axiomatizes the propositional modal logic
S5. No axioms or rules of inference have been added specifically to govern
the interaction of quantifiers and modal operators. Nevertheless, theorems
- - are
governing their interaction in particular, BF and BFC derivable.6
Proofs are sketched below.
LPC=S5 is sound and complete with respect to an equally simple for?
mal semantics. A model is a triple (W, D, V), where W and D are sets and
V is a function mapping each n-place predicate letter 0 in the language to
a function
V^ from W to the power set of Dn. Intuitively, W is the set of
possible worlds, D the domain of quantification, and V(0) the intension of
0. With respect to (W, D, V), the truth of a formula a at a world w e W
relative to an assignment v of members of D to variables (w \=v a) is
defined by the obvious recursion:

Y=(w) = {{d,d) : deD}.

If (j) is an n-place predicate letter and x\,...,xn are any variables,

u;K 0*1. - xn iff {v(xi),..., v(xn)) <E
V^w).

w \=v ?*a iff not w \=v a.

w \=v a D ? iff either not w \=v a or w \=v ?.

w \=v Vxa iff w \=v* a for every assignment v* of members of

D to variables differing from v at most on x.
 BARE POSSIBILIA 261

w \=v Da iff for all w* e W, w* \=v a.

A formula is valid just in case with respect to every model it is true relative
to every assignment at every world.7 Just as no axiom or rule of inference
of LPC = S5 specifically governs the interaction of quantifiers and modal

operators, so the world parameter w is inert in the semantic clause for D,

and the assignment parameter v is inert in the clause for D. The clause for
V is motivated by quantificational
simply considerations; the clause for D
ismotivated simply by modal considerations. Nevertheless, it is routine to
check that BF and BFC are valid.
The formal semantics ismade particularly simple by the use of S5 as the
underlying propositional modal logic. In S5, it is never contingent whether
something is necessary, so necessity can be treated as truth in all of a fixed
set of possible worlds; there is no need to introduce an accessibility relation
and equate necessity with truth in all accessible worlds.
Although S5 is not

beyond dispute as a propositional logic for metaphysical necessity, it has

considerable plausibility; in this paper it will simply be assumed.
The formal semantics alone does not fix the meaning of D. It does
not specify which worlds are metaphysically possible. Perhaps we cannot
specify that without using the very modalities in question. But necessity
may still be equivalent to truth in all possible worlds, even if the latter

concept is less fundamental than the former, and this equivalence may use?
fully guide our thinking. The formal semantics provides a simple, intuitive

hypothesis about which formulas are logical truths, on which LPC = S5 is

theright quantifiedmodal logic with identity.
To derive BF
in LPC=S5, it is useful to establish the lemma that
I"lpc=S5 ? D UK just in case hLPC:=:S5 O? D y. IfKpc=s5 ? D Dy,
then r-LPC=s5 ?(-Oy D -.0) by N, so hLPC=S5 O? D ODy by K; but
I~lpc=S5 ODy d y from E and T, so r-LpC=s5 ?? D y. Conversely, if
I"lpc=s5 O? D y, then hLPC=S5 DO? D Dy by N and K; but r-LPC=S5
? D DO? from E and T, so h-LPC=s5 ? D Dy. Note that the proof of the
lemma makes no use of quantificational principles. Now VxDa z> Da is an
instance of VI, so by the lemma r-LPC=s5 OVxDa D a; since x is not free
in OVjcDa, r-LPC=S5 OVxDa D Vxa by V2, so r-LPC=s5 VxDa D DVxa
by another application of the lemma. Unlike BF, BFC is derivable in
LPC=S5 without appeal to E or T. For Vjca D a is an instance of VI,
so r-Lpc=s5 DVxa D Da by N and K; since x is not free in DVxa,
I~lpc=S5 DVxa D VxDa by V2. In light of these derivations, one might
be inclined to regard it as simply a discovery that BF and BFC are truths
of logic.
262 TIMOTHYWILLIAMSON

4.

Unfortunately, the apparent counterexamples to BF and BFC remain. It is

therefore prudent to look at systems of quantified modal logic in which BF
and BFC are not theorems.
In formal semantic terms, the source of the trouble is the use in LPC=S5
of a single domain of quantification, for intuition seems to indicate that it
is contingent what 'everything' comprises. Following Kripke, we might
therefore give each world w e W its own domain D(w;). Vxa is true at w
relative to an assignment v if and only if a is true at w relative to every

assignment v* differing from v at most on x such that v*(x) e D(w). An

assignment can assign a member of any domain to a variable; we must

be able to determine whether w \=v 0x even when v(x) is not in D(w),
otherwise we could not evaluate 3x00* properly. Given two worlds w\
and u)2 and an object d in D(w\) but not in D(w2X we can now construct
countermodels to both BF and BFC. Let 0 be a one-place predicate letter.
If the extension of (p is {d} at w\ and {} in every other world, then at
W2 03x<px is true and 3x00* false relative to any assignment, so BF is
invalidated. Similarly, if the extension of at
</> each world w is D(w), then
at w\ 3xO~,0* is true (because w2 \=v ~,0* if v(x) is d) and O3x-?0x
false relative to any assignment, so BFC is invalidated.8
If domains are relativized to worlds, some axiom schema or rule of
inference of LPC=S5 must be rejected. The culprit is VI. Its instance
Vx0x Z) (py can fail at a world w relative to an assignment v that as?

signs y an object outside D(w). The closed formula VyD(Vx0x D 4>y)

is correspondingly invalid; both BF and BFC are proved in LPC=S5
by applying necessitation (N) and then universal generalization (V2) to
VI (with intermediate steps). Without VI, the axiomatization of quanti?
fied modal logic becomes much harder; the required complications are

greater than in the formal semantics, and completeness proofs are more
convoluted.9 Such complications are a warning sign of philosophical error.
From a non-technicalperspective, however, the relativization of domains
to worlds appears to yield a straightforward approach to quantified modal
logic respectful of the intuitions behind the putative counterexamples to
BF and BFC. The burden of the next section is that that appearance is

quite misleading.

5.

If the informal arguments against BF show anything, they show that it has
false instances, not just that it could have had false instances. But con
 BARE POSSIBILIA 263

sider the quantifiers in the metalanguage used to formulate the evaluation

clauses in the formal semantics. The metalanguage is in effect an applied
non-modal predicate calculus. On the relativized domains approach, the
metalinguistic statement that BF has false instances implies that something
in the domain of some world is not in the domain of the actual world.
But the latter statement is true only if the domain of 'something' in the
metalanguage is not restricted to the domain of the actual world. Thus
the restriction on quantifiers in the object-language must not be applied
to quantifiers in the metalanguage. But then the restriction looks arbitrary.
Indeed, the metalanguage can itself be taken as the object-language relative
to a metametalanguage; the quantifiers in that object-language cannot be
given their intended sense if they are subjected to that restriction. Thus the
non-modal relativized domains approach is philosophically unsatisfying;
it quantifies in a way that its own theory of quantification cannot account
for.10

Certainly, if the quantifiers in BF and BFC are restricted by an arbitrary

predicate the
</>, result is not in general valid. BF becomes Vx(0x D Da) D
DVx(0x D a); we can substitute a contradiction for a, and the result boils
down to Vx-?0x D DVx-?0x, which is uncontroversially invalid (read (?>x
as 'x smiles'). Similarly, BFC becomes DVx(0x D a) D Vx(0x D Da);
we can substitute 0x for a, which makes the antecedent trivially true, with
the result Vx(0x D D0x), which is also uncontroversially invalid. But
the issue is whether
BF and BFC hold when the quantifiers in them are
not interpreted as the result of restricting other quantifiers with broader
domains.11 The non-modal relativized domains approach gives the quanti?
fiers an irrelevant interpretation. It is mathematically perspicuous, and can
be used to establish
philosophically significant independence results to the
effect formulas are not provable on the basis of various axioms
that various
and rules of inference; but it does not settle the informal status of BF and
BFC.
An approach that is not self-defeating in the way just considered is to
do the semantics for a modal object-language in a modal metalanguage.12
In such a metalanguage, the statement that BF has false instances does not
imply that something in the domain of some world is not in the domain of
the actual world; it merely implies that there could have been
something
such that actually nothing is it. Even in the modal metalanguage,
one can
define an operator that behaves formally like a broader quantifier than
'everything', in the words 'necessarily everything actually'. Everything
is something, but given the rejection of BF it does not follow that nec?
essarily everything is actually something. However, those who take modal
operators as primitive can insist that 'necessarily is
everything actually'
 264 TIMOTHYWILLIAMSON

not a quantifier over a broader domain than that of 'everything', because

there is no broader domain; rather, it is a complex built out of 'everything'
and modal operators. The structure of the modal metalanguage bears out
this contention; by contrast, the structure of the non-modal metalanguage
undermined the non-modal relativized domains approach.
The advantage in practice of a non-modal metatheory for a modal
object-language is that it allows us to settle questions of formal validity.
In effect, they are reduced to questions of non-modal set theory that its

accepted axioms decide. For instance, we know that the non-modal rela?
tivized domains approach invalidates BF and BFC. A modal metatheory
may well be more faithful to the intended meanings of modal terms; be?
cause it is so faithful, it does not reduce questions of formal validity to
non-modal questions, often with the result that we are left none the wiser
as to their answers. Indeed, the more faithfully the metatheory interprets an

object-language modal principle, the closer the metalinguistic statement of

the object-language principle's validity is likely to be to a metalinguistic
statement of the principle itself, and therefore the less likely it is to help us
settle its validity. There may be a general dilemma here. If we are unsure
whether to accept BF and BFC, a faithful metatheory will not resolve our
doubts.
Theforegoing remarks do not yet rehabilitate BF and BFC, for they
do not undermine the intuitive force of the supposed counterexamples.
What they undermine is only the idea that the relativized domains approach
constitutes a satisfactory theoretical underpinning for the supposed coun?

terexamples. It is indeed very hard to understand in principle how BF and

BFC could fail.
How, on reflection, could BFC fail? Suppose that necessarily every?
thing is F, but not everything is necessarily F. Thus something (d, say)
could have not been F. If d had not been F, everything would still have
been F. So this situation is possible: everything is F, and it is not the case
that d is F But that situation does not seem possible, for d is a counterex?

ample to the universal generalization; not absolutely everything is F if t??

isn't.13 Perhaps d would have lacked some special ontological status if it
had not been F, but how is that relevant? The generalization does not say
that everything with some special ontological status is F; it simply says that

everything is F, without restriction. Again, if it had not been the case that
d is F, perhaps it would not have been statable that d is not F, but that too
seems irrelevant. For it is we in our actual situation who are formulating
the statement that d is not F to describe a counterfactual situation; it is no
part of our claim that the statement could have been formulated in those
circumstances.14
 BARE POSSIBILIA 265

If we cannot understand could fail, we also cannot understand

how BFC
how BF could
fail, given S5, for BF is derivable in a system that results
from adding BFC as an axiom schema to axioms and rules of quantified
modal logic that seem compelling in this context.15 We thus have reason to
suspect that BF and BFC are something like logical truths. But how should
we handle the apparent counterexamples to them?

6.

Start with BFC. The putative counterexample is DWx3y x = y & 3x0~*3y

x = is something,
y: necessarily everything yet something (e.g., the Inn)
could have been nothing. The present strategy is to defend BF and BFC by
= S5. Since x = y is easily derivable in LPC = S5,
defending LPC DVx3y
x =
VxD3j y is also derivable. Thus the strategy is committed to the truth
of VxD3y x = y. But how can the Inn be necessarily
something? Clearly,
the Inn might not have had spatiotemporal location. How could some?
thing with spatiotemporal location have been something without having
spatiotemporal location?
The analogy between modality and tense may help here. Although we
cannot argue directly from one case to the other, because it is so unclear
how far the analogy extends, the comparison often has heuristic value.
Read D as 'always' and Vx as 'everything x
[without restriction] is such
that', with 'is' tensed. In brief,tense-logical BFC says that 'always every?
thing' implies 'everything always'. Equivalently, if there is something that
was, is or will be such-and-such, then there was, is or will be something
that is (then) such-and-such. Always, everything is something. Therefore,
by BFC, everything is always something. In particular, the Inn is always
something. But it was not the case a billion years ago that the Inn was
somewhere; it will not be the case a billion years from now that the Inn is
somewhere then; how can it be something at those times? These questions
are not devastating. Imagine a time when the Inn has disappeared, but the
name 'Inn' continues to be used, to lament its passing. Everyone knows
that it does not refer to a present river; sceptics begin to suggest that stories
about a past river of that name are merely myths. Someone asks whether
the name refers to something. Surely the right answer is that it does; it
refers to a past river. Thus the Inn is still something that can be referred to,
and therefore something. Presumably it is no longer a river, for it should
no longer be counted in answer to the question 'How many rivers are there
now?' ;but it should be counted in answer to the question 'How many rivers
have there ever been?'. What kind of thing has the Inn become, if it is no
longer a river? Given that abstractness is not a temporary property, it has
 266 TIMOTHYWILLIAMSON

not become an abstract object. The best and most natural answer is just that
the Inn was once a river; it is a past river. To insist that it is somehow some?

thing more than a past river would be obscurantist, by making it a ghost of

a river (past persons are not ghosts). Its characteristic properties concern
its past; whether it continues to leave traces in the present is inessential to
its nature. In particular, it will continue to be a past river even when there
is no one left to refer to it.16More controversially, why should it not have
been a future river even before it had ever run anywhere, and before there
was anyone to refer to it?
The modal analogue of a past river is a possible river. The proposal is
that even if the Inn had never run, it would still have been a possible river.
What is necessary to it is not being in space and time but the possibility
of being in space and time; furthermore, although it is not necessarily a

river, itmay well be necessarily such that if it is in space it is a river. If the

Inn had never run, itwould have had few interesting non-modal properties,
where property is one expressible
a modal only by use of modal terms. It
would not have been a modal river ghost. And given that abstractness is
not a contingent property, it would not have been an abstract object. With?
out a theory of abstractness, the classification of merely possible rivers as
abstract objects assimilates them to paradigmatically abstract objects, such
as numbers, sets and directions, which are not contingently abstract.17
What merely possible rivers may affront is the idea that an object's

general modal properties should somehow derive from its general non
modal This is a strong form of the claim that hypothetical
properties.18
truths need categorical bases. For if the Inn had never run, from what gen?
eral non-modal properties of it could its general modal property of being a

possible river have derived? Although itwould have had the universal non
modal property of being something, and negative general properties such
as being a non-river, it would have shared all those with merely possible
mountains, which are presumably not possible rivers. But we have no good
reason to believe that an object's general modal properties are traceable
to its general non-modal properties. Although we may feel that we can
visualize its general non-modal properties in a way in which we cannot
visualize its general modal properties, reliance on this kind of primitive

imagistic thinking is no way to do philosophy.

On the present view, there are no counterexamples to BFC, only coun?

terexamples to the different claim that if there is something in space and

time that could have been such-and-such (e.g., not in space and time), then
there could have been something in space and time that was such-and
such. Similarly, there are no counterexamples to BF, only counterexamples
to the different claim that if there could have been something in space
 BARE POSSIBILIA 267

and time that was such-and-such (e.g., fathered by Wittgenstein), then

there is something in space and time that could have been such-and-such.
Since Wittgenstein could have fathered someone, there is something that
Wittgenstein could have fathered; it is amerely possible person. More gen?
erally, there could not have been more or fewer things than there actually
19
are.

Merely possible members of a kind are needed to make sense of some

counting questions. Suppose, for simplicity, that a suit consists

of a jacket
and a pair of trousers, that necessarily they make a suit just in case they are
originally hung together, and that at most one possible suit can be made of
a given jacket and pair of trousers. Consider two jackets Jl and J2 and two

pairs of trousers Tl and T2, which actually constitute two suits, Jl +T1
and J2+ T2. If Jl had originally been hung with T2, there would have been
a differentsuit Jl + T2; as things actually are, Jl + T2 is not a suit; indeed,
Jl + T2 does not actually seem to be anywhere. Intuitively, the question
'How many possible suits could be made from Jl, J2, Tl and T2?' has a
reading on which the answer is four, even though it is impossible for more
than two suits to be made from the set.
In that example, we can refer to specific merely possible suits, using
counterfactuals in definite descriptions to fix the reference of 'Jl + T2' and
'J2 + Tl'.20 Such descriptions must be handled with
care; for example,
'the possible child of Wittgenstein' is improper,
because many possible
people could have been fathered by Wittgenstein.21 No such difficulty
faces treatment of a putative counterexample to BFC, because the object
at issue (e.g., the Inn) is actually in space and time, which allows us to
establish reference to it in the ordinary way; we use that referential link to
discuss the object with respect to situations in which it is not in space and
time.22 We can then reflect that people in those situations might be unable
to establish reference to the object, if no uniquely identifying description
were available to them, but could still express general propositions (e.g.,
'There is a merely possible river') made true by facts about the object. We
can further reflect that we may be in the same predicament with respect to
other possible members of the same kind, such as merely possible rivers,
to which we cannot establish reference but facts about which nevertheless
make true general propositions that we can express - such as BF.
Suppose, for the sake of argument, that referring to something pre?
cludes its being a merely possible river. It does not follow that no merely
possible river is capable of being referred to by us. For perhaps, if it had
been a river, we could have referred to it. Nevertheless, we should not
assume that every object is a possible object of reference by us, or anyone
else, for what motivates the assumption but ill-concealed verificationism?
268 TIMOTHYWILLIAMSON

Perhaps the world contains microscopic particles incapable of being sin?

gled out by any thinking being. Although every possible bearer of a name
is an object, something may belong to the same general kind as a possible
bearer of a name, and be an object too, without itself being a possible
bearer of a name.
The difficulty of singling out mere possibilia means that they are not

plausible objects of most intentional states. Seeking, for example, should

not be construed as a relation to possibilia. Someone may seek the answer
to all his problems, even though there is no answer to all his problems;
that is not to say that there is a possible answer to all his [actual] prob?
lems which he seeks. Perhaps there is no even possible answer to all his

problems; perhaps there are many possible answers, none of which is the
possible answer he seeks.23 Theories of modality should not be expected
to resolve issues about intentionality.

7.

How are possible rivers individuated? In LPC = S5, both the necessity of
x = =
identity (LI, y D Dx y) and the necessity of distinctness (LNI,
x jk y D Dx ^ y) are provable.24 Objects are either necessarily identical
or necessarily distinct. It is then easy to derive this schema in LPC = S5:
=
p= Op* D (x = y 0(px & py & x = y))
For x = =
y gives Dx y by LI and thence D(px z> py), from which
the right-hand side can be derived from Opx by ordinary modal reasoning.
Conversely, the right-hand side gives 0* = y, which gives x = y by LNI.
Now read px as 'x is a river'. Informally, p? says that a possible river
is identical with something just in case they could be the same river. The
significance of p? is that on the right-hand side the notion of identity is in
effect restricted to the identity of rivers, rather than possible rivers. 'Same
river' explains 'same possible river'.
The point may leave some doubts. Itmay still be asked: what makes the
river in one world the same thing as the genuinely flowing
merely possible
river in another world? But this doubt is more obscure than that which it
calls into question. Its focus is Op* rather than x = y, possibilities for

objects rather than identities. Why should one find 'x could be a river'
Doubts about modality often seem to be epistemological in
problematic?
can we know whether that is not so could have
origin: how something
been so? For non-verificationists, however, the limits of knowledge about

modality are not the limits of modality itself. Of course, it would be self
to make modal claims and then admit that one cannot discover
defeating
 BARE POSSIBILIA 269

whether they are true. But the central modal claims of this paper are
broadly logical claims. Their epistemology will be a special case of the

epistemology of logic. Until we have a better understanding of the epis?

temology of logic, we cannot say whether the epistemology of quantified
modal logic is fundamentally more problematic than the epistemology of
other branches of logic.
= S5
As for the metaphysics of quantified modal logic, LPC presents a
startlingly clear picture. LI and LNI fit well with BF and BFC; what ob?
jects there are and their identity, distinctness and number are all necessary
matters. Indeed, in LPC=S5 any proposition expressible in terms solely
of identity, negation, conjunction, the universal quantifier and necessity
is either necessarily true or necessarily false.25 Contingency has its place
within a necessary framework of objects.26

NOTES

1
The original Barcan formula inMarcus [= Barcan] 1946 is in existential form, with strict
rather than material implication. The difference is unimportant in this context.
2
For objections to BF and close related principles in tense logic, and intriguing refer?
ences to relevant earlier discussions, see Prior (1957: 26-36, 112-116 and 1967: 138-151).
Parsons (1995) is a more recent discussion of putative counterexamples to BF.
3
Marcus (1985/86) defends BF on actualist without the intuitions
grounds, explaining
underlying the apparent counterexamples. Linsky and Zalta (1994, 1996) aim to recon?
cile BF with actualism by broadening the domain of existents; their view has significant
similarities with that to be defended here. They argue against several versions of actualist
semantics that invalidate BF.
4 See furtherWilliamson
(1987/88, 1990b).
5
LPC=S5 and its semantics are slightly adapted from Hughes and Cresswell (1996: 243
244 and 312-313).
6 See Prior
(1956), Kripke (1963) and Hughes and Cresswell (1996: 245-247). Stalnaker
(1995) usesa similar of combining of non-modal first-order
methodology principles quan?
tified logic with principles of propositional modal logic. He constructs a system in which
not even a weakened version of BFC is provable; however, he bases quantified logic on a
predicate abstraction operator governed by principles too weak to ensure that abstraction
has its normal effect in modal contexts (Williamson, 1996b).
7
IfW or D is empty, every formula
is vacuously true relativeto every assignment at every
world in the model, for there are no worlds or assignments.
8
See Kripke (1963). In semantic frameworks in which Da is true at w just in case a is true
at every world accessible from w, BF corresponds to the constraint that D(w*) ? D(w;)
whenever w* is accessible from w and BFC to the constraint that D(w) ? D(w*) when?
ever w* is accessible from w. These constraints are equivalent if accessibility is symmetric.
In effect, the present semantics for S5 is the special case in which holds
accessibility
universally, and so is trivially symmetric. The B axiom schema aDDOa to
corresponds
the symmetry of the accessibility relation. As one might therefore BF is derivable
expect,
in the weak subsystem LPC=B of LPC=S5, axiomatized B for T and E
by substituting
 270 TIMOTHYWILLIAMSON

in the axiomatization of LPC=S5 in the text (in effect E. J. Lemmon this out in
pointed
1965; see Prior (1967: 146)). This is philosophically significant, because Hugh Chandler
(1976) and Nathan Salmon (1989) have given purported counterexamples to the 4 schema
Da D DDa, which is derivable in LPC=S5 but not in LPC=B. Salmon suggests that
instances of B are not logically true, but not that they are subject to counterexamples. He
allows that instances of T are logically true; the reason for this differential treatment of T
and B seems to be the greater obviousness of T, but obviousness is hardly a condition of

logical truth. Ofcourse, the simplicity of the semantics in the text depends on the use of
S5, so that mention of accessibility is unnecessary. For an argument the purported
against
counterexamples to 4 seeWilliamson (1990a: 126-143); a reply to Salmon (1993) must
await another occasion.
9
See Garson (1984). Cresswell (1991) uses the complications as an argument for BF.
10
Although the argument resembles arguments that actualism, the Kripke semantics and
the denial of BF form an inconsistent set, it does not use the informal notion of existence.
11
Even on accounts of the set-theoretic paradoxes on which every domain can be extended
to a wider domain, the narrower domain is not primarily given as a restriction of the
broader one; rather, the broader domain is given as an extension of the narrower one. In

any case, the intuitive objections to BF and BFC do not seem to have to do with
anything
the paradoxes.
12See
e.g., Fine (1977) and Peacocke (1979).
13
Objections toBFC are in effect objections to VyD(Vjca D cc[y/x]), the result of neces?
sitating and then universally generalizing VI, for BFC is derivable from the special case

VjcD(Vjca D a) of that formula in a system of axioms and rules that are comparatively
uncontentious in this context. In particular, consider the Kripke-inspired system LPCK+K,
axiomatized by the axiom sch?mas PC, VD(Vjc(a D ?) D (Vxa D Vjc/?) for any variable
= Vjca x not free in
*)> VQ (a for any variable a), V1K (VyV?(Vjca D ct[y/x]), where
jc, y and z are any variables, y is free for x in a and a[jc/y] results from replacing each
free occurrence of jc in a by y) and K, and the rules of inference are MP, N and UG

(if a is a theorem, so is Vjca for any variable jc). See Hughes and Cresswell (1996: 304

309) for more on such systems. By UG and then VD (twice) applied to an instance of K,

l-LPCK+K VjcD(Vjcq? D a) D (VjcDVjca D VjcDa); by VQ, l~LPCK+K VjcO(Vjca D a) D

BFC.
14 a value
Prior's system Q
postulates of unstatability between truth and falsity, distin?

guishing D
(always true) from -"O"*1 (never false). See Prior (1957: 41-54). On Prior's

view, D(0a D (pa) is false unless a exists at every world, because (pa D (pa is statable
a exists. This seems to confuse the context of utterance
only at worlds at which with the
context of evaluation. If a exists at this world and (p is an ordinary predicate, then at this
world (pa and (pa D (pa express conditions that each world may or may not meet. A world
w meets the condition a D ? expresses at this world just in case if w meets the condition
a expresses at this world, w meets the condition ? expresses at this world. Thus every
world meets the condition which (pa D (pa expresses at this world, so this world meets the
condition which U((pa D (pa) expresses at this world.
15BF is derivable in
LPCK+S5+BFC, the result of adding T, E and BFC as axiom sch?mas
to LPCK+S5 (see note 13). Since y and z can be the same variable as jc in V1K, by VQ
-LPCK+S5+BFC Vx(VxDa D Da), so l"LPCK+S5+BFCDVx(VxDa D Da) by N. By
BFC, HLpcK+S5+BFC VxD(VxDa D Da). By N and K hLpCK+S5+BFC D(VxDa D
Da) D (OVjcDa D ODa), so HLPCk+S5+BFC D(VjcDa D Da) d (OVjcDa D a) by
T and E. By VD and UG, hLPCK+S5+BFC VjcD(VjcDa D Da) D Vx?>VjcDa D a). We
 BARE POSSIBILIA 271

have already established the antecedent, so Vx (OVxDa D a). By Vo and

l~LPCK+S5+BFC
VQ, I-LPCK+S5+BFCOVxDa D Vxa. Then N andK give l"LPCK+S5+BFC DOVxDa D
DVxa and T and E give hLPCK+S5+BFC ^x^a ^ ^?Vxcx'QED. As in the proof of BF in
LPC = S5, B can replace T and E.
16
In a similar case, Prior (1957: 32-33) is driven to the desperate suggestion that 'Bu?
is no a proper name because Bucephalus no longer exists. He proposes
cephalus' longer
instead to analyse it as a definite description, 'the Bucephalizer'. Developments in the
of language over the last forty years make this proposal even less plausible than
philosophy
it was when made. Moreover, even if accepted, it would not obviously achieve anything. It
enables us to deny that 'Bucephalus' now refers to Bucephalus; the sentence 'Bucephalus
ran' is analysed as something like 'PAST ((THE x: x Bucephalizes) x runs)'. But any

problem in assigning Bucephalus to the name in 'PAST (Bucephalus runs)' is equally a

problem in assigning it to the free variable in 'jc Bucephalizes' and 'jc runs'. If either

assignment is impossible, we cannot use the corresponding sentence to say something

true; our ability to do that is what was to be explained. Thus nothing is lost in treating

'Bucephalus' as a name. But 'Bucephalus' refers now to Bucephalus. Note that a present
term t can refer to a past object o even if t was not used to refer to o when o was extant.
17
Linsky and Zalta, whose approach is in some ways very similar to the present one

(although they present it within the framework of actualism and take BF as primitive)
waver on this point. They hold, like the present approach, that there are contingently

objects. At Linsky and Zalta (1994: 446) they define 'concrete' as 'spa
non-spatiotemporal
tiotemporal' and 'abstract' as 'not concrete'; thus some abstract objects could have been
concrete. At Linsky and Zalta (1996: 293) they say 'abstract objects are not concrete at any
world'. The latter approach is preferable. The inconsistency may be largely terminological.
18
'General' is intended to exclude non-modal properties such as being the Inn, which is
not shared by anything that differs in modal properties from the Inn.
19
See Ramsey (1927) for an early argument to this effect.
20
In arguing against mere possibilia, Ruth Barcan Marcus assumes that we can refer only
towhat we have empirically encountered (1993: 207).
21
For a careful treatment of definite descriptions in relation to possibilia, although based
on an actualist reading of the quantifiers, see Salmon (1987).
22
Throughout the ensuing discussion, read 'reference' as 'singular reference'.
23
Linsky and Zalta, in their 1996 response to Tomberlin (1996), propose an account on
which 'the answer to all his problems' is analysed as 'the unique (abstract) object which,
according to his conception, answers all his problems'. Their account has difficulties with
sentences suchas 'He looked for the largest apple, found it and ate it'; although they also
allow the more usual reading of the definite description, the sentence does not seem to have

any reading on which it implies that he ate an abstract object.

24 See
Hughes and Cresswell (1996: 313-314). Williamson (1990b, 1996a) defends LI
and LNI.
25
Proof: One shows that I~lpc=s5 a 3 ^a induction on the complexity of a. LI and
by
LNI give the basis of the induction. The induction
steps for & and D are routine exercises in

propositional modal logic. For ->, note that if I~lpc=S5 ? D Da then I~lpc=S5 0~,Da D
so I~lpc=S5 ~>0i 3 D--a by E and T. For the induction
D-?a, step for V, note that if
? 3 Da then I~lpc=S5 ^xa 3 VxDa, so I~lpc=S5 ^xa 3 DVxa
l~LPC=S5 by BF.
26Thanks toWinfried L?ffler for his
helpful reply, other participants in the Innsbruck
Symposium on Analytical and a meeting of the Portuguese
Ontology Philosophical Society
 272 TIMOTHYWILLIAMSON

in Lisbon for discussion, and Rodrigo Bacellar, Paolo Crivelli, Alex Orenstein and Philip
Percival for detailed comments.

REFERENCES

Chandler, H.: 1976, 'Plantinga and the Contingently Possible', Analysis 36, 106-109.
Cresswell, M.: 1991, 'In Defence of the Barcan Formula', Logique et Analyse 135-136,
271-282.

Fine, K.: 1977, 'Postscript', in Prior 1977.

Garson, J.: 1984, 'Quantification inModal

Logic', in Gabbay, D. and Guenthner, F. (eds.),
Handbook of Philosophical Logic, Vol. II, Reidel, Dordrecht.

Hughes, G. and Cresswell, M.: 1996, A New Introduction to Modal Logic, Routledge,
London.
S.: 1963, 'Semantical Considerations on Modal Logic', Acta Philosophica Fennica
Kripke,
16, 83-94.

Linsky, B. and Zalta, E.: 1994, 'In Defense of the Simplest Quantified Modal Logic', in
James Tomberlin (ed.), Philosophical Perspectives 8: Logic and Language, Ridgeview,
Atascadero.

Linsky, B. and Zalta, E.: 1996, 'In Defense of the Contingently Nonconcrete', Philosophi?
cal Studies 84: 283-294.

Marcus, R. Barcan: 1946, 'A Functional Calculus of First Order Based on Strict Implica?
tion', The Journal of Symbolic Logic 11, 1-16.

Marcus, R. Barcan: 1985/86, 'Possibilia and Possible Worlds', Grazer Philosophische

Studien 25-26, 107-133; reprinted inMarcus 1993.

Marcus, R. Barcan: 1993, Modalities: Philosophical Essays, Oxford University Press,

Oxford.

Parsons, T.: 1995, 'Ruth Barcan Marcus and the Barcan Formula', in Sinnott-Armstrong,
Raffman and Asher 1995.

Peacocke, C: 1978, 'Necessity and Truth Theories', Journal of Philosophical Logic 7,

473-500.

Prior, A.: 1956, 'Modality and Quantification in S5', The Journal of Symbolic Logic 21,
60-62.

Prior, A.: 1957, Time and Modality, Clarendon Press, Oxford.

Prior, A.: 1967, Past, Present and Future, Oxford University Press, Oxford.

Prior, A.: 1977, Worlds, Times and Selves, University of Massachusetts Press, Amherst.

Ramsey, F.: 1927, 'Facts and Propositions', Aristotelian Society 7(Suppl.), 153-170.

Salmon, N.: 1987, 'Existence', in James Tomberlin (ed.), Philosophical Perspectives 1:

Metaphysics, Ridgeview, Atascadero.

Salmon, N.: 1989, 'The Logic of What Might Have Been', The Philosophical Review 98,
3-34.

Salmon, N.: 1993, 'This Side of Paradox', Philosophical Topics 21, 187-197.

Sinnott-Armstrong, W, Raffman, D., and Asher, N. (eds.): 1995, Modality, Morality

and Belief: Essays in Honor of Ruth Barcan Marcus, Cambridge University Press,

Cambridge.
Stalnaker, R.: 1995, 'The Interaction of Modality with Quantification and Identity', in

Sinnott-Armstrong, Raffman and Asher 1995.

Tomberlin, J.: 1996, 'Actualism or Possibilism?', Philosophical Studies 84, 263-281.

 BARE POSSIBILIA 273

Williamson, T.: 1987/88, 'Equivocation and Existence', Proceedings of the Aristotelian

Society 88, 109-127.

Williamson, T.: 1990a, Identity and Discrimination, Blackwell, Oxford.

Williamson, T.: 1990b, 'Necessary Identity and Necessary Existence', in Rudolf Haller
- a Re-Evaluation:
and Johannes Brandi (eds.), Wittgenstein Towards Proceedings of the
14th International Wittgenstein-Symposium, Vol. I, Holder-Pichler-Tempsky, Vienna.

Williamson, T.: 1996a, 'The Necessity and Determinacy of Distinctness', in Sabina Lovi
bond and Stephen Williams (eds.), Essays for David Wiggins: Identity, Truth and Value,

Blackwell, Oxford.

Williamson, T.: 1996b, 'Review of Sinnott-Armstrong, Raffman and Asher 1995', Philos?

ophy 71, 167-172.

University of Edinburgh
David Hume Tower

George Square
Edinburgh EH8 9JX
UK