You are on page 1of 29

Journal of Philosophical Logic (2005) 34: 275–303 © Springer 2005

DOI: 10.1007/s10992-004-7800-2

ROBERTA BALLARIN

VALIDITY AND NECESSITY

ABSTRACT. In this paper I argue against the commonly received view that Kripke’s for-
mal Possible World Semantics (PWS) reflects the adoption of a metaphysical interpretation
of the modal operators. I consider in detail Kripke’s three main innovations vis-à-vis Car-
nap’s PWS: a new view of the worlds, variable domains of quantification, and the adoption
of a notion of universal validity. I argue that all these changes are driven by the natural tech-
nical development of the model theory and its related notion of validity: they are dictated
by merely formal considerations, not interpretive concerns. I conclude that Kripke’s model
theoretic semantics does not induce a metaphysical reading of necessity, and is formally
adequate independently of the specific interpretation of the modal operators.

KEY WORDS: Carnap, Kripke, modal logic, necessity, possible world semantics, validity

1. I NTRODUCTION

The history of the model theory of modal logic, commonly known as


possible world semantics, did not unfold linearly. As a consequence, it is
particularly complex, rich and interesting in its own terms.1 In this paper
I will grossly oversimplify the historical development of possible world
semantics (from now on PWS) in order to focus on a theoretical question:
Is it true that some crucial formal developments that occurred from the
pioneering work of Carnap to the nowadays commonly accepted Kripkean
version of PWS are the natural formal reflections of philosophical consid-
erations, mainly of the switch of focus from Carnap’s logical modalities to
Kripke’s metaphysical necessity and possibility?
Most philosophers interested in the formal semantics of modal logic
seem inclined to answer positively the above question. Cocchiarella for
example labels the following (Kripkean) practice a ‘model-theoretic arti-
fice’: “[A]llowing modal operators to range over only some and not all
of the worlds. . .” and claims that such an ‘artifice’ is “quite appropriate
and may in fact be required for operators purportedly representing non-
logical modalities (e.g., temporal or causal modalities) . . . [H]owever, . . .
the employment of such an artifice is inappropriate in the semantics of
what one considers to be a purely formal . . . sign.”2
More recently, Hintikka and Sandu have pressed the same point. The
title of the final section of their “The Fallacies of the New Theory of Refer-
ence” states their position with the boldness of a slogan calling for a much
276 ROBERTA BALLARIN

needed reform program: Kripke Semantics Is Not the Right Semantics of


Logical Modalities. It is worth quoting extensively from this section:
[I]n its usual form, the so-called Kripke semantics is not the correct semantics for logical
modalities either. As has been pointed out repeatedly . . . Kripke semantics . . . is analogous
to the non-standard interpretations of higher-order logics, which is not equivalent with the
intended standard interpretation of these logics. In other words, the so-called Kripke’s se-
mantics does not provide us with the right model theory of logical (conceptual) necessities
in any case.
Hence the New Theorists either have to change the logic they are basing their discussion
on or else admit that they are not dealing with purely logical (alethic) modalities, but with
some kind of metaphysical necessity and possibility.3

Not everybody calls for a logical reform, nonetheless it is widely as-


sumed that the kind of necessity one is philosophically interested in deter-
mines the formal semantics one adopts – or at least should adopt. David
Kaplan for example speculates on the interaction between two alterna-
tive notions of necessity (logical and metaphysical) and two correspond-
ing approaches to validity (maximal and universal).4 And very recently
Sten Lindström has argued that Kanger’s formal semantics for modal logic
is adequate for the notion of logical necessity, while Kripke’s PWS is
adequate for Kripke’s own metaphysical view of necessity.5
In this paper I challenge this widespread opinion. My contention is that
the formal development of PWS is best understood as driven by technical
considerations intrinsic to the formal semantics itself, and not by overt or
covert philosophical agendas. To support my claim, I will consider three
crucial aspects in which Kripke’s formal semantic apparatus for modal
logic altered the received Carnapian PWS. These three changes are crucial
both intrinsically and insofar as they might be, and have been, viewed as
dictated by philosophical considerations. I will argue on the contrary that
these changes are due to merely formal reasons. A purely formal explana-
tion of these formal developments is called for. We can and must detach
alternative philosophical views of necessity from logical considerations
regarding (intrinsically) the model theory of modal systems. Finally, I will
argue that Kripke’s (formal) semantics is in some important sense the right
(formal) semantics for modal systems independently of one’s logical or
metaphysical understanding of the modalities.6

2. C ARNAP : S TATE -D ESCRIPTIONS , A NALYTICITY,


AND L OGICAL T RUTH

I will start by introducing some of the basic ideas behind Carnap’s formal
PWS. In 1946 Carnap published “Modalities and Quantification”, and thus
VALIDITY AND NECESSITY 277

proved Quine de facto wrong insofar as his early well-known criticisms to


quantified modal logic seemed to suggest some kind of technical unfeasi-
bility. In this work, Carnap presents a quantified modal system and offers
some ideas concerning its proper interpretation, ideas that one year later
he will develop in Meaning and Necessity.7
Carnap states explicitly that the notion of necessity he has in mind is
logical/analytic:
[T]he guiding idea in our constructions of systems of modal logic is this: a proposition p
is logically necessary if and only if a sentence expressing p is logically true. That is to say,
the modal concept of the logical necessity of a proposition and the semantical concept of
the logical truth or analyticity of a sentence correspond to each other.8

Carnap endorse the analytic interpretation of the modal operators prevalent


at the time, according to which modal operators represent in the object
language the analogue of the meta-predicate of analyticity. According to
such a view, the modal operators reflect at the object level a semantic
predicate, i.e., a predicate of sentences; moreover, such operators operate
on intensional entities, viz., on propositions.9
Carnap speaks of “the logical truth or analyticity of a sentence,” imply-
ing their equivalence. However, already in 1943 Quine pointed out that the
class of logical truths is at best a proper subclass of the class of analytic
truths.10 I believe that Carnap’s disregard of the distinction is a manifes-
tation of what Quine called the ‘epistemologically biased’ view of logical
truth, according to which the key semantic feature of logical truths consists
in their (presumed) analyticity, while their specific difference from analytic
truths in general plays no significant semantic role. In this framework, the
interesting semantic notion is analyticity, and – argues Quine – ultimately
a priority.11 Carnap introduces the apparatus of state-descriptions to eluci-
date the relatively unclear notions of analyticity and (analytic) necessity.
However, the apparatus can be adopted to represent different semantic
interpretations of necessity, i.e., to capture different semantic properties
to which necessity may correspond at the object-language level.
A state-description for a language L is a set-theoretic entity, more pre-
cisely a class of sentences of L such that, for every atomic sentence S
of L, either S or its negation, but not both, is contained in the class. A state-
description does not contain non-atomic sentences (other than negations of
atomic sentences). The semantic values of non-atomic sentences relative
to a state-description are calculated on the basis of the values of atomic
sentences in the customary inductive way, under the basic assumption that
an atomic sentence holds in a state-description if and only if it belongs to
it. For example, ∼S (where S need not be atomic) holds (is true) in a state-
description R if and only if S does not hold in R; (S ∧ T ) holds in R if and
278 ROBERTA BALLARIN

only if both S and T hold in R; (∀x)P x holds in R if and only if all the
substitution instances of P x hold in R.
By definition, a sentence S of L is L-true just in case it holds in every
state-description for L. Therefore, the proposition expressed by S is nec-
essary just in case it is true in every possible world (given that state-
descriptions are taken to describe possible worlds).
The appeal to possible worlds should not mislead us into believing that
some metaphysical, rather than semantic, notion of necessity is at stake.
As we have seen, Carnap explicates necessity explicitly in terms of the
semantic notion of analyticity, where this last is represented in terms of
L-truth. Possible worlds, through their linguistic representations (state-
descriptions), are just used to elucidate this semantic notion.
To be an adequate formal representation of analyticity, L-truth has to
reflect the basic idea behind analyticity: truth in virtue of meaning alone.
Hence, L-truths must be such that semantic rules alone need to be em-
ployed to establish their truth. With this purpose in mind, L-truth for a
sentence S of a language L is defined as truth in all the state-descriptions
of L.12
According to Carnap, intuitively a state-description is supposed to rep-
resent something like a Leibnizian possible world or a Wittgensteinian
possible state of affairs.13 The entire range of state-descriptions for a cer-
tain language is supposed to exhaust the range of alternative possibilities
(describable in that language). Clearly, it is the fact that state-descriptions
represent possible worlds – possible ways things might have been – that
makes Carnap’s formal apparatus intuitively apt to represent necessity and
possibility. Carnap’s appeal to Wittgenstein’s states of affairs adds the
further intuition that all combinatorially consistent combinations (of truth-
value assignments to atomic propositions) are indeed possible.
Carnap’s elucidation of the modalities by means of state-descriptions
encapsulates two distinct ideas. First there is the semantic ascent, the idea
that necessity corresponds to a semantic property. Insofar as state-des-
criptions serve the purpose of defining L-truth, they encode the interpretive
idea that necessity is ultimately analyzed in terms of a semantic property.
Second, Carnap’s reference to Wittgenstein and to the idea of logical con-
sistency suggests a combinatorial extension for necessity, i.e., what we
might characterize as a logical understanding of necessity.
However, the Wittgensteinian logical/combinatorial view of necessity
based on logical consistency on the one hand and the analytic interpre-
tation of necessity on the other are in conflict. The presence of a conflict
is witnessed by the conflict in extension between the two notions. Some
consistent combinations of truth-value assignments to atomic sentences
VALIDITY AND NECESSITY 279

are ruled out by the analytic interpretation. Meaning postulates are then
needed to exclude such combinations. As Quine points out:

In recent years Carnap has tended to explain analyticity by appeal to what he calls state-
descriptions. . . . The criterion in terms of state-descriptions is a reconstruction at best of
logical-truth, not of analyticity.14

Quine’s main concern is that the atomic statements of the language may
not be semantically independent. If that is the case, a state-description may
verify two incompatible statements, for example, “John is a bachelor” and
“John is married”. Hence, such a description is not suitable to represent
an (analytically) possible world: surely there is no world where John is
both a bachelor and married. To exclude cases of this kind, the atomic
terms of the language must be logically independent from one another
(in a way in which ‘married’ and ‘bachelor’ are not). Alternatively, the
logical connections between atomic terms must be spelled out by means,
for example, of Carnap’s meaning postulates, which rule out those state-
descriptions that do not correspond to authentic analytic possibilities, and
so reinstate analyticity as the main encoded notion.
In sum, state-descriptions can be taken to encapsulate at least two dis-
tinct notions: analyticity (analytic necessity) on the one hand and logi-
cal truth (logical necessity) on the other. To underline the distinction be-
tween analytic necessity and logical necessity, we need just notice the nat-
ural different extensions of these two notions. Analytic necessity excludes
some logically consistent state-descriptions (for example, states descrip-
tions containing both “John is a bachelor” and “John is married”). On the
other hand, if we disregard the meanings of all but the logical expressions
of the language all logically consistent combinations become possible.
From the logical point of view, the natural extension of necessity includes
all combinatorially consistent combinations. Further considerations may
induce us to restrict the class of state-descriptions. These considerations
however are semantic in nature, and not purely logical – in the sense
that they depend on the logical connections that hold between interpreted
non-logical terms (like ‘bachelor and ‘married’).15
It is important to notice that despite Carnap’s focus on (i) analytic-
ity and (ii) combinatorial/logical necessity, state-descriptions may also be
seen as playing the same role that Tarskian mathematical models play.16
We can view state-descriptions as logically consistent combinations of
truth-value assignments to the uninterpreted non-logical atomic sentences
and predicates of the language. In this way, analyticity, which has to do
with interpreted sentences, is set aside, and a new formal notion of validity
emerges. Such a notion applies to (partially) uninterpreted sentences.17
280 ROBERTA BALLARIN

3. F ORMAL VALIDITY AND S TATE -D ESCRIPTIONS

Once again, let us look at Quine to find a third possible semantic inter-
pretation of necessity. Quine points out that we may provide a semantic
reading of necessity (link necessity to a property of sentences) by linking
necessity to some formal notion of validity:
Something very much to the purpose of the semantical predicate ‘Nec’ is regularly needed
in the theory of proof. When, e.g., we speak of the completeness of a deductive system
of quantification theory, we have in mind some concept of validity as norm with which to
compare the class of obtainable theorems. The notion of validity in such contexts is not
identifiable with truth. A true statement is not a valid statement of quantification theory
unless not only it but also all other statements similar to it in quantificational structure are
true. Definition of such a notion of validity presents no problem, and the importance of the
notion for proof theory is incontestable.
A conspicuous derivative of the notion of quantificational validity is that of quan-
tificational implication. One statement quantificationally implies another if the material
conditional composed of the two statements is valid for quantification theory.
This reference to quantification theory is only illustrative. There are parallels for truth-
function theory: a statement is valid for truth-function theory if it and all statements like it
in truth-functional structure are true, and one statement truth-functionally implies another
if the material conditional composed of the two statements is valid for truth-functional
theory.
And there are parallels, again, for logic taken as a whole: a statement is logically valid if
it and all statements like it in logical structure are true, and one statement logically implies
another if the material conditional formed of the two statements is logically valid.18

[I]t is at the semantical or proof-theoretic level, where we talk about expressions and their
truth values under various substitutions, that we make clear and useful sense of logical
validity; and it is logical validity that comes nearest to being a clear explication of ‘Nec’,
taken as a semantical predicate.19

We see here the suggestion that necessity be linked to validity. But what
is validity? Validity might prima facie be confused with logical truth, after
all “a statement is logically valid if it and all statements like it in logical
structure are true”. However, this characterization is too narrow. Validity
is system relative, and not all systems are naturally seen as ‘logical’. Some
candidate examples of non-logical systems are second order logic, first-
order set theory, or (closer to home) modal systems. Validity is crucial to
proof-theory in providing a standard to which to compare the theorems
of the system, and if possible obtain soundness and completeness results.
But clearly not all truths provable in any formal system may be plausibly
regarded as logical truths (expressible in the language of that system).
Quine viewed this formal understanding of necessity as a reduction of
the obscure notion of necessity to the clear notion of validity. In fact, Quine
thought it illuminating because reductive:
VALIDITY AND NECESSITY 281

As long as necessity in semantical application is construed simply as explicit truth-function-


al validity, on the other hand, or quantificational validity, or set-theoretic validity, or va-
lidity of any other well-determined kind, the logic of the semantical necessity predicate
is a significant and very central strand of proof-theory. But it is not modal logic, even
unquantified modal logic, as the latter ordinarily presents itself; for it is a remarkably
meager thing, bereft of all the complexities which are encouraged by the use of ‘nec’ as
a statement operator. It is unquantified modal logic minus all principles which, explicitly
or implicitly . . . involve iteration of necessity; and plus, if we are literal-minded, a pair of
question marks after each ‘Nec’.20

To sum up, we find in Carnap’s definition of L-truth the seeds of three


alternative semantical understandings of necessity. The apparatus of state-
descriptions may be taken to encode the notion of analyticity, and in this
sense meaning postulates are a crucial part of the apparatus itself. Alter-
natively, it may be taken as a codification of the notion of logical truth,
schematically characterized as truth resistant to substitution of the non-
logical constants, and independently of any further thesis one may hold
concerning the nature of logical truths. In this second interpretation, mean-
ing postulates are best seen as additions to the basic apparatus. Finally, it
can be taken as codifying the validities of some formal system. Validities
are still schematically characterized, yet we do have validities of systems
that may not be considered part of logic proper.
These three semantic understandings of necessity share a common pre-
supposition. After picking a relevant semantic property, they treat state-
descriptions as a means to codify such a property. But given that necessary
truths are assumed to be extensionally equivalent with the semantically
privileged truths (be they analytic, logical or valid), the necessary truths of
the object language must match exactly the truths across all states descrip-
tions (questions of iteration aside).21
However, as suggested earlier, state-descriptions are early versions of
Tarskian models, as such apt to capture a formal notion of validity for the
object language at hand – formal insofar as it applies to formal sentences,
i.e., uninterpreted sentences. The question to be considered is the follow-
ing: What is the proper role of this model theoretic notion of validity? If
it is, as Quine suggests, a notion central to proof theory, its central role
consists in providing a class of truths with which to compare the class of
theorems of the system.
However, insofar as state-descriptions serve this proof-theoretic pur-
pose for modal systems themselves, they are not meant to provide a class
of L-truths extensionally equivalent to the class of necessary truths. Their
role consists rather in providing a model theoretic match to the theorems
of a modal system. But as Quine says: “The notion of validity in such
contexts is not identifiable with truth. A true statement is not a valid state-
282 ROBERTA BALLARIN

ment of quantification theory unless not only it but also all other statements
similar to it in quantificational structure are true.” In this perspective, there
is no guarantee that the theorems of a modal system include all the true
necessities expressible in the modal language of the system. In this formal,
Tarskian perspective, the purpose of the model theory of modal logic is to
find model theoretic matches to the theorems of the modal systems, not to
provide a reductive interpretation of the object language operators in terms
of a Carnapian match of validities to necessities.

4. F ROM C ARNAP TO K RIPKE

Quine’s 1947 paper “The Problem of Interpreting Modal Logic” starts by


saying,
There are logicians, myself among them, to whom the ideas of modal logic . . . are not
intuitively clear until explained in non-modal terms.22
We may think of Carnap’s work in the forties and of the work of other
logicians in the late-fifties as an attempt to respond to this complaint by
putting modal logic on equal footing with the familiar non-modal systems
of logic. The idea was to extend to it a form of the Tarski-style extensional
semantics of first order logic.23 Carnap’s work on the modalities starts the
important model theoretic approach to the semantics of modality. State-
descriptions are the precursors of the model theoretic apparatus of possible
worlds that will be so fruitfully employed in the late fifties and early six-
ties to provide a proof theoretically adequate model theoretic semantics
for modal logic.24 This, I claim, is a completely different role than the
one played by state-descriptions to provide an interpretation of the modal
operators.
Carnap’s work attempts both:
(i) To link necessity to the semantic notion of truth across all state de-
scriptions and explicate it in terms of validity;
and
(ii) To provide a proof theoretically adequate extensional formal seman-
tics of modal logic.25
I claim that these two attempts are intrinsically at odds with each other.
To support my claim, in what follows I shall analyze the evolution of
PWS from Carnap’s work to Kripke’s 1959 “A Completeness Theorem
in Modal Logic” and 1963 “Semantical Considerations on Modal Logic”
and “Semantical Analysis of Modal logic I”. In particular, I will concern
myself with three main points:
VALIDITY AND NECESSITY 283

(1) The nature of the ‘worlds’;


(2) The domain(s) of quantification;
(3) The notion of validity.
According to a very widespread view – explicitly upheld by Cocchia-
rella, Hintikka and Sandu, Lindström, and at least implicitly by Kaplan –
Kripke’s development of the model theory produces a new notion of va-
lidity appropriate for a metaphysical understanding of necessity. But if the
Kripkean formal semantics is to impose a non-logical notion of necessity,
it must be in its changes concerning the worlds, the domains of quantifica-
tion, and/or the notion of validity itself.26 I will argue however that all the
above developments are best understood as driven by the Quinean task of
developing a formal notion of validity central to proof theoretic purposes.
My focus on Kripke’s model theoretic innovations is not meant to sug-
gest that such innovations were suggested and adopted by him exclusively.
In the fifties and early sixties, Hintikka, Montague, and Kanger all mod-
ified Carnap’s formal semantics in ways similar to Kripke’s.27 However,
it is Kripke who is standardly characterized as providing a model theory
inadequate for logical necessity. Moreover, Kripke is the only one who
adopts all the three main modifications discussed in this paper.
Concerning the nature of the worlds, Kripke was the only one who
characterized them as simple points of evaluation. Such a characterization
is crucial in affording a link between the model theoretic semantics and
algebraic treatments of modal logic.28 On the other hand, variations of
the domain of quantification are considered both by Kanger and Hintikka.
Finally, and interestingly enough, Hintikka like Kripke also adopted a new
notion of validity that required truth in all arbitrary sets of worlds. Despite
his own adoption of this universal notion of validity, in his later work, as
we have seen, Hintikka criticizes Kripke’s model theory as inadequate for
logical necessity exactly because of the notion of validity it endorses.

5. T HE NATURE OF THE ‘W ORLDS ’

Carnap’s notion of validity for a given language is truth across all state-
descriptions for that language. State-descriptions are collections of sen-
tences. They are taken to represent possible worlds, or possible states of
affairs. But what possible states of affairs there are is determined by the
state-descriptions (unless as we have seen one wants to restrict them in or-
der to capture an antecedently given idea of possibility, as in Carnap’s case
of analytic possibility). In this sense, the span of possibilities represented
by the state-descriptions is very much bound to the language whose sen-
284 ROBERTA BALLARIN

tences they collect. Both in the obvious sense that only the (combinatorial)
possibilities expressible in the language can be represented, but also in the
less trivial sense that all combinatorially consistent sets of sentences of the
language are taken to describe possible states of affairs.
Later on, in 1963, Carnap himself adopts models in place of state-
descriptions.29 Models are assignments of values to the primitive non-
logical constants of the language. In Carnap’s case predicate constants
are the only primitive constants to which the models assign values, since
individual constants are given a fixed pre-model interpretation. Value as-
signments to variables are done independently of the models.
In 1959, Kripke also uses models, i.e., complete assignments of values,
as representatives of possible worlds. The terminology however can be
misleading: what Kripke in 1959 calls ‘models’ are not such assignments.
To avoid confusion, I will reserve the term ‘model’ for the assignments of
values, and use instead ‘M-model’ for Kripke’s (modal) models.
Given a domain D of individuals, a model is an assignment of values to
the variables of the language (Kripke’s language has no non-logical con-
stants), such that each propositional variable is assigned a truth-value, each
individual variable is assigned an element of D, and each n-adic predicate
variable is assigned a set of ordered n-tuples of elements of D. A Kripke
M-model instead is an ordered pair (G, K), such that K is a set of complete
assignments (models), G is an element of K, and all elements of K agree
in their assignments to individual variables.
Consider Kripke’s 1959 models, representatives of possible worlds. As
seen, they have evolved from Carnap’s sets of sentences (state-descriptions)
to assignments of values, i.e., mathematical functions that correlate syntac-
tical entities to values. Nonetheless, such assignments are still very much
language driven, in the sense that all combinatorially consistent assign-
ments of values are possible.
In 1963, a further evolution takes place concerning the nature of the
worlds in PWS. Worlds are not anymore represented by models, rather
simply by points of evaluation. Of the set K of worlds in a model structure,
Kripke just says that it is a non-empty set. Nothing is assumed concerning
the nature of the elements of K.30 What is the significance of this change?
Does this technical change in the model theory have philosophical reper-
cussions? In particular, does it reflect the adoption of a new interpretation
of the modal operators of the object language?31 I claim that it does not.
Rather than reflecting a new philosophical interpretation of the modal
operators, the adoption of points in place of models reflects a more abstract
understanding of the formal semantics. The formal semantics of a modal
system does not formally represent or encode some antecedent semantic
VALIDITY AND NECESSITY 285

conjecture (e.g., that “necessarily” means true in all alternative possible


worlds). Instead it provides an algebraic characterization of the system.
Kripke himself recognizes the connection between his formal semantics
for modal logic and work in the algebras of modal systems:
The most surprising anticipation of the present theory, discovered just as this paper was al-
most completed, is the algebraic analogue in JÓNSSON and TARSKI [“Boolean Algebras
with Operators”]. Independently and in ignorance of [“Boolean Algebras with Operators”]
(though of course much later), the present writer derived its main theorem by an algebraic
analogue of his semantical methods . . .32

Moreover, as Kripke notices, the changes in his formal semantics are


generalizations of his own 1959 treatment – “The present treatment gen-
eralizes that of [1959] in the following respects”33 – but surely such a
generalization can hardly be viewed as apt to or designed for capturing
independent specific assumptions on the proper interpretation of the modal
operators. The generalizations Kripke is referring to are the introduction
of the accessibility relation R which makes it possible to deal with a whole
range of modal systems, not only S5; and, more importantly for our present
concern, the switch from worlds as models to worlds as points. Here is
what Kripke says:
For in [1959], we did not have an auxiliary function  to assign a truth value to P in
a world H; instead H itself was a “complete assignment”, that is a function assigning a
truth-value to every atomic subformula of a formula A. On this definition, “worlds” and
complete assignments are identified; so distinct worlds give distinct complete assignments.
This last clause means that there can be no two worlds in which the same truth-value is
assigned to each atomic formula. Now this assumption turns out to be convenient perhaps
for S5, but it is rather inconvenient when we treat normal [Modal Propositional Calculi] in
general.34

In this passage, Kripke connects the switch from worlds-as-models to


worlds-as-points to his current concern with a larger spectrum of modal
systems. In S5 the accessibility relation can be taken to connect every
world with every other world including itself; hence there is no need of
world-duplicates. Once the accessibility relation gets restricted, however,
world-duplicates become important in achieving model theoretic general-
ity. They make it possible to have model structures in which worlds differ
in the way they relate to each other, while being intrinsically indistinguish-
able – both in the sense of having intrinsically indistinguishable worlds
in the same model structure, and in the sense of having different model
structures differing only for what concerns the accessibility relation R.
Once the worlds are not identified with assignments, an external func-
tion  is needed to assign values to variables relative to worlds. An
M-model is now a model structure (G, K, R) plus an evaluation function .
There is absolutely no sense in which it is natural to think of such model
286 ROBERTA BALLARIN

theoretic constructions (vis-à-vis the 1959 M-models) as better suited to


represent a non-semantic notion of metaphysical necessity. The function 
is there to do the job of the old assignments. The suggestion that if the as-
signments are done by independent functions (themselves members of K)
then a logical interpretation of necessity is natural, while if the assignments
are all dealt with by a function  on the elements of K, then a restriction on
all the combinatorial combinations of assignments is more natural (and so
a metaphysical interpretation of necessity is implied) is in my view highly
implausible.
I believe that the idea behind such a suggestion is the following: While
assignments are language driven, points are given independently of the
language; hence they lend themselves more naturally to excluding combi-
natorial assignments that do not correspond to metaphysical possibilities
(“There is just no such point!”).
Such a view thrives on the identification of points with worlds and on
the further idea that the worlds are the real possibilities.35 But if worlds
are supposed to have any kind of metaphysical reality, they just cannot be
algebraic points to which a function  assigns values. If we are inclined
to think of worlds as somehow real possibilities, we should make sure
not to identify them with points of evaluation, and just think of the points
as world representatives. But then why would the points of evaluation be
better fit to represent all the metaphysically possible worlds, rather than all
the combinatorially consistent assignments of values (or anything else for
that matter)? If on the other hand, we are not inclined to think of the worlds
as anything other than the elements of K, then we may identify the worlds
with the points, but then a world is just that: a locus of evaluation, in no
way better fitted to represent one interpretation of necessity over another.

6. T HE D OMAIN ( S ) OF Q UANTIFICATION

In 1946, Carnap assumes a fixed domain of quantification for his quantified


system (functional calculus) and consequently for his modal functional
calculus. He is well aware that such an assumption raises the question of
completeness, even before the addition of modal operators. Gödel proved
completeness for the first order predicate calculus with identity, but he
employed a notion of validity as truth in every (non-empty) domain of
quantification. Carnap instead adopts one unique denumerable domain of
quantification.36 The adoption of a fixed denumerable domain of individu-
als generates some additional validities already at the pre-modal level. For
example (Carnap’s example), it becomes valid that there are at least two in-
dividuals. Carnap raises the question of the completeness of the functional
VALIDITY AND NECESSITY 287

calculus with identity vis-à-vis his own notion of validity, nonetheless he


adopts a fixed denumerable model. This is a clear sign that his notion of
validity is driven by external, philosophical rather than technical consider-
ations – he chooses a unique domain of a fixed size despite the consequent
loss of completeness.
In 1959, Kripke adopts what we may in retrospect consider as a middle
position between Carnap and Kripke’s later 1963 work. Domains may vary,
and only the empty domain is excluded, but each M-model is defined on
an antecedently given domain. This amounts to positing no variability of
domains between the models inside a modal structure (an M-model), while
postulating variability across M-models. This change consents to regain
Gödel’s indifference to the domain of quantification in establishing the
validity of non-modal formulas. Concerning modal formulas, this kind of
limited variability of the domain of quantification already generates some
differences from Carnap.
Assuming Carnap’s fixed domain, it is not only valid that there are
at least two individuals, but also that it is possible that there are at least
two individuals, and that it is necessary that there are at least two indi-
viduals. There is no world with fewer than two individuals to invalidate
these claims. In Kripke’s semantics the variability of the domains across
M-models is enough to invalidate all of these three claims. It is not valid
that there are at least two individuals, that possibly there are at least two
individuals, nor that necessarily there are at least two individuals. The kind
of domain variability assumed in 1959 is already by itself, even if no other
changes in the model theory are adopted, sufficient to produce a significant
change in the notion of validity, and a first detachment of validity from
necessity. Validity is truth across all Domains (hence across all M-models
on each Domain),37 but necessity is now relative to a given Domain for an
M-model.
This has two kinds of consequences. In the first place, non-modal sen-
tences that are not valid may still turn out to be necessary in a certain
Domain. For example, M-models with domains with two or more individ-
uals will make “There are at least two individuals” necessary, while the
presence of M-models with one-individual domains will suffice to make
this same sentence not valid. In the second place, the validity of modal
sentences is affected too. For example, neither “Necessarily, there are at
least two individuals” nor “Possibly, there are at least two individuals” will
turn out valid, but they will still be true, and necessarily so, in M-models
defined on Domains containing more than one individual.38
In 1963, a further change takes place. Kripke introduces variability of
domains not only across distinct M-models, but also across worlds in the
288 ROBERTA BALLARIN

same M-model. Kripke does support such a change with informal argu-
ments. For example, concerning the function ψ that assigns domains to
the worlds H in an M-model, he says:
Notice, of course, that ψ(H) need not be the same set for different arguments H, just as,
intuitively, in worlds other than the real one, some actually existing individuals may be
absent, while new individuals, like Pegasus, may appear.39

Despite Kripke’s informal, philosophical motivation, this change too seems


in line with the general trend of moving towards a more general, algebraic
model theory, where fewer and fewer restrictions are placed on the com-
binations of worlds and domains. Moreover, there is a particular technical
problem with Kripke’s 1959 semantics.
Thanks to the assumption of the variability of domains across (intra-
M-model) worlds, Kripke is able to construct a counter-example both to
the ‘Barcan Formula’ and its converse. Let us just consider the Barcan
Formula:

(BF) (∀x)2F x → 2(∀x)F x

Kripke considers a structure with two worlds, the actual G point and
one possible world H extending it. The domain of G is the individual a,
which is F (and thus all things in G are F ). The domain of H is the
set {a, b}. a is still F at H and so we get that the antecedent of (BF) is true
at G. But the consequent is false. It is false because (∀x)F x is false at H.
And this last is false because the new individual b – a mere possibilium
from G’s point of view – is not-F at H.
However, as Kripke mentions, Prior seems to have proved the Barcan
formula in quantified S5.40 If this is the case, the 1963 model theory in-
validates S5-theorems. However, Kripke suggests that neither the Barcan
formula nor its converse is really provable in S5. He reconstructs an al-
leged proof for the converse Barcan formula, and shows how the proof
goes through only by allowing the necessitation of a sentence containing
a free variable. But if free variables are to be considered as universally
bound, then necessitating directly an open formula, without first closing it,
amounts to assuming the derivability of the necessitated open formula from
the necessitation of its closure, which is what was to be proved. I.e., from
“F x” we should only derive “2(∀x)F x”, given that “F x” is to be read
as “(∀x)F x”. If instead from “F x” we are allowed to derive “2F x”, and
understand it as “(∀x)2F x”, we are implicitly assuming the derivability
of this last from “2(∀x)F x”.
The question to be considered is the following: If the Barcan formula
and its converse are not theorems of S5, how could Kripke have proved
completeness in 1959 while adopting a model theory that provided no
VALIDITY AND NECESSITY 289

countermodel to these formulas? Recall that the 1959 model theory does
not allow domains to vary inside M-models. Hence, both the Barcan for-
mula and its converse hold in every 1959 M-model. If so, they are validities
of the 1959 model theory. But they aren’t provable in S5. The fact is that
in giving his completeness theorem in 1959, Kripke had assumed Prior’s
alleged result (see p. 9 of Kripke’s 1959 paper).
The realization of the improvability of (BF) and its converse under the
standard reading of free variables as universally bound is sufficient to jus-
tify the 1963 revision of the model theory. If these formulas are not prov-
able, in order to have completeness we need model theoretic constructions
that provide counterexamples to their alleged validity.41
Independently of what came first in Kripke’s mind – be it the selection
of modal structures with variable domains, the proof theoretic considera-
tion that F x must be first universally closed and then necessitated, or even
a philosophical intuition concerning possible objects – what is ultimately
essential is that in 1963 like before in 1959 the essential task is to provide
a match between a certain class of structures and a particular proof sys-
tem. Changes in the model theory and in the proof theory proceed hand in
hand. The logician’s formal interest lies in finding the appropriate class of
structures corresponding to a certain system, or the right system to capture
a certain class of structures.

7. M AXIMAL VALIDITY VERSUS U NIVERSAL VALIDITY

The above considerations naturally bring us to our next topic: validity. As


we have seen, Kripke claims to have derived on his own the main theorem
of “Boolean Algebras with Operators” by an algebraic analogue of his own
semantical methods. The main theorem of Jónsson and Tarski’s work is a
general representation theorem for Boolean algebras with operators. Such
a theorem is the algebraic analogue of a model theoretic completeness
theorem for modal systems. This brings us to a crucial shortcoming of
Carnap’s notion of validity.
We have seen how Carnap takes inspiration from Wittgenstein’s combi-
natorial view according to which a logical truth is characterized by holding
for all possible distributions of truth-values. In propositional logic, this
amounts to a logical truth having the value true at all rows of its truth-
table. Once modal operators are added, Carnap assumes an interpreta-
tion of necessity according to which 2p is true just in case p is valid,
i.e., true at all truth-table rows. Carnap’s notion of validity (and neces-
sity) is maximal validity, i.e., truth across all the truth-table rows/state-
descriptions/assignments of values.
290 ROBERTA BALLARIN

Given the way state-descriptions are built out of atomic sentences of


the language, it follows that each atomic sentence and its negation turn
out to be true at some, but not all, state-descriptions. Hence, given that a
sentence 2ϕ is true in a state-description if and only if ϕ is true in every
state-description, it follows that neither 2p nor 2∼p is ever going to be
true for an atomic p. Hence, their negations ∼2p and ∼2∼p (♦p) are
validities in Carnap’s maximal sense. Moreover, given any two atomic
sentences p and q there is surely a state-description in which both turn
out to be true.
This notion of validity has a significant logical consequence: the prin-
ciple of substitution fails. If, for example, given atomic p and q we sub-
stitute ∼p for q, then no state-description will assign them the same truth-
value. Similarly, we cannot substitute for p a logically complex sentence
which is either logically true or logically false, e.g., (r ∨ ∼r) or (r ∧ ∼r).
In Carnap’s semantics, for every atomic sentence p both ♦p and ∼2p are
valid. But we do not want ∼2(r ∨ ∼r) and ♦(r ∧ ∼r) to turn out valid.
According to Quine’s schematic characterization of logical truth/vali-
dity, in order to qualify as logically true/valid a sentence has to remain
true even when its simple elements are substituted by logically complex
elements. Hence, some of Carnap’s maximal validities do not qualify as
schematically valid (for example ♦p for atomic p). If we accept Quine’s
schematic characterization of validity, and we think of the logic of neces-
sity as aiming at providing the schematic validities of a modal language,
we will have to conclude that Carnap’s model theory is inadequate to cap-
ture the schematic validities of even logical necessity: No matter which
interpretation of the modal operators one adopts, ♦p is not schematically
valid.42
Technically, this situation calls for a choice. In a modal propositional
system that adopts as an axiom ♦p for every atomic sentential letter p, the
rule of substitution must be restricted. Alternatively, if the rule of substitu-
tion is not restricted (or the axioms are given as axiom schemes), sentences
like ♦p for atomic p cannot be added as axioms. In other words, we cannot
have a system that has both axioms like ♦p with atomic p and a uniform
rule of substitution; otherwise we could derive as a theorem ♦(p ∧ ∼p).43
This point was noticed by Makinson who argued that (Carnap’s) naïve
logical understanding of the modal operators leads to failure of substi-
tutivity. If one however considers the schematic nature of logical truths
and revises accordingly the naïve notion of logical necessity, one will be
natural brought to reinstate substitutivity.44
Burgess endorses and develops Makinson’s point. He essentially points
out that for propositional S5, the universal notion of validity matches in
VALIDITY AND NECESSITY 291

extension the schematic notion. Hence one may well adopt a Kripkean
model theory not because of interpretational questions on necessity, but
because one adopts a standard fully substitutional view of propositional
variables.45
Concerning quantified modal logic, Carnap’s notion of maximal va-
lidity makes it impossible to prove completeness. According to maximal
validity, the models of a modal language L∗ are exactly the models of its
non-modal part L.46 Given a first order language L, all its models are taken
to represent possible states of affairs. A necessary truth is a truth across all
such models. So in a modal extension L∗ of a first order language L, 2ϕ is
true (in each model) if and only if ϕ is valid. Hence, for any non-valid first-
order sentence ϕ (any first-order sentence ϕ that is true at most in some
but not all of the models), ∼2ϕ will be true in all the models, i.e., valid.
(Carnap’s models all agree in the evaluation of modal sentences.) But
the non-logically true first-order sentences are not recursively enumerable,
hence neither are the validities of the modal language. Hence, quantified
modal logic is incomplete vis-à-vis Carnap’s maximal validity.47
It may reasonably be conjectured that it was primarily the search for a
completeness proof for quantified S5 that led Kripke to revise Carnap’s
model theory and introduce his 1959 notion of universal validity. The
crucial change consists in the introduction of M-models. An M-model,
remember, is an ordered pair (G, K), with G an element of K, and K an
arbitrary subset of models/assignments of values. Validity is not anymore
defined as truth across all models, rather as truth across all M-models.
Once again, this is a move that adds generality to the model theory. Instead
of considering only the maximal structure that contains all models, all the
subsets of the maximal structure are considered. The maximal structure is
now one among others, and not all sentences that are valid in it (maximally
valid in Carnap’s sense) will turn out to be universally valid:
In trying to construct a definition of universal logical validity, it seems plausible to assume
not only that the universe of discourse may contain an arbitrary number of elements and
that predicates may be assigned any given interpretations in the actual world, but also that
any combination of possible worlds may be associated with the real world with respect
to some group of predicates. In other words, it is plausible to assume that no further
restrictions need be placed on D, G, and K, except the standard one that D be non-empty.
This assumption leads directly to our definition of universal validity.48

My account above simplifies matters in not considering the role of the


G model. In fact, to any set of models K there corresponds more than one
M-model, according to which element of K is selected as the actual world.
This however has no effect on which sentences turn out to be universally
valid.49 Moreover, to be careful, given that in 1959 M-models are defined
on Domains, not every subset of models counts as K in an M-model. Only
292 ROBERTA BALLARIN

subsets of models on the same domain form a set K in an M-model. That


is how universal validity is defined as validity in every non-empty domain
in 1959. As we have seen, by 1963 even this last restriction is lifted: now
every arbitrary subset of worlds can play the role of K in a model structure.
Kripke’s model theory disconnects validity from necessity. In 1959,
Kripke defines validity in an M-model as truth in the actual world of the
Model, and not as truth in all the worlds of the Model. Hence, even for
one single M-model necessity does not correspond to validity. In 1963,
Kripke revises his terminology and calls a sentence simply true in an M-
model when true in the actual world of the M-model.50 Nonetheless, the
original 1959 terminology suggests that Kripke, insofar as he was willing
to speak of validity for a single M-model (set of worlds), never assumed
a correspondence between necessity and validity. More importantly, uni-
versal validity is truth in all M-models (sets of worlds), while necessity
is always relative to an M-model. Carnap’s project of linking necessity to
validity is simply not pursued.

8. C ONCLUSION

Given the above considerations, I conclude that all the three main changes
in the model theory considered in this paper – the evolution from state-
descriptions to models and then to points of evaluation; the change from
one fixed domain to variable domains across model structures, and then to
variable domains inside model structures; finally, the switch from maximal
to universal validity – point towards a more general, combinatorial, alge-
braic model theory, and are justified by technical rather than philosophical
considerations, especially by the search for completeness results.51
This conclusion refutes the widespread claim that it is the switch to
metaphysical necessity and possibility in his philosophical work to dic-
tate Kripke’s adoption of a new notion of Universal Validity and his other
reforms of the received model theory for modal logic.
Contrary to Cocchiarella, Hintikka, Sandu, and Lindström I believe that
Kripke’s model theory reflects a change in the notion of model theoretic
validity itself, a new view of its role, not a change in the notion of ne-
cessity. In my view, Kripke’s model theory is the right model theory even
for the logical modalities.52 Once we abandon Carnap’s project of linking
the interpretation of the modal operators to the notion of validity adopted
for the modal system under consideration, we are free to start judging the
model theory in its own terms. In this sense, the right model theory is
not the model theory whose class of validities matches a preferred class
of necessities, rather the model theory that adequately represents the proof
VALIDITY AND NECESSITY 293

theoretic apparatus; hence that affords soundness and completeness results.


Kripke provided a completeness proof for quantified S5, something we saw
Carnap could not possibly achieve.53
It is worth noticing that the claim that Kripke’s universal validity cor-
responds to a metaphysical notion of necessity is not even plausible on
the face of it. Kripke’s model theory reduces the class of validities. For
example, ♦p for atomic p is not universally valid, given the presence of
(incomplete) structures where all worlds verify ∼p. If we were interested
in a notion of validity to be linked to Kripke’s metaphysical necessity, we
would have to pursue the project of increasing the number of validities, i.e.
necessities, rather than cutting them back.54
More importantly, once the project of linking the semantical property
of validity to the interpretation of the modal operators is finally abandoned,
one may conclude not only that Kripke’s Universal Validity does not reflect
a particular philosophical stand on necessity, but also that, from a formal
point of view, this is the right notion of model theoretic validity, regardless
of one’s metaphysical or logical interpretation of necessity.
Moreover, those who sympathize with Quine’s schematic characteriza-
tion of validity will find a whole bunch of Carnapian logical necessities
(for example ♦p for atomic p and perhaps even “Possibly there exist at
least two individuals”) very implausible candidates to the status of (modal)
validities. From this schematic point of view, there are not only Kripkean
metaphysical necessities, but also Carnapian logical necessities that are
not schematic in nature, hence unfit to play the role of (modal) logical
validities.
To sum up, once the project of linking the object language modal op-
erator of necessity of a certain modal system to the notion of validity for
the modal system under consideration is abandoned, validity starts playing
a crucial proof theoretic role and not an interpretational function. I have
argued that it is the proof theory, and the search for a matching class of
structures, which best explains Kripke’s development of PWS. Such de-
velopment is technically natural, and interpretationally irrelevant. It surely
does not capture as validities the new class of metaphysical necessities.
Neither does it capture the class of Carnap’s logical necessities. Kripke’s
validities are at best fit to capture the class of schematic validities for S5
(on this see the Appendix). By detaching necessity from validity, and not
pursuing the project of an extensional match between the two, Kripke has
freed his hands and made it possible to provide a more austere logic of
necessity, one that does not make all necessary truths of one kind or another
valid. His model theoretic apparatus, designed as it is to correspond to the
294 ROBERTA BALLARIN

proof theory, allows him to regain a schematic notion of validity even for
a modal language.55

ACKNOWLEDGEMENTS

I wish to thank Joseph Almog, David Kaplan, Tony Martin, and the SMU
Philosophy Discussion Group for helpful comments and conversations,
and an anonymous referee of this journal for his valuable suggestions.

A PPENDIX

The main thesis of this paper is that a model theoretically adequate notion of validity has
to match extensionally the proof theoretic notion of theorem, if anything at all. At the
end of the paper I also suggest that Kripke’s universal validities may well correspond to
Quine’s schematic validities. However, in “Opacity” David Kaplan argues that there are
some schematic (modal) validities that are not universally valid.56 Kaplan’s example is
“Possibly, there exist at least two things”. Given that this is true, it is a schematic validity
(there are no non-logical terms to substitute). However, it is not universally valid, given the
presence of M-models where all worlds contain only one individual. In such M-models,
“Possibly, there exist at least two things” is false. Hence this sentence is not universally
valid, i.e., not true in all Kripkean M-models. If Kaplan is right, my claim that Kripke’s
notion of universal validity corresponds to a (Quinean) notion of schematic validity for the
modal language cannot be correct.
Lindström argues that while Kripke models refute the validity of sentences like “Pos-
sibly, there exist at least two individuals” such sentences should be valid for the logic of
logical necessity.57 It should be clear by now that in my view the logic of logical necessity
should be no different from the logic of metaphysical necessity, and that I regard Kripke
models as apt to capture the right logic (notion of validity) in both cases.
Kaplan’s point with which I am taking issue now is a further one: it does not concern
which notion of model theoretic modal validity is the right one. It has to do instead with a
comparison between a schematic notion and Kripke’s model theoretic universal conception
of validity. Kaplan claims that these two conceptions differ in extension. I claim that they
do not, once schematic validity is extended to include full domain variability.
Questions concerning the cardinality of the domain (from which Kaplan’s example
is drawn) are of a rather special kind, already at the pre-modal level. Even in the case
of a non-modal first-order language, we have a coincidence between the model theoretic
and the schematic notions of validity only if we define schematic validity as truth for all
replacements of the non-logical signs and for all domains. Kaplan says of a schematic
validity that “it would be true no matter how we were to reinterpret its non-logical signs”,
but then he glosses this definition as follows: “i.e., no matter what grammatically appro-
priate expressions are substituted for the non-logical signs and no matter what domain of
discourse the variable[s] [are] taken to range over.” (“Opacity”, p. 275, emphasis added).
I regard the addition of domain variability as an adjunct to the purely schematic char-
acterization of validity. However, if such an addition is welcome, shouldn’t we in a sim-
ilar vein add a corresponding change to accommodate modal sentences? If non-modal
VALIDITY AND NECESSITY 295

schematic validities are independent of how many actual individuals there are, shouldn’t
modal schematic validities be similarly independent of how many possible individuals
there are?
Consider “There are at least two individuals.” This is not by Kaplan’s standards a
schematic validity, even if it is indeed true under all replacements of its non-logical signs
(it contains no such signs to be replaced). It is not a schematic validity because Kaplan
assumes that how many individuals there actually are is not a matter of logic, and hence
adds resistance to domain variability to the definition of schematic validity/logical truth.
But then shouldn’t we in a similar spirit deny schematic validity to “Possibly, there
are at least two individuals” (which Kaplan deems a schematic modal validity), given
that this claim does not remain true if we allow variability not only of the actual domain,
but also of the combination of possible domains connected to it? If non-modal schematic
logical truths are independent of the actual size of the universe, shouldn’t modal schematic
logical truths be similarly independent of its possible sizes? In other words, shouldn’t we
go combinatorial all the way when providing the schematic logic of necessity? According
to such an extreme combinatorialism, it is consistent that any combination of possible sizes
(not just all combinations) may accompany any actual size of the universe. In calculating
the schematic validities of a modal language, we must consider not only the independence
of logic from the actual number of individuals, but also its independence from the possible
number of individuals. Not only should we consider the possibility of the actual world
containing only one individual, but also the possibility that it so does of necessity. If logic
has to be indifferent to what exists, it has to be indifferent to what might or might not exist
too. If logic is not concerned with how many individuals there really are (definitely more
than one), it must equally be unconcerned with how many there could have been. If pure
from questions of cardinality, logic should not claim that given any actual domain all other
domains are possible. Rather, given any domain any other combinations of domains are
possible.58
The suggested amendment to the notion of schematic validity develops a full analogy
between the quantifiers and the modal operators. Independently of what the real cardinality
of the world is, to capture the logic of the quantifiers, we make use of domains of all
different cardinalities. Similarly, independently of what the right span of possible domains
is according to one’s preferred view of necessity and possibility, to capture the logic of the
modal operators, we are to make use of all different spans of possible domains.
I suspect that what drives Kaplan’s intuitions is once again the old Carnapian assump-
tion of a linkage between necessity and validity. Consider the non-modal sentence “There
are at least two individuals”. In such a non-modal case logical intuitions prevail. The
intuition is that such a sentence is consistent, hence (logically) possible, independently
of how many individuals there actually are. Even the false sentence “There are at most two
individuals” is deemed possibly true, given its consistency.
Moreover, any hypothesis about the possible cardinality of the universe is regarded to
be true, independently of the real actual cardinality. This is a form of maximal combinatori-
alism concerning necessity: whatever the actual size of the universe, all sizes are regarded
as (logically) possible. And given the possibility that there are at least two things (or at
most two things, or what not), it is then necessarily possible that it be so. Once necessity
is settled, given its assumed link to validity, Kaplan judges it (schematically) valid that
possibly there are two things.
The hypothesis of a necessity-validity link induces first a cut in basic necessities, i.e.,
the generation of too many possibilities (all consistent sentences are deemed possible); but
296 ROBERTA BALLARIN

then, given that all possibilities are necessarily possible, we have an over-generation of
validities for the modal language.
I conclude that we should revise the notion of schematic validity for a modal language
to account for the possibility of any combination of possible domains to a given domain,
not just a maximal combination. If the amendment is accepted, the class of schematic
validities for S5 coincides with Kripke’s class of universal validities.59

N OTES
1 For an excellent review of the historical development of possible worlds semantics,
see B. J. Copeland, “The Genesis of Possible Worlds Semantics.” See also R. Goldblatt,
“Mathematical Modal Logic: a View of its Evolution” which covers the history of modal
logic from its early beginnings up to its most recent contemporary developments.
2 N. B. Cocchiarella, “On the Primary and Secondary Semantics of Logical Necessity”,
p. 26.
3 J. Hintikka & G. Sandu, “The Fallacies of the New Theory of Reference”, p. 281.
4 Cf. D. Kaplan, “Opacity”, Appendix E.
5 See S. Lindström, “Modality Without Worlds: Kanger’s Early Semantics for Modal
Logic” (1996) and “An Exposition and Development of Kanger’s Early Semantics for
Modal Logic” (1998). In the more recent “Quine’s Interpretation Problem and the Early
Development of Possible Worlds Semantics,” (2001), Lindström is more cautious in linking
Kripke’s formal semantics to a metaphysical interpretation of necessity. He writes there:
“One reason for arguing that Kripke’s notion of necessity in 1959 is not logical necessity
is Kripke’s use of non-standard models. . . . This conclusion is however, not unavoidable
. . . Kripke’s reason for allowing non-standard models, in addition to standard ones, when
defining validity, could have been logical rather than philosophical.” (p. 209).
6 In “Which Modal Models are the Right Ones (for Logical Necessity)?” John Burgess
argues against the idea that Kripke’s models are adequate for metaphysical necessity, but
not apt to represent logical necessity. Burgess’s main argumentative strategy differs from
mine. His central argument against this widespread misconception focuses on the claim
that such a misconception arises from a subtle use-mention confusion. My main strategy
instead does not consist in directly attacking such a view, but rather in showing how each
change in Kripke’s PWS is better explained as due to logical reasons, not to interpretive hy-
potheses about the nature of necessity. I do however claim that the misconception depends
on the assumption that the notion of validity one adopts is linked to the interpretation of
the object language operator of necessity. The assumption of such a linkage may be seen
as presupposing a use-mention confusion.
7 R. Barcan (Marcus)’s “A Functional Calculus of First Order Based on Strict Implica-
tion” was published just a few months before Carnap’s work. It also contains a quantified
modal system, but no semantic considerations.
8 R. Carnap, “Modalities and Quantification”, p. 34.
9 The two points are separate. For example, in characterizing analytic necessity, Quine
just says that 2p is true if and only if p is analytic, with no further talk of propositions
as semantic entities for the operator to operate upon. A semantic view of necessity can be
adopted without reference to intensional entities such as propositions.
10 “At best” because once analyticity is not part of the definition of logical truth, there is
no guarantee that all logical truths be analytically true.
VALIDITY AND NECESSITY 297
11 See W. V. Quine, “Carnap and Logical Truth”.
12 R. Carnap, Meaning and Necessity, Chapter I, Section 2.
13 Cf. R. Carnap, Meaning and Necessity, p. 9. We will see later how possible worlds
on the one hand and Wittgensteinian states of affairs on the other encapsulate two distinct
intuitions. Carnap seems aware of the distinction: “In my search for an explication I was
guided, on the one hand, by Leibniz’ view that a necessary truth is one which holds in
all possible worlds, and on the other hand, by Wittgenstein’s view that a logical truth or
tautology is characterized by holding for all possible distributions of truth-values.” (The
Philosophy of Rudolf Carnap, p. 63. Emphasis added.)
14 Cf. W. V. Quine, “Two Dogmas of Empiricism”, pp. 23–4.
15 I am talking of the natural extension of these notions. This does not rule out the
possibility that further considerations may bring one to conclude that the real extensions
of these notions are different from their natural ones, perhaps even that the two notions
coincide in extension. This would be the case, for example, if one assumed that there is
no logical connection between atomic expressions of the language because atomic, logi-
cally independent meanings are assigned to atomic expressions. This further assumption
makes analytic necessity coincide in extension with logical necessity, but it is a further
assumption.
16 Carnap switches to models in his later work: “In my book on syntax . . . and still
in [Meaning and Necessity], the values assigned by the semantical rules to variables and
descriptive constants were linguistic entities, viz., expressions, classes of expressions, etc.
Today I prefer to use as values extra-linguistic entities, e.g., numbers, classes of num-
bers, etc. In an analogous way I now represent possible states of the universe of discourse
by models instead of state-descriptions, which are sentences or classes of sentences.”
(“Language, Modal Logic, and Semantics” in The Philosophy of Rudolf Carnap, p. 891,
footnote 10.)
17 Of course, someone can be epistemologically biased toward logical truth or validity,
but the point is that he need not be.
18 W. V. Quine, “Three Grades”, p. 165.
19 W. V. Quine, ibid., p. 168. Quine’s semantic ‘Nec’ corresponds to Carnap’s ‘L-true’.
20 W. V. Quine, ibid., p. 171.
21 For the purpose of this discussion I set aside iterated modalities.
22 W. V. Quine, “The Problem of Interpreting Modal Logic”, p. 43.
23 The use of such extensional techniques is not unique to Carnap and Kripke. In the
late fifties and early sixties, the suggestion was made by S. Kanger “Provability in Logic”,
R. Montague, “Logical Necessity, Physical Necessity, Ethics, and Quantifiers”, and J. Hin-
tikka, “Modality and Quantification”. I am focusing on Carnap and Kripke partly be-
cause Kripke’s version of the possible world model theory has become the standard one,
and also because, as we shall see, the evolution of the formal semantics from Carnap to
Kripke brings to the fore the question of whether the extensional model theory provides an
interpretation of the modal operators.
24 In fact, Carnap himself in his more logical, rather than philosophical, work employs
state-descriptions to tackle logical questions of soundness and completeness, rather than to
provide philosophical clarifications; cf. “Modalities and Quantification”.
25 Carnap faces the question of the soundness and completeness of his modal systems in
“Modalities and Quantification”.
26 I consider the introduction of an accessibility relation R between worlds when dis-
cussing validity.
298 ROBERTA BALLARIN

27 See note 23.


28 R. Goldblatt argues convincingly that Kripke’s adoption of points of evaluation in
his model structures was a particularly crucial innovation. Such a generalization opened
the door to different future developments of the model theory and made it possible to
provide model theories for intensional logics in general. Because of this and other features
of Kripke’s model theory, Goldblatt claims that Kripke’s work deserves a special place in
the development of possible world semantics. See “Mathematical Modal Logic: a View
of its Evolution,” part 4. See also J. Burgess, “Kripke Models”, where similar points are
emphasized and the overall superiority of Kripke’s work in formal semantics is argued for.
29 Cf. “Language, Modal Logic, and Semantics”, in The Philosophy of Rudolf Carnap,
pp. 889–900.
30 In 1963, a model structure contains also an accessibility relation R between worlds,
needed to represent a notion of relative possibility. We are not concerned here with R, nor
with G, which is taken to represent the actual world.
31 J. Almog suggests that this is the case. He says that worlds taken as primitive points,
rather than models, are better suited to represent Kripke’s idea of metaphysical possibil-
ity/necessity, rather than the combinatorial idea of truth across all consistent assignments
of values (cf. “Naming without Necessity”, pp. 217–8). Almog’s suggestion has recently
been supported by S. Neale (cf. “On a Milestone of Empiricism”, p. 321).
32 S. Kripke, “Semantical Analysis of Modal logic I. Normal Modal Propositional Cal-
culi”, p. 69, footnote 2.
See also Kripke’s Reviews of E. J. Lemmon’s “Algebraic Semantics for Modal Log-
ics”, where Kripke claims of Lemmon’s “algebraic method of proving completeness the-
orems for certain modal propositional logics” that it “can be obtained by straightforward
applications of the methods . . . of the reviewer.” (p. 1021). This, by the way, was exactly
the point Lemmon wanted to make.
33 S. Kripke, ibid., p. 69. Emphasis added.
34 S. Kripke, ibid., p. 69.
35 Cf. J. Almog: “But they still lacked the idea of the world-as-a-point, possible worlds
as primitives, each with a structure intrinsic to it.” (“Naming without Necessity”, p. 218).
Points, I would think, are intrinsically structure-less. It must be their identification with
worlds that drives the idea of intrinsic structure.
36 He also assumes a one-to-one relation between individual constants and the objects of
the domain. But this is a further point from the one under consideration.
37 And, as a consequence, across all models/worlds. Kripke defines truth in an M-model
as truth in the actual world of the M-model, not as truth in all the models/worlds of the
M-model. Nonetheless, when it comes to validity across all M-models, the two definitions
make no extensional difference, at least before an operator sensitive to the choice of the
actual world like “Actually” is introduced.
38 Notice that even if Carnap had assumed Gödel’s domain variability, his validities about
the cardinality of the universe would not have coincided with Kripke’s. For any finite
number of individuals, it would still be valid in Carnap’s maximal sense that possibly
there are that many things. No such sentence is valid in Kripke’s universal sense (except
for “Possibly, there is at least one individual”, given the exclusion of the empty domain).
39 Kripke, “Semantical Considerations on Modal Logic”, reprinted in Reference and
Modality, p. 65.
40 Cf. A. N. Prior, “Modality and Quantification in S5”.
VALIDITY AND NECESSITY 299
41 Kripke is very careful, and does not bluntly state that (BF) and its converse are not
provable. We have seen that the provability of the converse of (BF) depends on how one
decides to treat free variables. The notion of provability that one adopts (like the notion of
validity that one chooses) is relative. What is interesting is to find the right match between a
certain notion of provability and a certain notion of validity, in order to find out which sys-
tem characterizes which class of structures, whether our main interest is proof-theoretical
(i.e., in the system) or model-theoretical (i.e., in the set-theoretical structures).
42 Cf. D. Kaplan, “Opacity”, Appendix E “Schematic Validity and Modal Logic”.
43 Interestingly, in his own formal work Carnap adopts a system equivalent to proposi-
tional S5, for which he proves completeness, but that does not reflect his notion of maximal
validity and the philosophical assumption that atomic p’s abbreviate logically independent
sentences (cf. “Modalities and Quantification”). S. Thomason has proved completeness for
a system that adds to the axioms of S5 all formulas ♦p, for atomic p (more precisely, the
axioms added state the possibility of any conjunction of distinct propositional constants or
their negations – of which ♦p for atomic p is a limit case). This system of Thomason, and
not S5, captures the notion of validity philosophically endorsed by Carnap, i.e., validity
across all state-descriptions. Cf. Thomason, “A New Representation of S5”. David Kaplan
has done unpublished work on this subject, from knowledge of which I have benefited.
44 See Makinson, “How Meaningful are Modal Operators?”
45 Cf. Burgess, “Which Modal Logic is the Right One?”, pp. 86–87.
46 Cf. Carnap, p. 892 of “Language, Modal Logic, and Semantics”, in The Philosophy of
Rudolf Carnap.
47 Things are really slightly more complicated. First of all, given Carnap’s assumption of
a fixed denumerable domain, not all first order models qualify as worlds. Secondly, even
allowing variability of the domains, the result needs to be more carefully stated. For a full,
precise statement, cf. N. B. Cocchiarella, “On the Primary and Secondary Semantics of
Logical Necessity”. Cocchiarella attributes the result to D. Kalish and R. Montague. See
also D. Kaplan, “Opacity”, pp. 253–4, and S. Lindström, “Quine’s Interpretation Problem
and the Early Development of Possible Worlds Semantics”, p. 209.
48 S. Kripke, “A Completeness Theorem in Modal Logic”, p. 3. Emphasis added.
49 At least for the language under consideration, before an actuality operator is intro-
duced.
50 Cf. Kripke, “Semantical Analysis of Modal Logic I”, p. 70.
51 See E. J. Lemmon’s “Algebraic Semantics for Modal Logics I” and “Algebraic Se-
mantics for Modal Logics II” for a very clear exposition of the correspondence between
Kripke’s model theory and algebras for modal systems. Lemmon connects Tarski and
McKinsey’s algebraic method to Kripke’s model theoretical method, by showing how
Kripke’s 1963 completeness results for various propositional modal systems can be derived
form algebraic completeness results. Lemmon’s main theorem proves that algebras for
modal systems can be represented as algebras based on the power set of the set K in the
corresponding Kripke’s structures. As a consequence, algebraic completeness translates
into Kripke’s model theoretic completeness. See also D. Makinson, “On Some Complete-
ness Theorems in Modal Logic”, where completeness results for various modal systems
are provided making use of Lindenbaum sets, the modal equivalent of the Löwenhein-
Skolem theorem is proved, and the possibility of mirroring these results in algebraic form
is emphasized.
52 This point is also made by Burgess in “Kripke Models” and “Which Modal Models
are the Right Ones (for Logical Necessity)?”.
300 ROBERTA BALLARIN

53 In my view, while it is the job of philosophers to investigate the notion(s) of necessity,


to evaluate different modal systems as better apt to represent different readings of the modal
operators, and also to inquire into an intuitive notion of logical truth for English modal
sentences (that may be represented in the formal language), it is not their job to devise the
right model theory for the system itself. This last task pertains to the mathematical logician.
54 One possible view could be that when it comes to metaphysical necessity we are not
looking for an unqualified extensional match between necessities and validities, rather
for a match between necessity in a structure and the validities in that structure. While
extensionally more plausible, this view still suffers of the criticism presented in this paper
to the idea that the evolution of the notion of validity corresponds to philosophical changes
in the notion of necessity, not to formal changes of validity itself. Curiously, S. Lind-
ström recognizes that Kripke’s logic of necessity is relatively meager and that the class of
Kripke’s validities is not meant to capture the class of metaphysical necessities, but he still
argues that Kripke’s validities are appropriate for metaphysical necessity only. This is so
because the new notion of validity is taken to induce at the object language level a new
meaning for the modal operators: “Observe how the relativization of the definition to C
[an arbitrary class of worlds] changes the meaning of 2. Instead of 2ϕ meaning that ϕ
is logically true, it now means that ϕ is universally true relative to the given set of state
descriptions.” (“Modality without Worlds”, p. 270.) In this paper I have argued precisely
against this general idea – whatever specific form it may take – that changes in the notion of
validity reflect at the meta-level corresponding changes in the interpretation of the modal
operators.
55 This isn’t surprising considering (Quine’s argument) that validity in terms of substitu-
tion (schematic logical truth) and validity in terms of models (model theoretic validity)
coincide when completeness and the Löwenhein-Skolem theorem hold, and given that
the modal equivalents of these theorems hold vis-à-vis Kripke’s model theory. See Quine,
Philosophy of Logic, part 4 on Logical Truth.
56 Cf. D. Kaplan, “Opacity”, pp. 275–6.
57 See Lindström, “Modality Without Worlds”, pp. 282–3.
58 Notice that these are conditional claims. I do not intend to defend the claim that logic
should indeed be indifferent to questions of cardinality. But if such a claim is accepted –
as it generally is and surely so by Kaplan – then cardinality issues should be completely
disregarded and we should not stop halfway.
59 In “Which Modal Models are the Right Ones (for Logical Necessity)?”, Burgess
considers the formula “♦∃x∃y∼(x = y)” and argues against some possible arguments
to regard it as valid. Since I do not take such a formula to be valid in some intuitive
sense, I am not going to disagree with Burgess on this. However, Burgess also claims that
questions about the validity of this formula are often treated in too simplistic a manner.
His reason for this last claim is that such a formula should not be translated into English
as saying “Possibly, there are at least two individuals”, but rather as something like: “For
any domain it is (logically) possible that there should be more than two elements in it.”
Moreover, Burgess claims that questions concerning the validity of this last sentence are
not obviously meaningful, because of Quinean reasons about different possible ways of
specifying the domain in question and how these different specifications may alter the
correct answer. This is not the right place to take issue with Burgess’s claim. Let me
only clarify a few points. First, in this paper I am not specifically concerned with the
notion of logical possibility. My argument is about any notion of necessity one may be
interested in. I have in fact argued that the notion of necessity/possibility at stake should
VALIDITY AND NECESSITY 301

not determine the notion of validity one adopts. Presumably, Quinean arguments of the kind
Burgess is considering apply exclusively to a logical understanding of necessity. Second,
I am not in the appendix arguing that “♦∃x∃y∼(x = y)” should indeed be regarded as
valid. Such a sentence is, as a matter of fact, not valid in Kripke models. But, Kaplan
claims, it is schematically valid. Contra Kaplan, my claim is simply that it might well not
be schematically valid given certain adjustments to the notion of schematic validity that
one should want to endorse. Finally, I do have a lot of sympathy for Burgess’s desire to
keep a clear distinction between formal sentences and English sentences, and I agree that
questions concerning the validity of formal sentences are to be kept apart from questions
concerning the intuitive logical truth of their English counterparts. However, I am not sure
that the right translation, if any, for “♦∃x∃y∼(x = y)” is the one Burgess proposes. But to
argue for this last point would take me far beyond the limits of this work.

R EFERENCES

Almog, J. (1986) Naming without necessity, J. Philos. 83, 210–242.


Barcan, R. C. (1946) A functional calculus of first order based on strict implication,
J. Symbolic Logic 11, 1–16.
Burgess, J. P. (1999) Which modal logic is the right one?, Notre Dame J. Formal Logic 40,
81–93.
Burgess, J. P., Kripke Models, available at http://www.princeton.edu/~jburgess/Kripke1.
doc.
Burgess, J. P., Which Modal Models are the Right Ones (for Logical Necessity)?, available
at http://www.princeton.edu/~jburgess/Hintikka.doc.
Carnap, R. (1946) Modalities and quantification, J. Symbolic Logic 11, 33–64.
Carnap, R. (1947) Meaning and Necessity, 2nd ed with supplements, The University of
Chicago Press, Chicago, 1956.
Carnap, R. (1963) Language, modal logic, and semantics, in P. A. Schilpp (ed.), The
Philosophy of Rudolf Carnap, Open Court, La Salle, IL, pp. 889–944.
Cocchiarella, N. B. (1975) On the primary and secondary semantics of logical necessity,
J. Philos. Logic 4, 13–27.
Copeland, B. J. (2002) The genesis of possible worlds semantics, J. Philos. Logic 31,
99–137.
Gödel, K. (1930) The completeness of the axioms of the functional calculus of logic,
reprinted in Heijenoort (ed.), 1967: From Frege to Gödel, Harvard University Press,
Cambridge, MA, pp. 582–591.
Goldblatt, R. (2003) Mathematical modal logic: A view of its evolution, J. Appl. Logic 1,
309–392.
Hintikka, J. (1961) Modality and quantification, Theoria 27, 119–128.
Hintikka, J. (1963) The modes of modality, Acta Philosophica Fennica 16, 65–79;
reprinted in M. J. Loux (ed.), The Possible and the Actual, Cornell University Press,
Ithaca, pp. 65–79.
Hintikka, J. and Sandu, G. (1995) The fallacies of the new theory of reference, Synthese
104, 245–283.
Jónsson, B. and Tarski, A. (1951) Boolean algebras with operators. Part I, Amer. J. Math.
73, 891–939.
Jónsson, B. and Tarski, A. (1952) Boolean algebras with operators. Part II, Amer. J. Math.
74, 127–162.
302 ROBERTA BALLARIN

Kanger, S. (1957) Provability in Logic, Almqvist and Wiksell, Stockholm.


Kaplan, D. (1986) Opacity, in Hahn and Schilpp (eds.), The Philosophy of W.V. Quine,
Open Court, La Salle, IL, pp. 229–289.
Kripke, S. A. (1959) A completeness theorem in modal logic, J. Symbolic Logic 24, 1–14.
Kripke, S. A. (1962) The undecidability of monadic modal quantification theory, Z. Math.
Logik Grundlag. Math. 8, 113–116.
Kripke, S. A. (1963a) Semantical analysis of modal logic I, Z. Math. Logik Grundlag.
Math. 9, 67–96.
Kripke, S. A. (1963b) Semantical considerations on modal logic, Acta Philosophica Fen-
nica 16, 83–94; reprinted in L. Linsky (ed.), 1971: Reference and Modality, Oxford
University Press, Oxford, pp. 63–72.
Kripke, S. A. (1965) Semantical analysis of modal logic II, in J. W. Addison, L. Henkin
and A. Tarski (eds.), Symposium on the Theory of Models, North-Holland Publ. Co.,
Amsterdam, pp. 206–220.
Kripke, S. A. (1967a) Review of Lemmon, Mathematical Reviews 34, 1021–1022.
Kripke, S. A. (1967b) Review of Lemmon, Mathematical Reviews 34, 1022.
Lemmon, E. J. (1966a) Algebraic semantics for modal logics I, J. Symbolic Logic 31,
46–65.
Lemmon, E. J. (1966b) Algebraic semantics for modal logics II, J. Symbolic Logic 31,
191–218.
Lindström, S. (1996) Modality without worlds: Kanger’s early semantics for modal logic,
in S. Lindström, R. Sliwinski and J. Österberg (eds.), Odds and Ends. Philosophical Es-
says Dedicated to Wlodek Rabinowicz on the Occasion of his Fiftieth Birthday, Uppsala,
Sweden, pp. 266–284.
Lindström, S. (1998) An exposition and development of Kanger’s early semantics for
modal logic, in P. W. Humphreys and J. H. Fetzer (eds.), The New Theory of Reference:
Kripke, Marcus, and its Origins, Kluwer Academic Publishers, Dordrecht, pp. 203–233.
Lindström, S. (2001) Quine’s interpretation problem and the early development of possible
worlds semantics, in E. Carlson and R. Sliwinski (eds.), Omnium-Gatherum. Philosophi-
cal Essays Dedicated to Jan Österberg on the Occasion of his Sixtieth Birthday, Uppsala,
Sweden, pp. 187–213.
Makinson, D. (1966a) How meaningful are modal operators?, Australasian J. Philos. 44,
331–337.
Makinson, D. (1966b) On some completeness theorems in modal logic, Z. Math. Logik
Grundlag. Math. 12, 379–384.
Montague, R. (1960) Logical necessity, physical necessity, ethics, and quantifiers, Inquiry
3, 259–269.
Neale, S. (2000) On a Milestone of empiricism, in Orenstein and Kotakto (eds.), Knowl-
edge, Language, and Logic, Kluwer Academic Publishers, Great Britain, pp. 237–346.
Prior, A. N. (1956) Modality and quantification in S5, J. Symbolic Logic 21, 60–62.
Quine, W. V. (1947) The problem of interpreting modal logic, J. Symbolic Logic 12, 43–48.
Quine, W. V. (1951) Two dogmas of empiricism, Philosophical Review 60, 20–43;
reprinted in Quine, 1953: From a Logical Point of View, Harvard University Press,
Cambridge, MA; 2nd revised edition, 1961; 3rd edition with a new Foreword, 1980,
pp. 20–46.
Quine, W. V. (1953a) Reference and modality, in From a Logical Point of View, Harvard
University Press, Cambridge, MA; 2nd revised edition, 1961; 3rd edition with a new
Foreword, 1980, pp. 139–59; reprinted in L. Linsky (ed.), 1971: Reference and Modality,
Oxford University Press, Oxford, pp. 17–34.
VALIDITY AND NECESSITY 303

Quine, W. V. (1953b) Three grades of modal involvement, in Proceedings of the XIth In-
ternational Congress of Philosophy, 14, Brussels, North-Holland Publishing Co., Am-
sterdam; reprinted in Quine, 1966: Selected Logic Papers, Random House, New York,
pp. 156–74.
Quine, W. V. (1954) Carnap and logical truth, first published in English in 1960, Synthese
12; reprinted in Quine, 1966: The Ways of Paradox, Random House, New York, pp. 107–
132.
Quine, W. V. (1970) Philosophy of Logic, Prentice-Hall, Englewood Cliffs, NJ.
Thomason, S. K. (1973) A new representation of S5, Notre Dame J. Formal Logic 14,
281–284.

Department of Philosophy
SMU
PO Box 750142
Dallas, TX 75275-0142, USA
E-mail: Rballari@smu.edu

You might also like