O W F F

Greg Restall
Philosophy Department
The University of Melbourne
[email protected]
1.01
February 27, 2004
Abstract: Stephen Neale takes theories of facts, truthmakers, and non-extensional connectives to be threatened by triviality in the face of powerful slingshot arguments. In this
paper I rehearse the most powerful of these arguments, and then show that friends of facts
have resources sufficient to not only resist slingshot arguments but also to be untroubled by
them. If a fact theory is provided with a model, then the fact theorist can be sure that this
theory is secure from triviality arguments.

Stephen Neale presents, in Facing Facts (Oxford: Clarendon Press, 2001), one convenient package containing his reasoned complaints against theories of facts and
non-extensional connectives. The slingshot is a powerful argument (or better, it is a
powerful family of arguments) which constrains theories of facts, propositions and
non-extensional connectives by showing that some of these theories are rendered
trivial. This book is the best place to find the state of the art on the slingshot
and its implications for logic, language and metaphysics. It provides a useful starting point for anyone who has wondered what all of the fuss about the slingshot
amounts to. Neale shows that the fuss does amount to something, and that theories of facts must face facts and present an adequate response to the slingshot.
However, Neales evaluation of the state of play for theories of facts is too pessimistic. As the book draws near to a close, Neale writes:
As I have stressed, Russells Theory of Facts, according to which facts
have properties as components, is safe. It is certainly tempting to draw
the moral that if one wants non-collapsing facts one needs properties
as components of facts. I have not attempted to prove this here, but I
suspect it will be proved in due course. (page 210)
Neale concludes that while theories which take facts to be structured entities are
safe from slingshot arguments, and he suspects that this is the only kind of fact
theory safe from slingshot-style collapse. If this were the case, then theories such
as situation theories or accounts of truthmakers may well be threatened. However,
Neales suspicion is ill-founded, as I shall soon show. Not only do Russellian theories of facts survive the slingshot unscathed, but so can theories of facts which take
them to be unstructured entities. Furthermore, the way that this may be not only
argued for, but proved can provide a new weapon in the armoury of the theorist
investigating fact theories.
Many thanks to Allen Hazen, Graham Priest and Barry Taylor for helpful comments and encouragement, and to anonymous referees of this journal for useful critical discussion.

http://consequently.org/publications/

This criticism of Neales understanding of the terrain does not mean that the
book is not worth the time and effort required for a close reading. There is a great
deal of good sense between its covers. The historical discussion of slingshot arguments is measured and accurate, and the chapter on descriptions is lucid, thorough
and convincing, as one would expect from Neale. I recommend this chapter especially to anyone who wants a cogent argument to the conclusion that descriptions
can be treated fruitfully as quantifiers rather than as referring expressions.
* * *
The core of the book is to be found in the formal presentation of various slingshot
arguments. The highlight is a chapter on Gdels slingshot, which, after years of development and simplification by Neale is a finely honed instrument. I will present
it here, with a further simplification, to present the proof in one paragraph rather
than over a couple of pages as Neale does. The core idea will be that, allowing
some seemingly-innocuous inference principles involving descriptions, we will be
able to make seemingly illicit substitutions inside seemingly non-extensional contexts, such as claims about the identity of facts or modal statements.
Neales version of Gdels proof requires two different kinds of substitution,
one inside a term context, and another inside a sentential context. A term context
( ) (that is, a context ( ) such that (t) is a well-formed formula if t is a term)
is said to be +- if and only if it allows the following substitutivity principles
for definite descriptions.
x = x

(x)

(x)

x = a

(x)

(a)

x = a

(a)

(x)

The context ( ) is - if and only if it is not +-. The other substitution

required is in sentential or formula position: A formula context C[ ] (that is, if
is a formula, so is C[]) is +- if and only if the inference



C a = x x = a (x)
====================
C[(a)]
is allowed in both directions. The context C[ ] will be said to be - when it
is not +-. This inference will feature repeatedly in what follows, so it will pay
to introduce a shorthand for the kind of substitution used. Given the formula (x)


with x free, and a name a, let (a) be shorthand for the term x x = a (x) .
Then the inference is

C a = (a)]
==========
C[(a)]
Note that the empty context is +- on any reasonable theory of descriptions. If (a) is true, then a = (a) is also true, for a is indeed the unique object
which is identical to a and is such that (a) is true. The converse inference is also
clearly valid.

Greg Restall, [email protected]

1.01

February 27, 2004

http://consequently.org/publications/

Given these two inferences, we can present a powerful slingshot argument.

[Fa
[(a = Fa)

+-

Fa

Rab

a = Fa

a = Rab

[(a = Rab)
[Rab
[(b = Rab)

Fa = Rab

+-

+-
+-

Gb

Rab

b = Gb

b = Rab

Gb = Rab

[(b = Gb)
[Gb

+-

+-

This shows that if the connective [ is +- and if the contexts [(a = [ ]) and
[(b = [ ]) are +-, then Fa, Gb, Rab, [Fa ` [Gb. Since we can take Rab to be
each of a = b and a 6= b in turn, we have , a = b ` and , a 6= b ` , and
it follows that ` . It follows that the assumption of Rab is unnecessary, and
we have Fa, Gb, [Fa ` [Gb, a terrible collapse of the context [. If [ is to be read
the fact that Fa = the fact that then if Fa and Gb are true, the premises of our
reasoning all seem true (as the fact that Fa = the fact that Fa), but the conclusion
would be a terrible consequence for a putative theory of facts: The fact that Fa
should not be the fact that Gb simply because they both happen to be true.
* * *
What are we to make of this? Its clearly not a happy result for the fact-theorist,
as the troublesome inferences seem plausible at first blush. Neale does a good
job of showing where fact-theorists have been committed to them. However, the
arguments do not mean that we must abandon all talk of facts. It is open for the
fact-theorist to reject the collapsing argument. Neale recognises this.
There is no knock-down argument against facts in this; but it is now
abundantly clear that unless a theory of facts is presented with an accompanying theory of descriptions and an accompanying logic of [fact identity conditions], there is every reason to treat it with caution. The task for the
friend of facts is to put together a theory according to which facts are
not so fine-grained that they are unhelpfully individuated in terms of
true sentences, and not so coarse-grained that they collapse into one.
(page 223, emphasis mine)
This diagnosis of the situation overstates the case, and it will be illuminating to
consider why. It is quite possible for a friend of facts to present a theory giving no
regard to a theory of descriptions. Nonetheless, they may still be sure that their
theory is consistent and resistant to collapse. This is because the fact-theorist may
present and defend a theory of facts by providing a model of that theory. Models
have many virtues. They provide consistency proofs, and they provide a systematisation of logical consequence. Neale does not consider this route to a theory of
facts at all, and his account suffers as a result of it.
Let me illustrate this point by giving a simple back-of-the-envelope model for
a theory of facts. The result will be a demonstrably non-collapsing theory of facts,
which does not take facts to be individuated by constituent properties but allows
Greg Restall, [email protected]

1.01

February 27, 2004

http://consequently.org/publications/

them to be totally unstructured entities, and for which the facts corresponding
to the sentences and differ if it is possible for and to differ in truth value.
Take a formal language appropriate for quantified modal logic, with some stock
of primitive predicates, terms (constants, function symbols and variables) and with
the connectives (material implication), (negation) and  (necessitation), and
the universal quantifier x as primitive (for each variable x). We leave the other
connectives of the language to be defined in the usual fashion. We extend this
language with a new connective . to be read as is a fact that. This connective
is a hybrid between a true connective and a predicate, because it takes a term to its
left and a formula to its right to make a new formula. The formula a . is read
a is a fact that . We will express our theory of facts in this language.
To interpret this language we will use models for the constant domain quantified modal logic 5. I choose this model theory not because I think that constant
domain quantified 5 gets things right, but because it is simple, and because it has
been taken seriously by others as providing the logic of necessity and quantification. For our purposes we need treat the model theory merely as an uninterpreted
algebra. The consistency proof will stand even if we totally reject the philosophical
significance of possible worlds models as a semantics. As a purely algebraic construction with no interpretational significance whatsoever, a possible worlds model will
still act as a guarantee that the resulting theory is safe from collapse and from
slingshot argument. (Nonetheless, any interpretation we might find for the model
theory has application as an interpretation of the language we use it to model.)
A model for the language we have chosen involves a nonempty set W of worlds
and a nonempty domain D of objects. Each formula will be true at some set of
worlds and false at the complement of that set. We aim to have different facts for
formulas which differ in truth value at some worlds. That is, if and are not
true in exactly the same worlds, then the fact that should differ from the fact
that . One easy way to manage this is for there to be a fact for every set of worlds,
so we will do this. We require that the domain D includes the set P(W) of all sets
of worlds. It may well include other objects as well, but we do not require this.
Each n-place predicate F is interpreted as function [[F]] from W to subsets of
Dn . Given an assignment of values to the variables (a function from the set V
of variables to the domain D, so (x) D is the value of the variable x on the
assignment ), we assign the denotation d,w (t) of each term t in each world w in
the usual recursive fashion. This allows for terms to vary in denotation from world
to world, but not to have no denotation in a world, as is usual in constant domain
modal logics. An x-variant of an assignment is another assignment of variables
0 which assigns the same values to every variable except possibly x. [x d] is
the x-variant of where the variable x is now assigned the value d D.
Given a recursive assignment of denotations for terms, we can then assign truth
conditions to all of the formulas of the language, relative to assignments of values
to variables and to worlds, as follows. (, w is to be read as relative to
assignment and at world w, is true).
(tc0) , w F(t1 , . . . , tn ) iff hd,w (t1 ), . . . , d,w (tn )i [[F]](w).
(tc) , w iff , w 6 or , w .
(tc) , w iff , w 6 .
(tc) , w x iff 0 , w for all x-variants 0 of .
Greg Restall, [email protected]

1.01

February 27, 2004

http://consequently.org/publications/

(tc) , w  iff , v for all v W.

(tc.) , w t . iff d,w (t) = {w : , w } and , w .
The only clause which is at all innovative is (tc.). The formula t . is true at a
world w if and only if the term t denotes the set of worlds at which is true, and
w is one of those worlds. So t . is true at w only when is true at w.
Our formal theory of facts is the collection of all formulas true at every world
in every model. This theory includes some oft-claimed truisms about facts. Here
are two examples:
(x)(x . )
is true if and only if there is a fact that .
(a . ) (a . ) ( )
a fact can be a fact that and a fact that only if and necessarily stand or
fall together. Many other such claims may be verified by reasoning about all models
in the usual fashion. We get a simple theory, extending the familiar modal logic
of constant domain quantified 5 with a new operator ..
This theory of facts may be verified to be non-collapsing by providing a
simple model. Take a model featuring two worlds w1 and w2 , with a statement Fa
true at both w1 and w2 but Gb true only at w1 . Then w1 Fa and w1 Gb but
, w1 t . Fa only when d,w1 (t) = {w1 , w2 } (t . Fa is true only when t denotes
the set of all worlds in the model, as Fa is true everywhere) and , w1 t . Gb
only when d,w1 (t) = {w1 } (t . Gb is true only when t denotes the set {w1 }), so
this model is a guard against collapse of the theory of facts.
So, commitment to any of the inferences validated by these models will never
result in a trivial theory of facts. Of this we can be completely sure. If any collapse
threatens, it must come from outside this theory. This theory demonstrably does
not collapse, while at the same time, this theory demonstrably does not take facts
to be composed of properties or any other such thing. The theory is silent about
the composition of facts. The models of the theory take them to be sets of worlds,
but that is not a part of the theory. The model theory is a technique to provide a
tool for separating valid and invalid inferences in the formal language, and it does
this job even if we take the model theory to be merely an algebraic construction
devoid of other semantic significance.
This theory is a toy, and I do not mean to propose it as a serious theory of facts.
Nonetheless, it has a very serious consequence for any genuine theory of facts. The
class of models we have seen assures us that for any genuine theory of facts, which
restricts itself to claims endorsed in these models, is demonstrably non-collapsing.
So, perhaps we do not endorse all of constant domain quantified 5. Perhaps we do
not endorse all of the properties this theory takes . to have. If our genuine theory
of facts is properly weaker than this theory, it is still demonstrably non-collapsing.
Slingshot arguments can have no effect against any such theory.
Notice what we have done: We have proved that this account of facts is noncollapsing, even though we have said nothing about descriptions. This means that
our theory is incomplete and it needs supplementation if it is to tell us what to
say about the validity of arguments involving descriptions, but it does not mean
that this theory, in its incomplete state, is under any suspicion of collapse. We
can be completely confident that if we add a theory of descriptions which can be
Greg Restall, [email protected]

1.01

February 27, 2004

http://consequently.org/publications/

interpreted using the models we have presented then this extension of the theory
is secure against slingshot arguments. I will demonstrate this by providing two
different interpretations for descriptions.
The first interpretation of descriptions is Russellian. The phrase the fact that
can be translated in a Russellian fashion, taking [thex : (x)](x) to be shorthand for the pre-existing formula in our language: (x)((x) (y)((y) y =
x) (x)). This interprets descriptions in without extending the original language
in any way at all, and we are no nearer collapse than we were before.
The second interpretation of descriptions is referential. In this case, we add a
denotation clause for the new term x as follows:
(d) d,w (x) is the unique d D where [x d], w if there is such a d,
or it is otherwise.
This takes definite description terms to refer to the unique object satisfying them,
if there is any such object, or in cases where the standard reference fails, it takes
them to refer to the object in P(W). (This seems like a natural choice because
, the empty set of worlds, will never be the denotation of a succesful attribution
of fact-hood, as t . is true only when is actually true at some world.) This
choice for definite descriptions makes them genuinely referential, and it is a proper
extension of the language. It also introduces no new threat of collapse, because the
rest of the language is interpreted as before, and our counterexamples still stand.
In making these definitions for descriptions, it was not as if I had to struggle
to find accounts for descriptions which would break either - or -. I
simply took two pre-existing accounts of definitions off the shelf and applied
them. The models themselves dictated that one of - and - would fail.
In these cases it is +-. All

contexts definable in our language are +-
provided that x x = a (x) and (a) are true at the same worlds (they agree
in intension as well as extension on our models). This holds for both of our accounts
of descriptions, so - will hold for every context expressible in our language.
The mistake in the collapse inference, according to our theory, is therefore and a straightforward counterexample is not difficult to find. Consider the
putative inference
xFx = a t . (a = xFx)
t . (a = a)
which is an - inference for the context t . (a = [ ]). This inference fails in
our theory (on either account of descriptions, whether they are referential or not)
because xFx = a may be true in some worlds (where the denotation of the name
a is the unique object satisfying Fx) and not others (where the denotation of the
name a is no longer the only object satisfying Fx). So, if t . (a = xFx) is true, then
the denotation of t is the set of all worlds where a = xFx is true. But this is not
necessarily the set of all worlds whatsoever, which is what is required if t . (a = a)
is to be true. So, it is straightforward to show that the context t . (a = [ ])
is -, irrespective of whether descriptions are treated referentially or in a
Russellian manner.
This model theory is not to be taken too seriously as a genuine contender for a
model theory for a genuine theory of facts. I leave it to the fact theorist to show

Greg Restall, [email protected]

1.01

February 27, 2004

http://consequently.org/publications/

how more comprehensive and interesting theories may be shown to be consistent.1

It is merely a single example of a general technique. It shows to prove that many
plausible inference principles involving facts and descriptions are unproblematic
and secure against slingshots, no matter how they are refined. Neale seems either
ignorant or unimpressed with this reasoning: he contents himself with sharpening
up the slingshot arguments, and he leaves any non-triviality proofs to others. This
leaves the uncommitted readerwho is agnostic on the matter of the triviality of
fact theories in general and who wishes to gain an understanding of what works and
what doesntfeeling distinctly unsatisfied after reading the book. After all, logic
is not just proof theory, it is also model theory. One can use ever-more sophisticated
slingshot arguments to approach the boundary between the fact theories which
work and the fact theories which dont from one side, but no matter how far you
advance, this will not tell you as much as if you also advance to that boundary from
the other side. A more balanced work on the topic would have approached this
boundary from both sides. Neales Facing Facts reminds you of the boxer who fights
with only one fist. It is capable as far as it goes, and it is remarkable how well he
does with the tools he has allowed himself. Nonetheless, it is ungainly. A more
deft work, at the one time more measured and judicious, yet more interesting and
definitive, would have resulted had Neale availed himself of the other fist.
* * *
In the abstract I claimed that friends of facts need not be troubled by slingshot
arguments. Here is how I can make good this claim: Provided that a theory of
facts has a model, it is resistant to collapse arguments. Any slingshot argument
to a repugnant conclusion not true in the model must appeal to a principle not
endorsed in that model. Models, then, can provide a guide to the options available
to the friend of facts in resisting slingshot arguments. To make the point using a
different metaphor, triviality arguments on the one hand, and models on the other,
mark out different kinds of territory on the map of theories of facts. Slingshot
arguments show that certain places depicted on that map are uninhabitable. They
show that particular combinations of principles are incoherent. Models, on the
other hand, show that other places on the landscape are safe. Given a model, the
principles endorsed in that model are coherent and non-collapsing.
Of course, the coherence or consistency of a collection of principles is one
thing, and its truth is another. Models for fact theories give us all the assurance we
need that those theories are consistent, that life on that patch of land is possible.
It is another thing altogether to decide that we should take up residence there. We
will need more than just a slingshot to make that decision.

1 This is not merely a request for future development of fact theories. There are extant theories of
possibilities, facts, truthmakers and the like, developed with models which are given a treatment such as
this. Neale never addresses this material. Not once does Neale broach the use of models as a technique
for explaining where a slingshot argument might go wrong.

Greg Restall, [email protected]

1.01

February 27, 2004