New - Work David Lewis

Australasian Journal o f Philosophy
Vol. 61, No. 4, December 1983
NEW WORK FOR A THEORY OF UNIVERSALS
D a v i d Lewis
Introduction. D. M. A r m s t r o n g offers a t h e o r y o f universals as the o n l y

Downloaded by [Yale University Library] at 10:25 04 June 2013
adequate a n s w e r to a ' c o m p u l s o r y q u e s t i o n ' for s y s t e m a t i c p h i l o s o p h y : the

problem o f O n e over Many.1 I find this line o f a r g u m e n t u n p e r s u a s v i e . But
I think there is m o r e to be said for A r m s t r o n g ' s t h e o r y t h a n he h i m s e l f has
said. F o r as I bear it in m i n d considering various topics in p h i l o s o p h y , I notice
time a n d a g a i n t h a t it offers s o l u t i o n s to m y p r o b l e m s . W h a t e v e r we m a y
think o f the p r o b l e m o f O n e over M a n y , universals can earn their living d o i n g
other m u c h - n e e d e d w o r k .
I do n o t say t h a t t h e y are i n d i s p e n s a b l e . T h e services t h e y r e n d e r c o u l d
be m a t c h e d using r e s o u r c e s t h a t are N o m i n a l i s t i c in letter, if p e r h a p s n o t
in spirit. 2 But neither d o I h o l d a n y p r e s u m p t i o n a g a i n s t universals, to t h e
effect that t h e y are to b e a c c e p t e d o n l y if we h a v e n o a l t e r n a t i v e . I t h e r e f o r e
suspend j u d g e m e n t a b o u t universals themselves. I o n l y insist t h a t , one w a y
or another, their w o r k m u s t be d o n e .
I shall investigate the benefits o f a d d i n g universals to m y o w n usual
ontology. T h a t o n t o l o g y , t h o u g h N o m i n a l i s t i c , is in o t h e r respects generous.
It consists o f possibilia - p a r t i c u l a r , i n d i v i d u a l things, s o m e o f which
comprise o u r a c t u a l w o r l d a n d o t h e r s o f w h i c h a r e u n a c t u a l i s e d 3 -- t o g e t h e r
with the iterative h i e r a r c h y o f classes built u p f r o m t h e m . T h u s I a l r e a d y
have at m y d i s p o s a l a t h e o r y o f p r o p e r t i e s as classes ofpossibilia. P r o p e r t i e s ,
so u n d e r s t o o d , are n o t m u c h like universals. N o r can they, u n a i d e d , t a k e
over the w o r k o f universals. Nevertheless t h e y will figure i m p o r t a n t l y in w h a t
D. M. Armstrong, Universals and Scientific Realism (Cambridge University Press, 1978),

henceforth cited as "Universals"see also his "Against 'Ostrich' Nominalism: A Reply to Michael
Devitt", Pacific Philosophical Quarterly 61 (1980) pp. 440-449.
z In this paper, I follow Armstrong's traditional terminology: 'universals' are repeatable entities,
wholly present wherever a particular instantiates them; 'Nominalism' is the rejection of such
entities. In the conflicting modern terminology of Harvard, classes count as 'universals' and
'Nominalism' is predominantly the rejection of classes. Confusion of the terminologies can
result in grave misunderstanding; see W. V. Quine, 'Soft Impeachment Disowned', Pacific
Philosophical Quarterly 61 (1980) pp. 450-451.
A,laong 'things' I mean to include all the gerrymandered wholes and undemarcated parts
admitted by the most permissive sort of mereology. Further, I include such physical objects
as spatiotemporal regions and force fields, unless an eliminative reduction of them should
prove desirable. Further, I include such nonphysical objects as gods and spooks, though
not - I hope -- as parts of the same world as us. Worlds themselvesneed no special treatment.
They are things -- big ones, for the most part.
343
344 New Work for a Theory of Universals
follows, since for me they are part of the environment in which universals
might operate.
The friend o f universals m a y wonder whether they would be better
employed not as an addition to m y ontology o f possibilia and classes, but
rather as a replacement for parts of it. A fair question, and an urgent one;
nevertheless, not a question considered in this paper.
In the next section, I shall sketch Armstrong's theory of universals,
contrasting universals with properties understood as classes ofpossibilia. Then
I shall say why I am unconvinced by the One over M a n y argument. Then
I shall turn to my principal topic: how universals could help me in connection
with such topics as duplication, supervenience, and divergent worlds; a
minimal form of materialism; laws and causation; and the content of language
and thought. Perhaps the list could be extended.
Universals and Properties. Language offers us several more or less

interchangeable words: 'universal'; 'property', 'quality', 'attribute', 'feature',
and 'characteristic'; 'type', 'kind', and 'sort'; and perhaps others. And
philosophy offers us several conceptions of the entities that such words refer
to. My purpose is not to fix on one of these conceptions; but rather to
distinguish two (at opposite extremes) and contemplate helping myself to
both. Therefore some regimentation of language is called for; I apologise
for any inconvenience caused. Let me reserve the word 'universal' for those
entities, if such there be, that mostly conform to Armstrong's account. And
let me reserve the word 'property' for classes - any classes, but I have
foremost in mind classes o f things. T o have a property is to be a member
of the class. 4
W h y call them 'properties' as well as 'classes'? - Just to underline the fact
that they need not be classes of actual things. The property of being a donkey,
for instance, is the class of all the donkeys. This property belongs to - this
class contains - not only the actual donkeys of this world we live in, but
also all the unactualised, otherworldly donkeys.
Likewise I reserve the word 'relation' for arbitrary classes of ordered pairs,
triples . . . . . Thus a relation a m o n g things is a property of 'tuples of things.
Again, there is no restriction to actual things. Corresponding roughly to
the division between properties and relations of things, we have the division
between 'monadic' and 'polyadic' universals.
Universals and properties differ in two principal ways. The first difference
concerns their instantiation. A universal is supposed to be wholly present
wherever it is instantiated. It is a constituent part (though not a spatiotemporal
part) of each particular that has it. A property, by contrast, is spread around.
The property of being a donkey is partly present wherever there is a donkey,
4 My conception of properties resembles the doctrine of Class Nominalism considered in

Universals, I, pp. 28-43. But, strictly speaking, a Class Nominalist would be someone who
claims to solve the One over Many problem simply by means of properties taken as classes,
and that is far from my intention.
David Lewis 345
in this or any other world. Far f r o m the p r o p e r t y being part o f the donkey,
it is closer to the truth to say that the d o n k e y is part o f the property. But
the precise truth, rather, is that the d o n k e y is a m e m b e r o f the property.
Thus universals would unify reality (Cf. Universals, I, p. 109) in a way
that properties do not. Things that share a universal have n o t just joined
a single class. They literally have something in c o m m o n . They are not entirely
distinct. T h e y overlap.
By occurring repeatedly, universals defy intuitive principles. But that is
no damaging objection, since plainly the intuitions were made for particulars.
For instance, call two entities copresent if b o t h are wholly present at one
position in space and time. W e might intuit o f f h a n d that copresence is
transitive. But it is not so, obviously, for universals. Suppose for the sake
of argument that there are universals: round, silver, golden. Silver and r o u n d
are copresent, for here is a silver coin; golden and r o u n d are copresent, for
there is a gold coin; but silver and golden are not copresent. Likewise, if
we add universals to an o n t o l o g y ofpossibilia, for the relation o f being part
of the same possible w o r l d : I and some otherworldly d r a g o n are not
worldmates; but I a m a w o r l d m a t e o f the universal golden, and so is the
dragon. P r e s u m a b l y I needed a mixed case involving b o t h universals and
particulars. F o r w h y should any two universals ever fail to be worldmates?
Lacking such failures, the w o r l d m a t e relation a m o n g universals alone is
ffivially transitive.
The second difference between universals and properties concerns their
abundance. This is the difference that qualifies t h e m for different work, and
thereby gives rise to m y interest in having universals and properties both.
A distinctive feature o f A r m s t r o n g ' s theory is that universals are sparse.
There are the universals that there must be to g r o u n d the objective
resemblances and the causal powers o f things, and there is no reason to believe
in any more. All o f the following alleged universals would be rejected:
not golden, first examined before 2000 A . D . ;
golden or w o o d e n , being identical,
metallic, being alike in some respect,
self-identical, being exactly alike,
owned by Fred, being part of,
belonging to class C, owning,
grue, being paired with by some part in R
If universals are to do the new work I have in store for them, they must be capable of repeated
occurrence not only within a world but also across worlds. They would then be an exception
to my usual principle - meant for particulars, of course - that nothing is wholly present
as part of two different worlds. But I see no harm in that. If two worlds are said to overlap
by having a coin in common, and if this coin is supposed to be wholly round in one world
and wholly octagonal in the other, I stubbornly ask what shape it is, and insist that shape
is not a relation to worlds. (See my 'Individuation by Acquaintance and by Stipulation',
PhilosophicalReview92 (1983), pp. 3-32.) I do not see any parallel objection if worlds are
said to overlap by sharing a universal. What contingent, nonrelational property of the universal
could we put in place of shape of the coin in raising the problem? I cannot think of any.
346 New Work for a Theory o f Universals
(where C and R are utterly miscellaneous classes). The guiding idea, roughly,
is that the world's universals should comprise a minimal basis for
characterising the world completely. Universals that do not contribute at all
to this end are unwelcome, and so are universals that contribute only
redundantly. A satisfactory inventory of universals is a non-linguistic
counterpart o f a primitive vocabulary for a language capable of describing
the world exhaustively.
(That is rough: A r m s t r o n g does not dismiss redundant universals out of
hand, as the spirit of his theory might seem to demand. Conjunctive universals
- as it might be, golden-and-round - are accepted, though redundant; so
are analysable structural universals. The reason is that if the world were
infinitely complex, there might be no way to cut down to a minimal basis.
The only alternative to redundancy might be inadequacy, and if so we had

better tolerate redundancy. But the redundancy is mitigated by the fact that
complex universals consist of their simpler -- if perhaps not absolutely simple
- constituents. They are not distinct entities. See Universals, II, pp. 30-42
and 67-71.)
It is quite otherwise with properties. Any class o f things, be it ever so
gerrymandered and miscellaneous and indescribable in thought and language,
and be it ever so superfluous in characterising the world, is nevertheless a
property. So there are properties in immense abundance. (If the number of
things, actual and otherwise, is beth-2, an estimate I regard as more likely
low than high, then the n u m b e r of properties of things is beth-3. A n d that
is a big infinity indeed, except to students of the outer reaches o f set theory.)
There are so m a n y properties that those specifiable in English, or in the brain's
language o f synaptic interconnections and neural spikes, could be only an
infinitesimal minority.
Because properties are so abundant, they are undiscriminating. Any two
things share infinitely m a n y properties, and fail to share infinitely many
others. That is so whether the two things are perfect duplicates or utterly
dissimilar. Thus properties do nothing to capture facts of resemblance. That
is work more suited to the sparse universals. Likewise, properties do nothing
to capture the causal powers o f things. Almost all properties are causally
irrelevant, and there is nothing to m a k e the relevant ones stand out from
the crowd. Properties carve reality at the joints -- and everywhere else as
well. I f it's distinctions we want, too much structure is no better than none.
It would be otherwise if we had not only the countless throng of all
properties, but also an 61ite minority o f special properties. Call these the
natural properties. 6 I f we had properties and universals both, the universals
See Universals, I, pp. 38-41; Anthony Quinton, 'Properties and Classes', Proceedings of
the Aristotelian Society 48 (1957) pp. 33-58; and W. V. Quine, 'Natural Kinds', in his
Ontological Relativity (Columbia University Press, 1969). See also George Bealer, Quality
and Concept (Oxford University Press, 1982), especially pp. 9-10 and 177-187. Like me,
Bealer favours an inegalitariantwofold conception of properties: there are abundant 'concepts'
and sparse 'qualities', and the latter are the ones that 'determine t.he logical, causal, and
phenomenal order of reality'. (p. 10) Despitethis point of agreement, however, Bealer'sviews
and mine differ in many ways.
David Lewis 347
could serve to pick out the natural properties. Afterwards the universals could
retire if they liked, and leave their jobs to the natural properties. Natural
properties would be the ones whose sharing makes for resemblance, and the
ones relevant to causal powers. Most simply, we could call a property perfectly
natural if its members are all and only those things that share some one
universal. But also we would have other less-than-perfectly natural properties,
made so by families of suitable related universals. 7 Thus we might have an
imperfectly natural property of being metallic, even if we had no such single
universal as metallic, in virtue of a close-knit family o f genuine universals
one or another of which is instantiated by any metallic thing. These
imperfectly natural properties would be natural to varying degrees.
Let us say that an adequate theory of properties is one that recognises an
objective difference between natural and unnatural properties; preferably,

a difference that admits o f degree. A combined theory o f properties and
universals is one sort of adequate theory of properties.
But not the only sort. A Nominalistic theory of properties could achieve
adequacy by other means. Instead of employing universals it could draw
primitive distinctions among particulars. Most simply, a Nominalist could
take it as a primitive fact that some classes of things are perfectly natural
properties; others are less-than-perfectly natural to various degrees; and most
are not at all natural. Such a Nominalist takes 'natural' as a primitive
predicate, and offers no analysis of what he means in predicating it of classes.
His intention is to select the very same classes as natural properties that the
user of universals would select. But he regards the universals as idle
machinery, fictitiously superimposed on the primative objective difference
between the natural properties and the others. 8
Alternatively, a Nominalist in pursuit of adequacy might prefer to rest with
primitive objective resemblance among things. (He might not think that
'natural' was a very natural primitive, perhaps because it is to be predicated
of classes.) Then he could undertake to define natural properties in terms
of the mutual resemblance of their members and the failure o f resemblance
between their members and their non-members. Unfortunately, the project
meets with well-known technical difficulties. These can be solved, but at a
daunting price in complexity and artificiality of our primitive. We cannot
get by with the familiar dyadic 'resembles'. Instead we need a predicate of
resemblance that is both contrastive and variably polyadic. Something like
xl, x2 . . . . resemble one another and do not likewise resemble any of yl,
Y2, . . .
7 Here I assume that some solution to the problem of resemblance of universals is possible,
perhaps along the lines suggested by Armstrong in Universals, II, pp. 48-52 and 101-131;
and that such a solution could be carried over into a theory of resemblance of perfectly
natural properties, even if we take naturalness of properties as primitive.
8 This is the Moderate Class Nominalism considered in Universals, I, pp. 38-41. It is akin
to the view of Quinton, op. cit.; but plus the unactualised members of the natural classes,
and minus any hint that 'natural' could receive a psychologisticanalysis.
(where the strings o f variables may be infinite, even uncountable) must be

taken as understood without further anaiysis. 9 If adequate Nominalism
requires us to choose between this and a primitive predicate o f classes, we
might well wonder whether the game is worth the candle. I only say we might
wonder; I know o f no consideration that seems to me decisive.
At this point, you may see very well why it could be a good idea to believe
in universals as well as properties; but you may see no point in having
properties as well as universals. But properties have work o f their own, and
universals are ill-suited to do the work o f properties.
It is properties that we need, sometimes natural and sometimes not, to
provide an adequate supply o f semantic values for linguistic expressions.
Consider such sentences as these:
(1) Red resembles orange more than it resembles blue.

(2) Red is a colour.
(3) Humility is a virtue.
(4) Redness is a sign o f ripeness.
Prima facie, these sentences contain names that cannot be taken to denote
particular, individual things. What is the semantic role of these words? If
we are to do compositional semantics in the way that is best developed, we
need entities to assign as semantic values to these words, entities that will
encode their semantic roles. Perhaps sometimes we might find paraphrases
that will absolve us from the need to subject the original sentence to semantic
Such a theory is a form of Resemblance Nominalism, in Armstrong's classification, but it

is unlike the form that he principally considers. See Universals, I, pp. 44-63. For discussions
o f the problem o f defining natural classes in terms of resemblance, and of the trickery that
proves useful in solving this problem, see Nelson Goodman, T h e S t r u c t u r e o f A p p e a r a n c e
(Harvard University Press, 1951), Chapters IV-VI; W. V. Quine, 'Natural Kinds'; and Adam
Morton, 'Complex Individuals and Multigrade Relations', N o D s 9 (1975) pp. 309-318.
To get from primitive resemblance to perfectly natural properties, I have in mind a definition
as follows. We begin with R as our contrastive and variably polyadic primitive. We want
it to turn out that x~,x2, • • . Ryl,Y2 . . . . iff some perfectly natural property is shared by
all of x~, x2, • . • but by none of yl, Y2,. • • • We want to define iV, another variably polyadic
predicate, so that it will turn out that N x b x 2 . . . . iff x~, x2 . . . . are all and only the members
of some perfectly natural property. Again we must allow for, and expect, the case where
there are infinitely many x's. We define N x l , x 2 , . . . as:
3yl,y2 . . . . Y z (z, x l , x 2 , . • . R y e , y 2 . . . . ~ z = x ~ v z = x 2 v . . . ).
Then we finish the job by defining a perfectly natural property as a class such that, if xl,
x2 . . . . are all and only its members, then N x l , x2 . . . . .
We might have taken N as primitive instead o f R. But would that have been significantly
different, given the interdefinability of the two? On the other hand, taking N as primitive
also seems not significantly different from taking perfect naturalness of classes as primitive.
It is only a difference between speaking in the plural of individuals and speaking in the singular
o f their classes, and that seems no real difference. Is plural talk a disguised form of class
talk? Or vice versa? (See the discussion in Universals, I, pp. 32-34; also Max Black, 'The
Elusiveness of Sets', R e v i e w o f M e t a p h y s i c s 24 (1971) pp. 614-636; Eric Stenius, 'Sets',
S y n t h e s e 27 (1974), pp. 161-188; and Kurt G6del, 'Russell's Mathematical Logic', in P. A.
Schilpp, ed., T h e P h i l o s o p h y o f B e r t r a n d R u s s e l l (Cambridge University Press, 1944).) At
any rate, it is not at all clear to me that Moderate Class Nominalism and Resemblance
Nominalism in its present form are two different theories, as opposed to a single theory
presented in different styles.
David Lewis 349
analysis. That is the case with (1), for instance.I° But even if such paraphrases
sometimes exist -- even if they always exist, which seems unlikely - they
work piecemeal and frustrate any systematic approach to semantics.
Armstrong takes it that such sentences provide a subsidiary argument for
universals, independent o f his main argument from the One over Many
problem. (Universals, I, pp. 58-63; also "Against 'Ostrich' Nominalism". n)
I quite agree that we have here an argument for something. But not for
universals as opposed to properties. Properties can serve as the requisite
semantic values. Indeed, properties are much better suited to the job than
universals are. That is plain even from the examples considered. It is unlikely
that there are any such genuine universals as the colours (especially
determinable colours, like red, rather than determinate shades), or ripeness,
or humility. Armstrong agrees (Universals, I, P. 61) that he cannot take (1)-(4)

as straightforwardly making reference to universals. He must first subject
them to paraphrase. Even if there always is a paraphrase that does refer to,
or quantify over, genuine universals, still the need for paraphrase is a threat
to systematic sematics. The problem arises exactly because universals are
sparse. There is no corresponding objection if we take the requisite semantic
values as properties.
Other sentences make my point more dramatically.
(5) Grueness does not make for resemblance among all its instances.
(6) What is common to all who suffer pain is being in some or another
state that occupies the pain role, presumably not the same state in
all cases.
The point is not that these sentences are true -- though they are -- but that
they require semantic analysis. (It is irrelevant that they are not ordinary
language.) A universal of grueness would be anathema; as would a universal
such that, necessarily, one has it if he is in some state or other that occupies
10 In virtue of the close resemblance of red and orange, it is possible for a red thing to resemble
an orange one very closely; it is not possible for a red thing to resemble a blue one quite
so closely. Given our ontology ofpossibilia, all possibilities are realised. So we could paraphase
(1) by
(1 ') Some red thing resembles some orange thing more than any red thing resembles
any blue thing.
so long as it is understood that the things in question needn't be part of our world, or of
any one world. Or if we did not wish to speak of unactualised things, but we were willing
to take ordinary-language modal idioms as primitive, we could instead give the paraphrase:
( 1 " ) A red thing can resemble an orange thing more closely than a red thing can resemble
a blue thing.
It is necessary to use the ordinary-language idioms, or some adequate formalisation of them,
rather than standard modal logic. You cannot express (1' ') in modal logic (excluding an
enriched modal logic that would defeat the point of the paraphrase by quantifying over degrees
of resemblance or whatnot) because you cannot express cross-world relations, and in particular
cannot express the needed cross-world comparison of similarity.
n He derives the argument, a n d a second semantic argument to be considered shortly, from
Arthur Pap, 'Nominalism, Empiricism, and Universals: I', Philosophical Quarterly 9 (1959)
pp. 330-340, and F. C. Jackson, 'Statements about Universals', Mind 86 (1977) pp. 427-429.
the pain role in his case.lZ But the corresponding properties are no problem.
Indeed, we have a comprehension schema applying to any predicate phrase
whatever, however complicated. (Let it even be infinitely long; let it even
include imaginary names for entities we haven't really named.) Let x range
over things, P over properties (classes) o f things. Then:
31PV7¥x (x has P =-- l~x).
We could appropriately call this 'the property of It-ing' in those cases where
the predicate phrase is short enough to f o r m a gerund, and take this property
to be the semantic value of the gerund. Contrast this with the very different
relationship of universals and predicates set forth in Universals, II, pp. 7-59.
Consider also those sentences which prima facie involve second-order
quantification. From Universals, I, p. 62, and "Against 'Ostrich' Nominalism"

we have these.
(7) H e has the same virtues as his father.
(8) The dresses were o f the same colour.
(9) There are undiscovered fundamental physical properties.
(10) Acquired characteristics are never inherited.
(11) Some zoological species are cross-fertile.
Prima facie, we are quantifying either over properties or over universals.

Again, paraphrases might defeat that presumption, but in a piecemeal way
that threatens systematic semantics. In each case, properties could serve as
the values of the variables of quantification. Only in case (9) could universals
serve equally well. T o treat the other cases, not to mention
(12) Some characteristics, such as the colours, are more disjunctive than
they seem.
as quantifications over universals, we would again have to resort to some

preliminary paraphrase. (Armstrong again agrees: Universals, I, p. 63.) This
second semantic argument, like the first, adduces work for which properties
are better qualified than universals.
Which is not to deny that a partnership might do better still. Let it be
granted that we are dealing with quantifications over properties. Still, these
quantifications - like most o f our quantifications - m a y be tacitly or
explicitly restricted. In particular, they usually a r e restricted to natural
properties. N o t to perfectly natural properties that correspond to single
universals, except in special cases like (9), but to properties that are at least
somewhat more natural than the great majority of the utterly miscellaneous.
That is so for all our examples, even (12). Then even though we quantify
over properties, we still need either universals or the resources of an adequate
Nominalism in order to say which o f the properties we mostly quantify over.
12 Or better, in the case of creatures of his kind. See my 'Mad Pain and Martian Pain', in Ned
Block, ed., Readings in Philosophy of Psychology, I (Harvard University Press, 1980).
David Lewis 351
I also think that it is properties that we need in characterising the content

of our intentional attitudes. I believe, or I desire, that I live in one o f the
worlds in a certain class, rather than any world outside that class. This class
of worlds is a property had by worlds. I believe, or I desire, that my world
has that property. (The class o f worlds also m a y be called a proposition,
in one of the legitimate senses of that word, and m y 'propositional attitude'
of belief or desire has this proposition as its 'object'.) More generally,
subsuming the previous case, I believe or I desire that I myself belong to
a certain class ofpossibilia. I ascribe a certain property to myself, or I want
to have it. Or I might ascribe a property to something else, or even to myself,
under a relation of acquaintance I bear to that thing.13 Surely the properties
that give the content of attitudes in these ways cannot be relied on to be
perfectly natural, hence cannot be replaced by universals. It is interesting

to ask whether there is any lower limit to their naturalness (see the final section
of this paper), but surely no very exacting standard is possible. Here again
properties are right for the job, universals are not.
One Over Many. Armstrong's main argument for universals is the 'One
over Many'. It is because I find this argument unconvincing that I am
investigating alternative reasons to accept a theory of universals.
Here is a concise statement of the argument, taken by condensation f r o m
"Against 'Ostrich' Nominalism", pp. 440-441. A very similar statement could
have been drawn f r o m the opening pages o f Universals.
I would wish to start by saying that m a n y different particulars can all have
what appears to be the same nature and draw the conclusion that, as a
result, there is a p r i m a facie case for postulating universals. We are
continually talking about different things having the same property or
quality, being of the same sort or kind, having the same nature, and so
on. Philosophers draw the distinction between sameness of token and
sameness of type. But they are only making explicit a distinction which
ordinary language (and so, ordinary thought) perfectly recognises. I suggest
that the fact of sameness of type is a Moorean fact: one o f the m a n y facts
which even philosophers should not deny, whatever philosophical account
or analysis they give of such facts. Any comprehensive philosophy must
try to give some account of Moorean facts. They constitute the compulsory
questions in the philosophical examination paper.
From this point of departure, Armstrong makes his case by criticising rival
attempts to answer the compulsory question, and by rejecting views that
decline to answer it at all.
Still more concisely, the One over M a n y problem is presented as the
problem of giving some account of M o o r e a n facts o f apparent sameness of
type. Thus understood, I agree that the question is compulsory; I agree that
1t See my 'Attitudes De Dicto and De Se', Philosophical Review 88 (1979) pp. 513-543; and
'Individuation by Acquaintance and by Stipulation'.
Armstrong's postulation of shared universals answers it; but I think that an
adequate Nominalism also answers it.
An effort at systematic philosophy must indeed give an account o f any
purported fact. There are three ways to give an account. (1) 'I deny it' -
this earns a failing mark if the fact is really Moorean. (2) 'I analyse it thus'
- this is Armstrong's response to the facts of apparent sameness o f type.
-
Or (3) 'I accept it as primitive'. Not every account is an analysis! A system

that takes certain Moorean facts as primitive, as unanalysed, cannot be
accused of failing to make a place for them. It neither shirks the compulsory
question nor answers it by denial. It does give an account.
An adequate Nominalism, o f course, is a theory that takes Moorean facts
o f apparent sameness of type as primitive. It predicates mutual resemblance
o f the things which are apparently of the same type; or it predicates

naturalness of some property that they all share, i.e. that has them all as
members; and it declines to analyse these predications any further. That is
why the problem of One over Many, rightly understood, does not provide
more than a prima facie reason to postulate universals. Universals afford
one solution, but there are others.
I fear that the problem does not remain rightly understood. Early in
Universals it undergoes an unfortunate double transformation. In the course
of a few pages (Universals, I, pp. 11-16) the legitimate demand for an account
o f Moorean facts o f apparent sameness o f type turns into a demand for an
analysis of predication in general. The analysandum becomes the schema
'a has the property F'. The turning point takes only two sentences (p. 12):
How is [the Nominalist] to account for the apparent (if usually partial)
identity of numerically different particulars? How can two different things
both be white or both be on a table?
And very soon (pp. 16-17) those who 'refuse to countenance universals but
who at the same time see no need for any reductive analyses [of the schema
of predication]', those according to whom 'there are no universals but the
proposition that a is F i s perfectly all right as it is' stand accused o f dodging
the compulsory question.
When the demand for an account - for a place in one's system - turned
into a demand for an analysis, then I say that the question ceased to be
compulsory. And when the analysandum switched, from Moorean facts of
apparent sameness o f type to predication generally, then I say that the
question ceased to be answerable at all. The transformed problem of One
over Many deserves our neglect. The ostrich that will not look at it is a wise
bird indeed.
Despite his words, I do not think that Armstrong really means to demand,
either from Nominalists or from himself, a fully general analysis of
predication. For none is so ready as he to insist that not just any shared
predicate makes for even apparent sameness o f type. (That is what gives his
theory its distinctive interest and merit.) It would be better to put the
transformed problem thus: one way or another, all predication is to be
David Lewis 353
analysed. Some predications are to be analysed away in terms of others. Here

we have one-off analyses for specific predicates - as it might be, for 'grue'.
But all those predications that remain, after the one-off analyses are finished,
are to be analysed wholesale by means o f a general analysis o f the schema
'a has property F ' .
There is to be no unanalysed predication. Time and again, Armstrong
wields this requirement against rival theories. One theory after another falls
victim to the 'relation regress': in the course of analysing other predications,
the theory has resort to a new predicate that cannot, on pain of circularity,
be analysed along with the rest. So falls Class Nominalism (including the
version with primitive naturalness that I deem adequate): it employs
predications o f class membership, which predications it cannot without
circularity analyse in terms of class membership. So falls Resemblance

Nominalism: it fails to analyse predications o f resemblance. So fall various
other, less deserving Nominalisms. A n d so fall rival forms of Realism, for
instance Transcendent, Platonic Realism: this time, predications of
participation evade analysis. Specific theories meet other, specific objections;
suffice it to say that I think these inconclusive against the two Nominalisms
that I called adequate. But the clincher, the one argument that recurs
throughout the m a n y refutations, is the relation regress. And this amounts
to the objection that the theory under attack does not achieve its presumed
aim of doing away with all unanalysed predication and therefore fails to solve
the transformed problem of One over Many.
Doing away with all unanalysed predication is an unattainable aim, and
so an unreasonable aim. No theory is to be faulted for failing to achieve it.
For how could there be a theory that names entities, or quantifies over them,
in the course of its sentences, and yet altogether avoids primitive predication?
Artificial tricks aside, 14 the thing cannot be done.
What's true is that a theory m a y be faulted for its overabundant primitive
predications, or for unduly mysterious ones, or for unduly complicated ones.
These are not fatal faults, however. They are to be counted against a theory,
along with its faults of overly generous ontology or o f disagreement with
less-than-Moorean commonsensical opinions. Rival philosophical theories
have their prices, which we seek to measure. But it's all too clear that for
philosophers, at least, there ain't no such thing as a free lunch.
How does Armstrong himself do without primitive predication? - He
doesn't. Consider the predicate 'instantiates' (or 'has'), as in 'particular a
14 Let S be the syntactic category of sentences, let N be the category of names, and for any
categories x and y, let x/y be the category of expressions that attach to y-expressions to
makex-expressions.Predicates, then, are categoryS/N. (Or (S/N)/Nfor two-placepredicates,
and so on.) To embed names (or variables in the category of names) into sentences without
primitive predication, take any category q which is neither S nor N, nor S/N, and let there
be primitives of categories Q/N and S/Q. Or take Q~ and Qz, different from S and N and
S/N and each other, and let the primitives be of categories Q1/N, Qz/Q~, and S/Q2. Or
. . . . I cannot see how this trickery could be a genuine alternative to, rather than a disguise
for, primitive predication.
instantiates universal F ' or 'this electron has unit charge'. No one-off analysis
applies to this specific predicate. 'Such identity in nature [as results from the
having o f one universal in many particulars] is literally inexplicable, in the
sense that it cannot be further explained.' (Universals, I, p. 109.) Neither
do predications of 'instantiates' fall under Armstrong's general analysis of
(otherwise unanalysed) predication. His is a non-relational Realism: he
declines, with good reason, to postulate a dyadic universal of instantiation
to bind particulars to their universals. (And if he did, it would only postpone
the need for primitive predication.) So let all who have felt the bite of
Armstrong's relation regress rise up and cry 'Tu quoque!' And let us mark
well that Armstrong is prepared to give one predicate 'what has been said
to be the privilege o f the harlot: power without responsibility. The predicate
is informative, it makes a vital contribution to telling us what is the case,

the world is different if it is different, yet ontologically it is supposed not
to commit us. Nice work: if you can get it.' (Compare Armstrong on Quine's
treatment of predication, "Against 'Ostrich' Nominalism", p. 443.)
Let us dump the project of getting rid of primitive predication, and return
to the sensible - though not compulsory - project of analysing Moorean
facts of apparent sameness o f type. Now does the relation regress serve
Armstrong better? I think not. It does make better sense within the more
sensible project, but it still bites Armstrong and his rivals with equal force.
Let the Nominalist say 'These donkeys resemble each other, so likewise do
those stars, and there analysis ends.' Let the Platonist say 'This statue
participates in the Form o f beauty, likewise that lecture participates in the
F o r m of truth, and there analysis ends.' L e t Armstrong say 'This electron
instantiates unit charge, likewise that proton instantiates tripartiteness, and
there analysis ends.' It is possible to complain in each case that a fact of
sameness of type has gone unanalysed, the types being respectively
resemblance, participation, and instantiation. But it is far from evident that
the alleged facts are Moorean, and still less evident that the first two are more
Moorean than the third. None of them are remotely the equals of the genuine
Moorean fact that, in some sense, different lumps of gold are the same in
kind.
Michael Devitt has denounced the One over Many problem as a mirage
better left unseen. 15 I have found Devitt's discussion instructive and I agree
with much o f what he says. But Devitt has joined Armstrong in transforming
the One over Many problem. He takes it to be the problem of analysing the
schema
a and b have the same property (are of the same type), F-ness
otherwise than by means o f a one-off analysis for some specific F. To that

problem it is fair to answer as he does that
" 'Ostrich Nominalism' or 'Mirage Realism'?", Pacific Philosophical Quarterly 61 (1980)

pp. 433-439.Devitt speaks on behalf of Quine as well as himself; Quine indicatesagreement
with Devitt in 'Soft Impeachment Disowned'.
David Lewis 355
a is F; b is F
is analysis enough, once we give over the aim of doing without primitive
predication. But Devitt has set himself too easy a problem. If we attend to
the modest, untransformed One over Many problem, which is no mirage,
we will ask about a different analysandum:
a and b have some common property (are somehow of the same type)
in which it is not said what a and b have in common. This less definite
analysandum is not covered by what Devitt has said. If we take a clearly
Moorean case, he owes us an account: either an analysis or an overt resort
to primitive predication of resemblance.
Duplication, Supervenience, and Divergent Worlds. Henceforth I shall

speak only o f my need for the distinction between natural and unnatural,
or more and less natural, properties. It is to be understood that the work
I have in store for an adequately discriminatory theory of properties might
be new work for a theory of universals, or it might instead be work for the
resources of an adequate Nominalism.
I begin with the problem of analysing duplication. We are familiar with
cases of approximate duplication, e.g. when we use copying machines. And
we understand that if these machines were more perfect than they are, the
copies they made would be perfect duplicates o f the original. Copy and
original would be alike in size and shape and chemical composition of the
ink marks and the paper, alike in temperature and magnetic alignment and
electrostatic charge, alike even in the exact arrangement of their electrons
and quarks. Such duplicates would be exactly alike, we say. They would match
perfectly, they would be qualitatively identical, they would be indiscernible.
But they would not have exactly the same properties, in my sense of the
word. As in the case of any two things, countless class boundaries would
divide them. Intrinsically, leaving out their relations to the rest of the world,
they would be just alike. But they would occupy different spatio-temporal
positions; and they might have different owners, be first examined in different
centuries, and so on.
So if we wish to analyse duplication in terms of shared properties, it seems
that we must first distinguish the intrinsic (or 'internal') properties from the
extrinsic (or 'external' or 'relational') properties. Then we may say that two
things are duplicates iff they have precisely the same intrinsic properties,
however much their extrinsic properties might differ. But our new problem
of dividing the properties into intrinsic and extrinsic is no easier than our
original problem o f analysing duplication. In fact, the two problems are
joined in a tight little circle of interdefinability. Duplication is a matter of
sharing intrinsic properties; intrinsic properties are just those properties that
never differ between duplicates. Property P is intrinsic if f, for any two
duplicate things, not necessarily from the same world, either both have P
356 N e w W o r k f o r a Theory o f Universals
o r n e i t h e r d o e s . P is e x t r i n s i c i f f t h e r e is s o m e s u c h p a i r o f d u p l i c a t e s o f which
one has P and the other lacks p.16
If we relied on our physical theory to be accurate and exhaustive, we might
think to define duplication in physical terms. We believe that duplicates must
b e a l i k e i n t h e a r r a n g e m e n t o f t h e i r e l e c t r o n s a n d q u a r k s - - w h y n o t put
t h i s f o r w a r d as a d e f i n i t i o n ? B u t s u c h a ' d e f i n i t i o n ' is n o a n a l y s i s . It
p r e s u p p o s e s t h e p h y s i c s o f o u r a c t u a l w o r l d ; h o w e v e r p h y s i c s is c o n t i n g e n t
a n d k n o w n a p o s t e r i o r i . T h e d e f i n i t i o n d o e s n o t a p p l y t o d u p l i c a t i o n at
p o s s i b l e w o r l d s w h e r e p h y s i c s is d i f f e r e n t , o r t o d u p l i c a t i o n b e t w e e n w o r l d s
t h a t d i f f e r i n t h e i r p h y s i c s . N o r d o e s it c a p t u r e w h a t t h o s e i g n o r a n t o f physics
m e a n w h e n t h e y s p e a k - as t h e y d o - o f d u p l i c a t i o n .
T h e p r o p e r c o u r s e , I s u g g e s t , is t o a n a l y s e d u p l i c a t i o n i n t e r m s o f s h a r e d
properties; but to begin not with the intrinsic properties but rather with natural
p r o p e r t i e s . T w o t h i n g s a r e q u a l i t a t i v e d u p l i c a t e s i f t h e y h a v e e x a c t l y t h e same
p e r f e c t l y n a t u r a l p r o p e r t i e s . 17
P h y s i c s is r e l e v a n t b e c a u s e it a s p i r e s t o g i v e a n i n v e n t o r y o f n a t u r a l
16 Given duplication, we can also subdivide the extrinsic properties, distinguishing pure cases
from various mixtures of extrinsic and intrinsic. Partition the things, of this and other worlds,
into equivalence classes under the relation of duplication. A property may divide an
equivalence class, may include it, or may exclude it. A property P is extrinsic, as we said,
if it divides at least some of the classes. We have four subcases. (1) P divides every class;
then we may call Ppurely extrinsic. (2) P divides some classes, includes some, and excludes
none; then P is the disjunction of an intrinsic property and a purely extrinsic property. (3)
P divides some, excludes some, and includes none; then P is the conjunction of an intrinsic
property and a purely extrinsic property. (4) P divides some, includes some, and excludes
some; then P is the conjunction of an intrinsic property and an impurely extrinsic property
of the sort considered in the second case, or equivalently is the disjunction of an intrinsic
property and an impurely extrinsic property of the sort considered in the third case.
We can also classify relations as intrinsic or extrinsic, but in two different ways. Take
a dyadic relation, i.e. a class or ordered pairs. Call the relation intrinsic to its relata iff,
whenever a and a' are duplicates (or identical) and b and b ' are duplicates (or identical),
then both or neither of the pairs < a,b > and < a ' , b ' > stand in the relation. Call the relation
intrinsic" to itspairs iff, whenever the pairs <a,b > and <a',b" > themselves are duplicates,
then both or neither of them stand in the relation. In the second case, a stronger requirement
is imposed on the pairs. For instance they might fail to be duplicate pairs because the distance
between a and b differs from the distance between a ' and b ' , even though a and a' are
duplicates and b and b ' are duplicates. In traditional terminology, 'internal relations' are
intrinsic to their relata; 'external relations' are intrinsic to their pairs but not to their relata;
and relations extrinsic even to their pairs, such as the relation of belonging to the same owner,
get left out of the classification altogether.
Our definition of intrinsic properties in terms of duplication closely resembles the definition
of 'differential properties' given by Michael Slote in 'Some Thoughts on Goodman's Riddle',
Analysis 27 (1967) pp. 128-132, and in Reason and Scepticism (George Allen & Unwin, 1970).
But where I quantify over possibilia, Slote applies modality to ordinary, presumably actualist,
quantifiers. That makes a difference. An extrinsic property might differ between duplicates,
but only when the duplicates inhabit different worlds; then Slote would count the property
as differential. An example is the property of being a sphere that inhabits a world where
there are pigs or a cube that inhabits a world without pigs.
See my 'Extrinsic Properties', Philosophical Studies 44 (1983) for further discussion of
the circle from duplication to intrinsicness and back.
17 Likewise <a,b> and <a',b" > and duplicate pairs iff a and a' have exactly the same
perfectly natural properties, and so do b and b', and also the perfectly natural relations
between a and b are exactly the same as those between a ' and b ' .
David Lewis 357
properties -- not a complete inventory, perhaps, but a complete enough

inventory to account for duplication among actual things. If physics succeeds
in this, then duplication within our world amounts to sameness o f physical
description. But the natural properties themselves are what matter, not the
theory that tells us what they are. If Materialism were false and physics an
utter failure, as is the case at some deplorable worlds, there would still be
duplication in virtue of shared natural properties.
On my analysis, all perfectly natural properties come out intrinsic. That
seems right. The converse is not true. Intrinsic properties may be disjunctive
and miscellaneous and unnatural, so long as they never differ between
duplicates. The perfectly natural properties comprise a basis for the intrinsic
properties; but arbitrary Boolean compounds of them, however unnatural,
are still intrinsic. Hence if we adopt the sort of adequate Nominalism that
draws a primitive distinction between natural and unnatural properties, that
is not the same thing as drawing a primitive distinction between intrinsic and
extrinsic properties. The former distinction yields the latter, but not vice versa.
Likewise if we adopt the sort of adequate Nominalism that begins with
a suitable relation of partial resemblance, that is not the same thing as taking
duplication itself as primitive. Again, the former yields the latter, but not
vice versa.
If instead we reject Nominalism, and we take the perfectly natural
properties to be those that correspond to universals (in the sense that the
members of the property are exactly those things that instantiate the
universal), then all the properties that correspond to universals are intrinsic.
So are all the Boolean compounds -- disjunctions, negations, etc. -- of
properties that correspond to universals. The universals themselves are
intrinsic e x officio, so to speak.
But here I must confess that the theory o f universals for which I offer new
work cannot be exactly Armstrong's theory. For it must reject extrinsic
universals; whereas Armstrong admits them, although not as irreducible. (See
Universals, II, pp. 78-79.) I think he would be better off without them, given
his own aims. (1) They subvert the desired connection between sharing of
universals and Moorean facts of partial or total sameness of nature.
Admittedly, there is such a thing as resemblance in extrinsic respects: things
can be alike in the roles they play vis-a-vis other things, or in the origins
they spring from. But such resemblances are not what we mean when we
say of two things that they are of the same kind, or have the same nature.
(2) They subvert the desired immanence of universals: if something
instantiates an extrinsic universal, that is not a fact just about that thing.
(3) They are not needed for Armstrong's theory of laws o f nature; any
supposed law connecting extrinsic universals of things can be equivalently
replaced by a law connecting intrinsic structures o f larger systems that have
those things as parts.
Thus I am content to say that if there are universals, intrinsic duplicates
are things having exactly the same universals. We need not say ' . . . exactly
the same intrinsic universals' because we should not believe in any other kind.
358 New Work f o r a Theory o f Universals
Not only is duplication of interest in its own right; it also is needed in

dealing with other topics in metaphysics. Hence such topics create a derived
need for natural properties. I shall consider two topics where I find need to
speak of duplication: supervenience and divergent worlds.
First, supervenience. A supervenience thesis is a denial of independent

variation. Given an ontology of possibilia, we can formulate such theses in
terms o f differences between possible individuals or worlds. To say that so-
and-so supervenes on such-and-such is to say that there can be no difference
in respect of so-and-so without difference in respect of such-and-such. Beauty
of statues supervenes on their shape, size, and colour, for instance, if no

two statures, in the same or different worlds, ever differ in beauty without
also differing in shape or size or colour. TM
A supervenience thesis is, in a broad sense, reductionist. But it is a stripped-
down form or reductionism, unencumbered by dubious denials o f existence,
claims of ontological priority, or claims of translatibility. One might wish
to say that in some sense the beauty of statues is nothing over and above
the shape and size and colour that beholders appreciate, but without denying
that there is such a thing as beauty, without claiming that beauty exists only
in some less-than-fundamental way, and without under taking to paraphrase
ascriptions o f beauty in terms of shape etc. A supervenience thesis seems
to capture what the cautious reductionist wishes to say.
Even if reductionists ought to be less cautious and aim for translation,
still it is a good idea to attend to the question of supervenience. For if
supervenience fails, then no scheme of translation can be correct and we
needn't go on Chisholming away in search of one. If supervenience succeeds,
on the other hand, then some correct scheme must exist; the remaining
question is whether there exists a correct scheme that is less than infinitely
complex. If beauty is supervenient on shape etc., the worst that can happen
is that an ascription of beauty is equivalent to an uncountably infinite
disjunction of maximally specific descriptions of shape etc., which descriptions
might themselves involve infinite conjunctions.
Interesting supervenience theses usually involve the notion o f qualitative
duplication that we have just considered. Thus we may ask what does or
doesn't supervene on the qualitative character of the entire world, throughout
all of history. Suppose that two possible worlds are perfect qualitative
duplicates - must they then also have exactly the same distributions of
objective probability, the same laws of nature, the same counterfactuals and
causal relations? Must their inhabitants have the same de re modal properties?
If so, it makes sense to pursue such projects as a frequency analysis of
probability, a regularity analysis of laws of nature, or a comparative similarity
~s For a generaldiscussionof supervenience,see JaegwonKim, 'Supervenienceand Nomological

Incommensurables', American Philosophical Quarterly 15 (1978) pp. 149-156.
David Lewis 359
analysis of causal counterfactuals and de re modality. If not, such projects

are doomed from the start, and we needn't look at the details of the attempts.
But we cannot even raise these questions of supervenience unless we can speak
of duplicate worlds. A n d to do that, I have suggested, we need natural
properties.
(Note that if possible worlds obey a principle o f identity of qualitative
indiscernibles, then all these supervenience theses hold automatically. If no
two worlds are duplicates, then afortiori no two are duplicates that differ
in their probabilities, laws . . . . . or anything else.)
We might also ask whether qualitative character supervenes on anything
less. For instance, we might ask whether global qualitative character
supervenes on local qualitative character. Say that two worlds are local
duplicates iff they are divisible into corresponding small parts in such a way
that (1) corresponding parts of the two worlds are duplicates, and (2) the
correspondence preserves spatiotemporal relations. (The exact meaning
depends, of course, on what we mean by 'small'.) I f two worlds are local
duplicates, then must they be duplicates simpliciter? Or could they differ in
ways that do not prevent local duplication - e.g. in external relations, other
than the spatiotemporal relations themselves, between separated things?
Again, we must make sense of duplication - this time, both in the large
and in the small - even to ask the question. 19
Next, divergent worlds. I shall say that two possible worlds diverge iff they
are not duplicates but they do have duplicate initial temporal segments. Thus
our world and another might match perfectly up through the year 1945, and
go their separate ways thereafter.
Note that we need no identity of times across worlds. Our world through
our 1945 duplicates an initial segment o f the other world; that otherworldly
segment ends with a year that indeed resembles our 1945, but it is part of
otherworldly time, not part of our time. Also, we need no separation of time
and space that contravenes Relativity - we have initial temporal segments,
of this or another world, if we have spatiotemporal regions bounded by
spacelike surfaces that cut the world in two.
I distinguish divergence of worlds from branching of worlds. In branching,
instead of duplicate segments, one and the same initial segment is allegedly
shared as a c o m m o n part by two overlapping worlds. Branching is
problematic in ways that divergence is not. First, because an inhabitant of
the shared segment cannot speak unequivocally of the world he lives in. What
if he says there will be a sea fight t o m o r r o w , meaning of course to speak
of the future of his own world, and one o f the two worlds he lives in has
19 Sucha thesis of supervenienceof the global on the local resemblesthe 'holographichypothesis'

considered and rejected by Saul Kripke in 'Identity Through Time', presented at the 1979
conference of the American Philosophical Association, Eastern Division, and elsewhere.
a sea fight the next day and the other doesn't? Second, because overlap of
worlds interferes with the most salient principle of demarcation for worlds,
viz. that two possible individuals are part of the same world iff they are linked
by some chain of external relations, e.g. of spatiotemporal relations. (I know
of no other example.) Neither of these difficulties seems insuperable, but both
are better avoided. That makes it reasonable to prefer a theory of
nonoverlapping divergent worlds to a theory of branching worlds. Then we
need to be able to speak of qualitative duplication of world-segments, which
we can do in terms of shared natural properties.
Divergent (or branching) worlds are of use in defining Determinism. The
usual definitions are not very satisfactory. If we say that every event has a
cause, we overlook probabilistic causation under Indeterminism. If we speak
of what could be predicted by a superhuman calculator with unlimited

knowledge of history and the laws of nature, we overlook obstacles that might
prevent prediction even under Determinism, or else we try to make
nonvacuous sense o f counterfactuals about what our predictor could do if
he had some quite impossible combination of powers and limitations.
A better approach is as follows. First, a system of laws of nature is
Deterministic iff no two divergent worlds both conform perfectly to the laws
of that system. Second, a world is Deterministic iff its laws comprise a
Deterministic system. Third, Determinism is the thesis that our world is
Deterministic. E°
(Alternative versions of Determinism can be defined in similar fashion.
For instance, we could strengthen the first step by prohibiting convergence
as well as divergence of law-abiding worlds. Or we could even require that
no two law-abiding worlds have duplicate momentary slices without being
duplicates throughout their histories. Or we could define a weaker sort of
Determinism: we could call a world f o r t u i t o u s l y Deterministic, even if its
laws do not comprise a Deterministic system, iff no world both diverges from
it and conforms to its laws. The laws and early history of such a world suffice
to determine later history, but only because the situations in which the laws
fall short of Determinism never arise. We might equivalently defirae fortuitous
Determinism as follows: for any historical fact F and any initial segment S
of the world, there are a true proposition H about the history of S-and a
true proposition L about the laws of nature, such that H and L together
strictly imply F. 21 Does this definition bypass our need to speak of duplication
of initial segments? Not so, for we must ask what it means to say that H
20 This approachis due, in essence, to Richard Montague,'DeterministicTheories',in Decisions,

Values and Groups, II (PergamonPress, 1962),and in his Formal Philosophy (Yale University
Press, 1974). But Montague did not speak as I have done of duplication of initial segments
of worlds in virtue of the sharing of certain 61iteproperties. Instead, he used sameness of
description in a certain vocabulary, which vocabularywas left as an unspecified parameter
of his analysis. For he wrote as a logician obliged to remain neutral on questions of
metaphysics.
21 A closelyrelated definition appears in Peter van Inwagen, 'The Incompatibilityof Free Will
and Determinism', Philosophical Studies 27 (1975) pp. 185-199.
David Lewis 361
is about the history of S. I take that to mean that H holds at both or neither
of any two worlds that both begin with segments that are duplicates o f S.)
Divergent worlds are important also in connection with the sort of
counterfactual conditional that figures in patterns of causal dependence. Such
counterfactuals tend to be temporally asymmetric, and this is what gives rise
to the a s y m m e t r y of causation itself. Counterfactuals o f this sort do not
'backtrack': it is not to be said that if the present were different a different
past would have led up to it, but rather that if the present were different,
the same past would have had a different outcome. Given a hypothesised
difference at a certain time, the events of future times normally would be
very different indeed, but the events of past times (except perhaps for the
very near past) would be no different. Thus actuality and its counterfactual
alternatives are divergent worlds, with duplicate initial segments, z2
M i n i m a l Materialism. There is a difficulty that arises if we attempt to

formulate certain reductionist views, for instance Materialism, as
supervenience theses. A solution to this difficulty employs natural properties
not only by way of duplication but in a more direct way also.
Roughly speaking, Materialism is the thesis that physics - something not
too different f r o m present-day physics, though presumably somewhat
improved - is a comprehensive theory of the world, complete as well as
correct. The world is as physics says it is, and there's no more to say. World
history written in physical language is all of world history. That is rough
speaking indeed; our goal will be to give a better formulation. But before
I try to say more precisely what Materialism is, let me say what it is not.
(1) Materialism is not a thesis of finite translatibility of all our language into
the language of physics. (2) Materialism is not to be identified with any one
Materialist theory of mind. It is a thesis that motivates a variety of theories
of mind: versions of Behaviourism, Functionalism, the mind-body identity
theory, even the theory that mind is all a mistake. (3) Materialism is not just
the theory that there are no things except those recognised by physics. To
be sure, Materialists don't believe in spirits, or other such nonphysical things.
But antimaterialists m a y not believe in spirits either - their complaint needn't
be that physics omits some of the things that there are. They m a y complain
instead that physics overlooks some o f the ways there are for physical things
to differ; for instance, they m a y think that physical people could differ in
what their experience is like. (4) That suggests that Materialism is, at least
in part, the thesis that there are no natural properties instantiated at our world
22 See my 'CounterfactualDependenceand Time'sArrow', Nofts 13 (1979)pp. 455-476; Jonathan

Bennett'sreviewof my Counterfactuals, CanadianJournal of Philosophy 4 (t974) pp. 381-402;
P. B. Downing, 'Subjunctive Conditionals, Time Order, and Causation', Proceedings of
the Aristotelian Society 59 (1959) pp. 125-140; Allan Gibbard and William Harper,
'Counterfactuals and Two Kinds of Expected Utility', in C. A. Hooker, J. T. Leach, and
E. F. McClennen, eds., Foundations and Applications of Decision Theory (Reidel, 1978),
and in W. L. Harper, R. Stalnaker, and G. Pearce, eds., Ifs(Reidel, 1981); and Frank Jackson,
'A Causal Theory of Counterfactuals',Australasian Journal of Philosophy 55 (1977)pp. 3-21.
except those recognised by physics. That is better, but I think still not right.
Couldn't there be a natural property X (in the nature of the case, it is hard
to name an exampled which is shared by the physical brains in worlds like
ours and the immaterial spirits that inhabit other worlds? Or by thisworldly
quarks and certain otherworldly particles that cannot exist under our physics?
Physics could quite properly make no mention of a natural property of this
sort. It is enough to recognise the special case applicable to our world, X-
cum-physicality, brainhood or quarkhood as it might be. Then if by physical
properties we mean those properties that are mentioned in the language of
physics, a Materialist ought not to hold that all natural properties instantiated
in our world are physical properties.
At this point, it ought to seem advisable to formulate Materialism as a
supervenience thesis: no difference without physical difference. Or,

contraposing: physical duplicates are duplicates simpliciter. Afortiori, no
mental difference without physical difference; physical duplicates are mental
duplicates. The thesis might best be taken as applying to whole possible
worlds, in order to bypass such questions as whether mental life is to some
extent extrinsic to the subject. So we have this first of several attempted
formulations of Materialism:
M1. Any two possible worlds that are exactly alike in all respects
recognised by physics are qualitative duplicates.
But this will not do. In making Materialism into a thesis about how just any
two worlds can and cannot differ, M1 puts Materialism forward as a necessary
truth. That is not what Materialists intend. Materialism is meant to be a
contingent thesis, a merit of our world that not all other worlds share. Two
worlds could indeed differ without differing physically, if a least one of them
is a world where Materialism is false. For instance, our Materialistic world
differs from a nonmaterialistic world that is physically just like ours but that
also contains physically epiphenomenal spirits.
There is a noncontingent supervenience thesis nearby that might appeal
to Materialists:
M2. There is no difference, afortiori no mental difference, without some
nonmental difference. Any two worlds alike in all nonmental respects
are duplicates, and in particular do not differ in respect of the mental
lives of their inhabitants.
This seems to capture our thought that the mental is a pattern in a medium,
obtaining in virtue of local features o f the medium (neuron firings) and
perhaps also very global features (laws of nature) that are too small or too
big to be mental themselves. But M2 is not Materialism. It is both less and
more. Less, obviously, because it never says that the medium is physical.
More, because it denies the very possibility of what I shall call Panpsychistic
Materialism.
It is often noted that psychophysical identity is a two-way street: if all
David Lewis 363
mental properties are physical, then some physical properties are mental. But
perhaps not just some but all physical properties might be mental as well;
and indeed every property of anything might be at once physical and mental.
Suppose there are indeed worlds where this is so. I f so, presumably there
are m a n y such worlds, not all duplicates, differing inter alia in the mental
lives of their inhabitants. But all differences between such worlds are mental
(as well as physical), so none are nonmental. These worlds will be vacuously
alike in all nonmental respects, for lack of any nonmental respects to differ
in. Then M 2 fails. A n d not just at the troublemaking worlds; M 2 is
noncontingent, so if it fails at any worlds, it fails at all - even decent
Materialistic worlds like ours. Maybe Panpsychistic Materialism is indeed
impossible - how do you square it with a broadly functional analysis of
mind? - but a thesis that says so is more than just Materialism.

A third try. This much is at least true:
M3. No two Materialistic worlds differ without differing physically; any
two Materialistic worlds that are exactly alike physically are
duplicates.
But M3 is not a formulation of Materialism, for the distinction between

Materialistic and other worlds appears within M3. All we learn is that the
Materialistic worlds comprise a class within which there is no difference
without physical difference. But there are m a n y such classes. In fact any
world, however spirit-ridden, belongs to such a class.
A fourth try. Perhaps we should confine our attention to nomologically
possible worlds, thus:
M4. A m o n g worlds that c o n f o r m to the actual laws o f nature, no two
differ without differing physically; any two such worlds that are
exactly alike physically are duplicates.
But again we have something that is both less and m o r e than Materialism.
Less, because M4 could hold at a world where Materialism is false but where
spiritual phenomen~ are correlated with physical p h e n o m e n a according to
strict laws. More, because M4 fails to hold at a Materialistic, spirit-free world
if the laws of that world do not preclude the existence of epiphenomenal
spirits. Our world might be such a world, a world where spirits are absent
but not outlawed. 23
So far, a supervenience formulation of Materialism seems elusive. But I
think we can succeed if we join the idea o f supervenience with the idea that
a nonmaterialistic world would have somthing extra, something that a
Materialistic world lacks. It might have spirits; or it might have physical things
that differ in nonphysical ways, for instance in what their experience is like.
2~ Thisobjection against M4 as a formulation of 'the ontologicalprimacy of the microphysical'

appearsin TerenceHorgan, 'Supervenienceand Microphysics',PacificPhilosophical Quarterly
63 (1982) pp. 29-43.
In either case there are extra natural properties, properties instantiated in

the nonmaterialistic world but nowhere to be found in the Materialistic world.
Let us say that a property is alien to a world iff (1) it is not instantiated by
any inhabitant o f that world, and (2) it is not analysable as a conjunction
of, or as a structural property constructed out of, natural properties all of
which are instantiated by inhabitants of that world. (I need the second clause
because I am following Armstrong, mutatis mutandis, in declining to rule
out perfectly natural properties that are conjunctive or structurally complex.
See Universals, II, pp. 30-42 and 67-71. It would be wrong to count as alien
a complex property analysable in terms of nonalien constituents.) If our world
is Materialistic, then it is safe to say that some of the natural properties
instantiated in any nonmaterialistic world are properties alien to our world.
Now we can proceed at last to formulate Materialism as a restricted and

contingent supervenience thesis:
M5. Among worlds where no natural properties alien to our world are
instantiated, no two differ without differing physically; any two such
worlds that are exactly alike physically are duplicates. 24
We took Materialism to uphold the comprehensiveness of 'something not

too different from present-day physics, though presumably somewhat
improved'. That was deliberately vague. Materialist metaphysicians want to
side with physics, but not to take sides within physics. Within physics, more
precise claims o f completeness and correctness may be at issue. Physics
(ignoring latter-day failures of nerve) is the science that aspires to
comprehensiveness, and particular physical theories may be put forward as
fulfilling that aspiration. If so, we must again ask what it means to claim
comprehensiveness. And again, the answer may be given by a supervenience
formulation: no difference without physical difference as conceived by such-
and-such grand theory. But again it must be understood as a restricted and
contingent supervenience thesis, applying only among worlds devoid of alien
natural properties.
Thus the businesss of physics is not just to discover laws and causal
explanations. In putting forward as comprehensive theories that recognise
only a limited range o f natural properties, physics proposes inventories of
the natural properties instantiated in our world. Not complete inventories,
24 This formulation resembles one proposed by Horgan, op. cir. The principal difference is
as follows. Horgan would count as alien (my term, not his) any property cited in the
fundamental laws of otherworldly microphysics that is not also explicitly cited in the
fundamental laws o f this worldly microphysics. Whether the property is instantiated in either
world doesn't enter into it. But must an alien property figure in laws of otherworldly physics?
Must it figure in any otherworldly laws at all? It seems that a Materialistic world might differ
without differing physically from a world where there are properties alien in my sense but
not in Horgan's -- perhaps a world where laws are in short supply.
David Lewis 365
perhaps. But complete enough to account for all the duplications and
differences that could arise in the absence of alien natural properties. O f
course, the discovery o f natural properties is inseparable f r o m the discovery
of laws. For an excellent reason to think that some hitherto unsuspected
natural properties are instantiated - properties deserving of recognition by
physics, the quark colours as they might be -- is that without them, no
satisfactory system of laws can be found.
This is reminiscent of the distinctive a posteriori, scientific character of
Armstrong's Realism (Universals, I, pp. 8-9, and passim). But in the setting
of an ontology of possibilia, the distinction between discovering what
universals or natural properties there actually are and discovering which ones
are actually instantiated fades away. And the latter question is a posteriori
on any theory. What remains, and remains important, is that physics discovers
properties. And not just any properties - natural properties. The discovery
is, for instance, that neutrinos are not all alike. That is not the discovery
that different ones have different properties in m y sense, belong to different
classes. We knew that much a priori. Rather, it is the surprising discovery
that some natural property differentiates some neutrinos f r o m others. That
discovery has in fact been made; I should like to read an account o f it by
some philosopher who is not prepared to adopt a discriminatory attitude
toward properties and who thinks that all things are equally similar and
dissimilar to one another.
Laws and Causation. The observation that physics discovers natural

properties in the course of discovering laws m a y serve to introduce our next
topic: the analysis o f what it is to be a law of nature. I agree with Armstrong
that we need universals, or at least natural properties, in explaining what
lawhood is, though I disagree with his account of how this is so.
Armstrong's theory, in its simplest form, 25 holds that what makes certain
regularities lawful are second-order states of affairs N(F, G) in which the two
ordinary, first-order universals F and G are related by a certain dyadic second-
order universal N. It is a contingent matter which universals are thus related
by the lawnmaker N. But it is necessary -- and necessary simpliciter, not
just nomologically necessary - that if N(F,G) obtains, then F and G
are constantly conjoined. There is a necessary connection between the
second-order state of affairs N(F,G) and the first-order lawful regularity
Vx(FxD Gx); and likewise between the conjunctive state of affairs N(F,G)
& Fa and its necessary consequence Ga.
A parallel theory could be set up with natural properties in place o f
Armstrong's first- and second-order universals. It would have m a n y of the
attractive features that Armstrong claims on behalf of his theory, but at least
Universals, II, pp. 148-157. A more developed form of the theory appears in D. M.
Armstrong, What Is a Law of Nature? (CambridgeUniversityPress, 1983). Similar theories
have been proposed in Fred I. Dretske, 'Laws of Nature', Philosophy of Science 44 (1977)
pp. 248-268, and in Michael Tooley, 'The Nature of Laws', CanadianJournal of Philosophy
4 (1977) pp. 667-698.
one merit would be lost. F o r A r m s t r o n g , the lawful necessitation o f Ga by

F a is a purely local matter: it involves only a, the universals F and G that
are present in a, and the second-order lawmaking universal t h a t is present
in t u r n in (or between) these two universals. I f we replace the universals by
properties, however natural, that locality is lost. F o r properties are classes
with their membership spread a r o u n d the worlds, and are not wholly present
in a. But I do not think this a conclusive objection, for o u r intuitions of
locality o f t e n seem to lead us astray. The selective regularity t h e o r y I shall
shortly advocate also sacrifices locality, as does any regularity theory o f law.
W h a t leads me (with some regret) to reject A r m s t r o n g ' s theory, whether
with universals or with natural properties, is that I find its necessary
connections unintelligible. W h a t e v e r N m a y be, I c a n n o t see h o w it could
be absolutely impossible to have N ( F , G ) a n d Fa w i t h o u t Ga. (Unless N just

is constant conjunction, or constant conjunction plus something else, in which
case Armstrong's theory turns into a f o r m o f the regularity theory he rejects.)
The mystery is s o m e w h a t hidden by A r m s t r o n g ' s terminology. H e uses
'necessitates' as a n a m e for the l a w m a k i n g universal N; and w h o would be
s u r p r i s e d ' t o hear that if F 'necessitates' G a n d a has F, then a m u s t have
G? But I say that N deserves the n a m e o f 'necessitation' only if, somehow,
it really can enter into the requisite necessary connections. It can't enter into
t h e m just by bearing a name, a n y m o r e t h a n one can have m i g h t y biceps
just b y being called ' A r m s t r o n g ' .
I a m t e m p t e d to c o m p l a i n in H u m e a n fashion o f alleged necessary
connections between distinct existences, especially when first-order states of
affairs in the past supposedly join with second-order states o f affairs to
necessitate first-order states o f affairs in the future. T h a t complaint is not
clearly right: the sharing o f Universals detracts f r o m the distinctness o f the
necessitating and the necessitated states o f affairs. But I a m n o t appeased.
I conclude that necessary connections can be unintelligible even when they
are supposed to obtain between existences that are n o t clearly and wholly
distinct. 26
T h u s I do not endorse A r m s t r o n g ' s way o f building universals, or
alternatively natural properties, into the analysis o f lawhood. Instead I favour
a regularity analysis. But I need natural properties even so.
Certainly n o t just any regularity is a law o f nature. S o m e are accidental.
26
Armstrong's more developed theory in What Is a Law of Nature? complicates the picture
in two ways. First, the second order state of affairs N(F,G) is itself taken to be a universal,
and its presence in its instances detracts yet further from the distinctness of the necessitating
and the necessitated states of affairs. Second, all laws are defeasible, It is possible after all
to have N(F,G) and Fa without Ga, namely if we also have N(E&F,H) and Ea, where H
and G are incompatible. (The law that F's are G's might be contingently indeafeasible, if
no such defeating state of affairs N(E&F,H) obtains; but no law has its indefeasibility built
in essentially.) It remains true that there are alleged necessary connections that I find
unintelligible, but they are more complicated than before. To necessitate a state of affairs,
we need not only the first- and second-order states of affairs originally considered, but also
a negative existential to the effect that there are no further states of affairs of the sort that
could act as defeaters.
David Lewis 367
So an adequate regularity analysis must be selective. Also, an adequate

analysis must be collective. It must treat regularities not one at a time, but
rather as candidates to enter into integrated systems. For a given regularity
might hold either as a law or accidentally, depending on whether other
regularities obtain that can fit together with it in a suitable system. (Thus
I reject the idea that lawhood consists of 'lawlikeness' plus truth.) Following
Mill and Ramsey, 27 I take a suitable system to be one that has the virtues
we aspire to in our own theory-building, and that has them to the greatest
extent possible given the way the world is. It must be entirely true; it must
be closed under strict implication; it must be as simple in axiomatisation as
it can be without sacrificing too much information content; and it must have
as much information content as it can have without sacrificing too much
simplicity. A law is any regularity that earns inclusion in the ideal system.
(Or, in case of ties, in every ideal system.) The ideal system need not consist
entirely of regularities; particular facts m a y gain entry if they contribute
enough to collective simplicity and strength. (For instance, certain particular
facts about the Big Bang might be strong candidates.) But only the regularities
of the system are to count as laws.
We face an obvious problem. Different ways to express the same content,
using different vocabulary, will differ in simplicity. The problem can be put
in two ways, depending on whether we take our systems as consisting of
propositions (classes of worlds) or as consisting of interpreted sentences. In
the first case, the problem is that a single system has different degrees of
simplicity relative to different linguistic formulations. In the second case,
the problem is that equivalent systems, strictly implying the very same
regularites, m a y differ in their simplicity. In fact, the content of any system
whatever m a y be formulated very simply indeed. Given system S, let F be
a predicate that applies to all and only things at worlds where S holds. Take
Fas primitive, and axiomatise S (or an equivalent thereof) by the single axiom
YxFx. I f utter simplicity is so easily attained, the ideal theory m a y as well
be as strong as possible. Simplicity and strength needn't be traded off. Then
the ideal theory will include (its simple axiom will strictly imply) all truths,
and a f o r t i o r i all regularities. Then, after all, every regularity will be a law.
That must be wrong.
The remedy, of course, is not to tolerate such a perverse choice of primitive
vocabulary. We should ask how candidate systems compare in simplicity when
each is formulated in the simplest eligible way; or, if we count different
formulations as different systems, we should dismiss the ineligible ones from
candidacy. An appropriate standard of eligibility not far to seek: let the
John Stuart Mill, A System of Logic (Parker, 1843) Book III, Chapter IV, Section 1;
F. P. Ramsey, 'Universals of Law and of Fact', in his Foundations (Roufledge & Kegan
Paul, 1978). Ramseyregarded this theory of law as superseded by the different theory in
his 'General Propositions and Causality', also in Foundations, but I prefer his first thoughts
to his second. 1 present a theory of lawhood along the lines of Ramsey's earlier theory in
my Counterfactuals (Blackwell, 1973) pp. 73-75. A revision to that discussion is needed in
the probabilistic case, which I here ignore.
primitive vocabulary that appears in the axioms refer only to perfectly natural
properties.
Of course, it remains an unsolved and difficult problem to say what
simplicity of a formulation is. But it is no longer the downright insoluble
problem that it would be if there were nothing to choose between alternative
primitive vocabularies.
(One might think also to replace strict implication by deducibility in some
specified calculus. But this second remedy seems unnecessary given the first,
and seems incapable of solving our problem by itself.)
If we adopt the remedy proposed, it will have the consequence that laws
will tend to be regularities involving natural properties. Fundamental laws,
those that the ideal system takes as axiomatic, must concern perfectly natural
properties. Derived laws that follow fairly straightforwardly also will tend
to concern fairly natural properties. Regularities concerning unnatural
properties may indeed be strictly implied, and should count as derived laws
if so. But they are apt to escape notice even if we someday possess a good
approximation to the ideal system. For they will be hard to express in a
language that has words mostly for not-too-unnatural properties, as any
language must. (See the next section.) And they will be hard to derive, indeed
they may not be finitely derivable at all, in our deductive calculi. Thus my
account explains, as Armstrong's does in its very different way, why the
scientific investigation of laws and of natural properties is a package deal;
why physicists posit natural properties such as the quark colours in order
to posit the laws in which those properties figure, so that laws and natural
properties get discovered together.
If the analysis of lawhood requires natural properties, then so does the

analysis of causation. It is fairly uncontroversial that causation involves laws.
That is so according to both of the leading theories of causation: the
deductive-nomological anaiysis, on which the laws are applied to the actual
course of events with the cause and effect present; and the counterfactual
analysis that I favour, on which the laws are applied to counterfactual
situations with the cause hypothesised away. These counterfactual alternatives
may need to break actual laws at the point where they diverge from actuality,
but the analysis requires that they evolve thereafter in accordance with the
actual laws. 2s
According to my counterfactual analysis, causation involves natural
properties in a second way too. We need the kind of counterfactuals that
avoid backtracking; else the analysis faces fatal counterexamples involving
epiphenomenai side-effects or cases of causal preemption. As I have already
noted, these counterfactuals are to be characterised in terms of divergent
28 See my'Causation',Journal of Philosophy 70 (1973)pp. 556-567; reprintedin Ernest Sosa,

ed., Causation and Conditionals (Oxford UniversityPress, 1975).
David Lewis 369
worlds, hence in terms of duplicate initial world-segments, hence in terms

of shared natural properties.
Causation involves natural properties in yet another way. (Small wonder
that I came to appreciate natural properties after working on the analysis
of causation!) Causation holds between events. Unless we distinguish genuine
from spurious events, we will be left with too many putative causes. You
put a lump of butter on a skillet, and the butter melts. What event causes
this? There is one event that we can call a moving of molecules. It occurs
in the region where the skillet is, just before the butter melts. This is an event
such that, necessarily, it occurs in a spatiotemporal region only if that region
contains rapidly moving molecules. Surely this event is a cause of the melting
of the butter.
Heat is that phenomenon, whatever it may be, that manifests itself in

certain familiar characteristic ways. Let us say: heat is that which occupies
the heat-role. (It won't matter whether we take the definite description plain,
as I prefer, or rigidified.) In fact, but contingently, it is molecular motion
that occupies the heat-role. It might have been molecular nonmotion, or
caloric fluid, or what you will. Now consider an alleged second event, one
that we may call a having-the-occupant-of-the-heat-role. This second event
occurs just when and where the first does, in the region where the hot skillet
is. It occurs there in virtue of the two facts (1) that the skillet's molecules
are moving rapidly, and (2) that the region in question is part of a world
where molecular motion is what occupies the heat-role. But this second event
differs from the first. The necessary conditions for its occurrence are different.
Necessarily, it occurs in a region only if that region contains whatever
phenomenon occupies the heat-role in the world o f which that region is part.
So in those worlds where caloric fluid occupies the heat-role and molecular
motion does not, the first event occurs only in regions with molecular motion
whereas the second occurs only in regions with caloric fluid.
Certainly the first event causes the melting of the butter, but shall we say
that the second event does so as well? No; that seems to multiply causes
beyond belief by playing a verbal trick. But if there really are two events
here, I cannot see why the second has less of a claim than the first to be a
cause of the melting of the butter. It is out of the question to say that the
first and the second events are one and the same - then this one event would
have different conditions of occurrence from itself. The best solution is to
deny that the alleged second event is a genuine event at all. If it isn't, o f course
it can't do any causing.
Why is the first event genuine and the second spurious? Compare the
properties involved: containing rapidly moving molecules versus containing
whatever phenomenon occupies the heat-role. (I mean these as properties
of the spatiotemporal region; other treatments of events would take instead
the corresponding properties o f the skillet, but my point would still apply.)
The first is a fairly natural, intrinsic property. The second is highly disjunctive
and extrinsic. For all sorts of different phenomena could occupy the heat-
role; and whether the phenomenon going on in a region occupies the role
370 New Work for a Theory of Universab
d e p e n d s n o t o n l y o n w h a t goes o n in t h e r e g i o n b u t also o n w h a t goes on

elsewhere in the s a m e w o r l d . T h u s t h e d i s t i n c t i o n b e t w e e n m o r e a n d less
n a t u r a l p r o p e r t i e s gives m e t h e d i s t i n c t i o n b e t w e e n g e n u i n e a n d spurious
events t h a t I need in o r d e r t o d i s o w n a n o v e r a b u n d a n c e o f causes. If a
p r o p e r t y is t o o u n n a t u r a l , it is inefficacious in the sense t h a t it c a n n o t figure
in the c o n d i t i o n s o f o c c u r r e n c e o f t h e events t h a t cause things. 29
The Content o f Language and Thought. H i l a r y P u t n a m has given an

a r g u m e n t which he r e g a r d s as a r e f u t a t i o n o f a ' r a d i c a l l y n o n - e p i s t e m i c ' view
o f t r u t h , b u t which I r e g a r d r a t h e r as a reductio against P u t n a m ' s premises. 3°
I n p a r t i c u l a r , it refutes his a s s u m p t i o n t h a t 'we i n t e r p r e t o u r l a n g u a g e s or
n o t h i n g d o e s ' ( ' M o d e l s a n d R e a l i t y ' , p. 482) so t h a t a n y c o n s t r a i n t on
reference m u s t be established b y o u r o w n s t i p u l a t i o n in l a n g u a g e o r thought.

G a r y M e r r i l l has suggested t h a t P u t n a m m a y b e a n s w e r e d b y a p p e a l to a
c o n s t r a i n t t h a t d e p e n d s o n a n o b j e c t i v e s t r u c t u r e o f p r o p e r t i e s a n d relations
in t h e w o r l d . 31 I agree, a n d find here a n o t h e r p o i n t at which we need natural
properties.
P u t n a m ' s a r g u m e n t , as I u n d e r s t a n d it, is as follows. First, s u p p o s e that
the o n l y c o n t r a i n t o n i n t e r p r e t a t i o n o f o u r l a n g u a g e (or p e r h a p s o u r language
o f t h o u g h t ) is given b y a d e s c r i p t i o n t h e o r y o f reference o f a g l o b a l and
futuristic sort. A n ' i n t e n d e d i n t e r p r e t a t i o n ' is a n y i n t e r p r e t a t i o n t h a t satisfies
a certain b o d y o f t h e o r y : viz. t h e idealised d e s c e n d a n t o f o u r c u r r e n t total
t h e o r y t h a t w o u l d e m e r g e at t h e e n d o f i n q u i r y , a n ideal t h e o r y refined to
p e r f e c t i o n u n d e r the g u i d a n c e o f all n e e d e d o b s e r v a t i o n a n d o u r best
t h e o r e t i c a l r e a s o n i n g . I f so, i n t e n d e d i n t e r p r e t a t i o n s are surprisingly
a b u n d a n t . F o r any w o r l d c a n satisfy any t h e o r y (ideal o r not), a n d can do
so in countless v e r y d i f f e r e n t ways, p r o v i d e d o n l y t h a t the w o r l d is n o t too
s m a l l a n d t h e t h e o r y is consistent. B e y o n d t h a t , it d o e s n ' t m a t t e r w h a t the
w o r l d is like o r w h a t t h e t h e o r y says. H e n c e we have r a d i c a l indeterminacy
o f reference. A n d we h a v e the c o i n c i d e n c e t h a t P u t n a m w e l c o m e s between
s a t i s f a c t i o n u n d e r all i n t e n d e d i n t r e p r e t a t i o n s a n d ' e p i s t e m i c t r u t h ' . F o r the
ideal t h e o r y is t h e w h o l e o f 'epistemic t r u t h ' , t h e i n t e n d e d i n t e r p r e t a t i o n s are
j u s t t h o s e i n t e r p r e t a t i o n s o f o u r l a n g u a g e t h a t satisfy t h e ideal t h e o r y , and
z9 See the discussion of impotence of dispositions in Elizabeth W. Prior, Robert Pargetter,

and Frank Jackson, 'Three Theses About Dispositions', American Philosophical Quarterly
19 (1982) pp. 251-257. If a disposition is not identified with its actual basis, there is a threat
of multiplication of putative causes similar to that in my example. We would not wish to
say that the breaking of a struck glass is caused both by its fragility and by the frozen-in
stresses that are the basis thereof; and if forced to choose, we should choose the latter. I
suggest that the fragility is inefficacious because it is too unnatural a property, too disjunctive
and extrinsic, to figure in the conditions of occurence of any event.
3o Hilary Putnam, 'Realism and Reason', in his Meaning and the Moral Sciences (Routledge
& Kegan Paul, 1978), and 'Models and Reality', Journal of Symbolic Logic 45 (1980)
pp. 464-482. The reader is warned that the argument as I present it may not be quite as
Putnam intended it to be. For I have made free in reading between the lines and in restating
the argument in my own way.
31 G. H. Merrill, 'The Model-Theoretic Argument Against Realism', Philosophy of Science
47 (1980) pp. 69-81.
David Lewis 371
(unless the world is too small or ideal theory is inconsistent) there are some
such interpretations.
I take this to refute the supposition that there are no further constraints
on reference. But P u t n a m asks: how c o u l d there be a further constraint?
How could we ever establish it? By stipulation, by saying or thinking
something. But whatever we say or think will be in language (or language
of thought) that suffers f r o m radical indeterminacy of interpretation. For
the saving constraint will not be there until we succeed in establishing it. So
the attempted stipulation must fail. The most we can do is to contribute a
new chapter to current and ideal theory, a chapter consisting o f whatever
we said or thought in our stipulation. And this new theory goes the way of
all theory. So we cannot establish a further constraint; and 'we interpret our
language or nothing does'; so there cannot be any further constraint. We

cannot lift ourselves by our bootstraps, so we must still be on the ground.
Indeed we cannot lift ourselves by our bootstraps, but we are o f f the
ground, so there must be another way to fly. Our language does have a fairly
determinate interpretation (a Moorean fact!) so there must be some constraint
not created ex nihilo by our stipulation.
What can it be? M a n y philosophers would suggest that it is some sort of
causal constraint. I f so m y case is made, given my arguments in the previous
section: we need natural properties to explain determinacy of interpretation.
But I doubt that it really is a causal constraint, for I am inclined to think
that the causal aspect of reference is established by what we say and think.
Thus: I think o f a thing as that which I am causally acquainted with in such-
and-such way, perhaps perceptually or perhaps through a channel of
acquaintance that involves the naming o f the thing and m y picking up of
the name. I refer to that thing in m y thought, and derivatively in language,
because it is the thing that fits this causal and egocentric description extracted
from m y theory o f the world and of my place in the world. 32
I would instead propose that the saving constraint concerns the referent
- not the referrer, and not the causal channels between the two. It takes
two to make a reference, and we will not find the constraint if we look for
it always on the wrong side o f the relationship. Reference consists in part
of what we do in language or thought when we refer, but in part it consists
in eligibility o f the referent. A n d this eligibility to be referred to is a matter
of natural properties.
That is the suggestion Merrill offers. (He offers it not as his own view,
but as what opponents of P u t n a m ought to say; and I gratefully accept the
offer.) In the simplest case, suppose that the interpretation of the logical
vocabulary somehow takes care of itself, to reveal a standard first-order
language whose nonlogical vocabulary consists entirely o f predicates. The
parts o f the world comprise a domain; and sets, sets o f pairs . . . . . f r o m this
domain are potential extensions for the predicates. Now suppose we have
32 See Stephen Schiffer, 'The Basis of Reference', Erkenntnis 13 (1978) pp. 171-206.
an all-or-nothing division o f properties into natural and unnatural. Say that

a set from the domain is eligible to be the extension of a one-place predicate
iff its members are just those things in the domain that share some natural
property; and likewise for many-place predicates and natural relations. An
eligible interpretation is one that assigns none but eligible extensions to the
predicates. A so-called 'intended' interpretation is an eligible interpretation
that satisfies the ideal theory. (But the name is misleading: it is not to be
said that our intentions establish the constraint requiring eligibility. That way
lies the futile bootstrap-tugging that we must avoid.) Then if the natural
properties are sparse, there is no reason to expect any overabundance of
intended interpretations. There may even be none. Even ideal theory runs
the risk of beings unsatisfiable, save in 'unintended' ways. Because satisfaction
is not guaranteed, we accomplish something if we manage to achieve it by

making a good fit between theory and the world. All this is as it should be.
The proposal calls for refinement. First, we need to provide for richer forms
of language. In this we can be guided by familiar translations, for instance
between modal language with higher-order quantification and first-order
language that explicitly mentions possibilia and classes built up from them.
Second, it will not do to take naturalness of properties as all-or-nothing. Here,
above all, we need to make naturalness - and hence eligibility - a
comparative matter, or a matter of degree. There are salient sharp lines, but
not in the right places. There is the line between the perfectly natural
properties and all the rest, but surely we have predicates for much-less-than-
perfectly natural properties. There is the line between properties that are and
that are not finitely analysable in terms of perfectly natural properties, but
that lets in enough highly unnatural properties that it threatens not to solve
our problem. We need gradations; and we need some give and take between
the eligibility of referents and the other factors that make for 'intendedness',
notably satisfaction of appropriate bits of theory. (Ideal theory, if we keep
as much of Putnam's story as we can.) Grueness is not an absolutely ineligible
referent (as witness my reference to it just now) but an interpretation that
assigns it is to that extent inferior to one that assigns blueness instead. Ceteris
paribus, the latter is the 'intended' one, just because it does better on eligibility.
Naturalness of properties makes for differences of eligibility not only
among the properties themselves, but also among things. Compare Bruce
with the cat-shaped chunk of miscellaneous and ever-changing matter that
follows him around, always a few steps behind. The former is a highly eligible
referent, the latter is not. (I haven't succeeded in referring to it, for I didn't
say just which such chunk 'it' was to be.) That is because Bruce, unlike the
cat-shaped chunk, has a boundary well demarcated by differences in highly
natural properties. Where Bruce ends, there the density of matter, the relative
abundance o f the chemical elements . . . . abruptly change. Not so for the
chunk. Bruce is also much more of a locus of causal chains than is the chunk;
this too traces back to natural properties, by the considerations of the previous
section. Thus naturalness of properties sets up distinctions among things.
The reverse happens also. Once we are away from the perfectly natural
David Lewis 373
properties, one thing that makes for naturalness of a property is that it is

a property belonging exclusively to well-demarcated things.
You might well protest that Putnam's problem is misconceived, wherefore

no need has been demonstrated for resources to solve it. Putnam seems to
conceive of language entirely as a repository of theory, and not at all as a
practice of social interaction. We have the language of the encyclopedia, but
where is the language of the pub? Where are the communicative intentions
and the mutual expectations that seem to have so much to do with what we
mean? In fact, where is thought? It seems to enter the picture, if at all, only
as the special case where the language to be interpreted is hard-wired,
unspoken, hidden, and all too conjectural.
I think the point is well taken, but I think it doesn't matter. If the problem
of inentionality is rightly posed there will still be a threat of radical
indeterminacy, there will still be a need for saving constraints, there will still
be a remedy analogous to Merrill's suggested answer to Putnam, and there
will still be a need for natural properties.
Set language aside and consider instead the interpretation of thought.
(Afterward we can hope to interpret the subject's language in terms of his
beliefs and desires regarding verbal communication with others.) The subject
is in various states, and could be in various others, that are causally related
to each other, to the subject's behaviour, and to the nearby environment that
stimulates his senses. These states fit into a functional organisation, they
occupy certain causal roles. (Most likely they are states of the brain. Maybe
they involve something that is language-like but hard-wired, maybe not. But
the nature of the states is beside the point.) The states have their functional
roles in the subject as he now is, and in the subject as he is at other times
and as he might have been under other circumstances, and even in other
creatures o f the same kind as the subject. Given the functional roles of the
states, the problem is to assign them content. Propositional content, some
would say; but I would agree only if the propositions can be taken as
egocentric ones, and I think an 'egocentric proposition' is simply a property.
States indexed by content can be identified as a belief that this, a desire for
that, a perceptual experience of seeming to confront so-and-so, an intention
to do such-and-such. (But not all ordinary ascriptions of attitudes merely
specify the content of the subject's states. Fred and Ted might be alike in
the functional roles of their states, and hence have states with the same content
in the narrowly psychological sense that is my present concern, and hence
believe alike e.g. by each believing himself to have heard of a pretty town
named 'Castlemaine ~. Yet they might be acquainted via that name with
different towns, at opposite ends o f the earth, so that Fred and not Ted
believes that Castlemaine, Victoria, is pretty.) The problem of assigning
content to functionally characterised states is to be solved by means o f
constraining principles. Foremost among these are principles of fit. If a state

is to be interpreted as an intention to raise one's hand, it had better typically
cause the hand to go up. If a state (or complex of states) is to be interpreted
as as a system of beliefs and desires - or better, degrees o f belief and desire
- according to which raising one's hand would be a good means to one's
ends, and if another state is to be interpreted as an intention to raise one's
hand, then the former had better typically cause the latter. Likewise on the
input side. A state typically caused by round things before the eyes is a good
candidate for interpretation as the visual experience of confronting something
round; and its typical impact on the states interpreted as systems of belief
ought to be interpreted as the exogenous addition o f a belief that one is
confronting something round, with whatever adjustment that addition calls
for.
So far, so good. But it seems clear that preposterous and perverse
misinterpretation s could nevertheless cohere, could manage to fit the
functional roles of the states because misassignment o f content at one point
compensates for misassignment at another. Let us see just how this could
happen, at least under an oversimplified picture o f interpretation as follows.
An interpretation is given by a pair of functions C and V. C is a probability
distribution over the worlds, regarded as encapsulating the subject's
dispositions to form beliefs under the impact of sensory evidence: if a stream
of evidence specified by proposition E would put the subject into a total state
S - for short, if E yields S - we interpret S to consist in part of the belief
system given by the probability distribution C(-/E) that comes from C by
conditionalising on E. V is a function from worlds to numerical desirability
scores, regarded as encapsulating the subject's basic values: if E yields S,
we interpret S to consist in part of the system of desires given by the C(-/E)-
expectations of V. Say that C and V rationalise behaviour B after evidence
E iff the system of desires given by the C(-/E)-expectations of V ranks B
at least as high as any alternative behaviour. Say that C and V f i t iff, for
any evidence-specifying E, E yields a state that would cause behaviour
rationalised by C and V after E. That is our only constraining prilxciple of
fit. (Where did the others go? -- We built them into the definitions whereby
C and V encapsulate an assignment of content to various states.) Then any
two interpretations that always rationalise the same behaviour after the same
evidence must fit equally well. Call two worlds equivalent iff they are alike
in respect of the subject's evidence and behaviour, and note that any decent
world is equivalent inter alia to horrendously counterinductive worlds and
to worlds where everything unobserved by the subject is horrendously nasty.
Fit depends on the total of C for each equivalence class, and on the C-
expectation of V within each class, but that is all. Within a class, it makes
no difference which world gets which pair of values of C and V. We can
interchange equivalent worlds ad lib and preserve fit. So, given any fitting
and reasonable interpretation, we can transform it into an equally fitting
perverse interpretation by swapping equivalent worlds around so as to enhance
David Lewis 375
the probabilities o f counterinductive worlds, or the desirabilities of nasty

worlds, or both. Q u o d erat d e m o n s t r a n d u m .
(My simplifications were dire: I left out the egocentricity of belief and desire
and evidence, the causal aspect of rationalised behaviour, the role of
intentions, change of basic values, limitations of logical competence . . . . .
But I doubt that these omissions matter to my conclusion. I conjecture that
if they were remedied, we could still transform reasonable interpretations
into perverse ones in a way that preserves fit.)
If we rely on principles o f fit to do the whole job, we can expect radical
indeterminacy of interpretation. We need further constraints, of the sort called
principles of (sophisticated) charity, or of 'humanity'. 33 Such principles call
for interpretations according to which the subject has attitudes that we would
deem reasonable for one who has lived the life that he has lived. (Unlike
principles of crude charity, they call for imputations of error if he has lived
under deceptive conditions.) These principles select a m o n g conflicting
interpretations that equally well conform to the principles of fit. They impose
a priori -- albeit defeasible - presumptions about what sorts of things are
apt to be believed and desired; or rather, about what dispositions to develop
beliefs and desires, what inductive biases and basic values, someone m a y
rightly be interpreted to have.
It is here that we need natural properties. The principles of charity will
impute a bias toward believing that things are green rather than grue, toward
having a basic desire for long life rather than for long-life-unless-one-was
born-on-Monday-and-in-that-case-life-for-an-even-number-of-weeks. In
short, they will impute eligible content, where ineligibility consists in severe
unnaturalness of the properties the subject supposedly believes or desires or
intends himself to have. They will impute other things as well, but it is the
imputed eligibility that matters to us at present.
Thus the threat of radical indeterminacy in the assignment of content to
thought is fended off. The saving constraint concerns the content - not the
thinker, and not any channels between the two. It takes two to index states
with content, and we will not find the constraint if we look for it always on
the wrong side of the relationship. Believing this or desiring that consists
in part in the functional roles o f the states whereby we believe or desire, but
in part it consists in the eligibility of the content. And this eligibility to be
thought is a matter, in part, of natural properties.
Consider the puzzle whereby Kripke illustrates Wittgenstein's paradox that
'no course of action could be determined by a rule, because every course of
action can be made out to accord with the rule'. 34 A well-educated person
working arithmetic problems intends to p e r f o r m addition when he sees the
' +' sign. He does not intend to perform quaddition, which is just like addition
33 See my 'Radical Interpretation', Synthese 23 (1974) pp. 331-344; and Richard E. Grandy,
'Reference, Meaning and Belief', Journal of Philosophy 70 (1973) pp. 439-452.
34 See Saul A. Kripke, 'Wittgensteinon Rules and Private Language:An ElementaryExposition',
in Irving Block, ed., Perspectives on Wittgenstein (Blackwell, 1981).
for small numbers but which yields the answer 5 if any o f the numbers to
be quadded exceeds a certain bound. Wherefore does he intend to add and
not to quadd? Whatever he says and whatever is written in his brain can be
perversely (mis)interpreted as instructing him to quadd. And it is not enough
to say that his brain state is the causal basis of a disposition to add. Perhaps
it isn't. Perhaps if a test case arose he would abandon his intention, he would
neither add nor quadd but instead would put his homework aside and
complain that the problems are too hard.
The naive solution is that adding means going on in the same way as before
when the numbers get big, whereas quadding means doing something
different; there is nothing present in the subject that constitutes an intention
to do different things in different cases; therefore he intends addition, not
quaddition. We should not scoff at this naive response. It is the correct

solution to the puzzle. But we must pay to regain our naivet6. Our theory
of properties must have adequate resources to somehow ratify the judgement
that instances of adding are all alike in a way that instances of quadding
are not. The property of adding is not perfectly natural, of course, not on
a par with unit charge or sphericality. And the property of quadding is not
perfectly unnatural. But quadding is worse by a disjunction. So quaddition
is to that extent less of a way to go on doing the same, and therefore it is
to that extent less of an eligible thing to intend to do.
It's not that you couldn't possibly intend to quadd. You could. Suppose
that today there is as much basis as there ever is to interpret you as intending
to add and as meaning addition by your word 'addition' and quaddition by
'quaddition'; and tomorrow you say to yourself in so many words that it
would be fun to tease the philosophers by taking up quadditon henceforth,
and you make up your mind to do it. But you have to go out of your way.
Adding and quadding aren't on a par. T o intend to add, you need only have
states that would fit either interpretation and leave it to charity to decree
that you have the more eligible intention. To intend to quadd, you must say
or think something that creates difficulties of fit for the more eligible intention
and thereby defeats the presumption in its favour. You must do something
that, taking principles of fit and presumptions of eligibility and other
principles of charity together, tilts the balance in favour of an interpretation
on which you intend to quadd. How ironic that we were worried to find
nothing positive to settle the matter in favour o f addition! For the lack of
anything positive that points either way just is what it takes to favour addition.
Quaddition, being less natural and eligible, needs something positive in its
favour. Addition can win by default.
What is the status of the principles that constrain interpretation, in
particular the charitable presumption in favour of eligible content? We must
shun several misunderstandings. It is not to be said (1) that as a contingent
psychological fact, the contents of our states turn out to be fairly eligible,
we mostly believe and desire ourselves to have not-too-unnatural properties.
Still less should it be said (2) that we should daringly presuppose this in our
interpreting o f one another, even if we haven't a shred of evidence for it.
David Lewis 377
Nor should it be said (3) that as a contingent psychological fact we turn out
to have states whose content involves some properties rather than others,
and that is what makes it so that the former properties are more natural.
(This would be a psychologistic theory of naturalness.) The error is the same
in all three cases. It is supposed, wrongly as I think, that the problem of
interpretation can be solved without bringing to it the distinction between
natural and unnatural properties; so that the natural properties might or might
not turn out to be the ones featured in the content of thought according to
the correct solution, or so that they can afterward be defined as the ones
that are so featured. I think this is overoptimistic. We have no notion how
to solve the problem of interpretation while regarding all properties as equally
eligible to feature in content. For that would be to solve it without enough
contraints. Only if we have an independent, objective distinction among

properties, and we impose the presumption in favour of eligible content a
priori as a constitutive constraint, does the problem of interpretation have
any solution at all. If so, then any correct solution must automatically respect
the presumption. There's no contingent fact o f psychology here to be believed,
either on evidence or daringly.
Compare our selective and collective theory o f lawhood: lawhood of a
regularity just consists in its fitting into an ideally high-scoring system, so
it's inevitable that laws turn out to have what it takes to make for high scores.
Likewise, I have suggested, contenthood just consists in getting assigned by
a high-scoring interpretation, so it's inevitable that contents tend to have what
it takes to make for high scores. And in both cases, I've suggested that part
of what it takes is naturalness of the properties involved. The reason natural
properties feature in the contents of our attitudes is that naturalness is part
of what it is to feature therein. It's not that we're built to take a special interest
in natural properties, or that we confer naturalness on properties when we
happen to take an interest in them. 35
Princeton University Received August 1982

Revised May 1983
3~ I am indebted to commentsby Gilbert Harman, Lloyd Humberstone, Frank Jackson, Mark

Johnston, Donald Morrison, Kim Sterelny, and others; and especially to discussion and
correspondence with D. M. Armstrong over several years, without which I might well have
believed to this day that set theory applied to possibilia is all the theory of properties that
anyone could ever need.

New - Work David Lewis

Uploaded by

Copyright:

Available Formats

New - Work David Lewis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

New - Work David Lewis

Uploaded by

Copyright:

Available Formats

Australasian Journal o f Philosophy

Vol. 61, No. 4, December 1983

NEW WORK FOR A THEORY OF UNIVERSALS

Introduction. D. M. A r m s t r o n g offers a t h e o r y o f universals as the o n l y

adequate a n s w e r to a ' c o m p u l s o r y q u e s t i o n ' for s y s t e m a t i c p h i l o s o p h y : the

D. M. Armstrong, Universals and Scientific Realism (Cambridge University Press, 1978),

and thought. Perhaps the list could be extended.

Universals and Properties. Language offers us several more or less

4 My conception of properties resembles the doctrine of Class Nominalism considered in

The only alternative to redundancy might be inadequacy, and if so we had

objective difference between natural and unnatural properties; preferably,

(where the strings o f variables may be infinite, even uncountable) must be

(1) Red resembles orange more than it resembles blue.

Such a theory is a form of Resemblance Nominalism, in Armstrong's classification, but it

or humility. Armstrong agrees (Universals, I, P. 61) that he cannot take (1)-(4)

quantification. From Universals, I, p. 62, and "Against 'Ostrich' Nominalism"

Prima facie, we are quantifying either over properties or over universals.

as quantifications over universals, we would again have to resort to some

I also think that it is properties that we need in characterising the content

perfectly natural, hence cannot be replaced by universals. It is interesting

Or (3) 'I accept it as primitive'. Not every account is an analysis! A system

o f the things which are apparently of the same type; or it predicates

analysed. Some predications are to be analysed away in terms of others. Here

circularity analyse in terms of class membership. So falls Resemblance

is informative, it makes a vital contribution to telling us what is the case,

otherwise than by means o f a one-off analysis for some specific F. To that

" 'Ostrich Nominalism' or 'Mirage Realism'?", Pacific Philosophical Quarterly 61 (1980)

to primitive predication of resemblance.

Duplication, Supervenience, and Divergent Worlds. Henceforth I shall

properties -- not a complete inventory, perhaps, but a complete enough

Not only is duplication of interest in its own right; it also is needed in

First, supervenience. A supervenience thesis is a denial of independent

of statues supervenes on their shape, size, and colour, for instance, if no

~s For a generaldiscussionof supervenience,see JaegwonKim, 'Supervenienceand Nomological

analysis of causal counterfactuals and de re modality. If not, such projects

19 Sucha thesis of supervenienceof the global on the local resemblesthe 'holographichypothesis'

of what could be predicted by a superhuman calculator with unlimited

20 This approachis due, in essence, to Richard Montague,'DeterministicTheories',in Decisions,

alternatives are divergent worlds, with duplicate initial segments, z2

M i n i m a l Materialism. There is a difficulty that arises if we attempt to

22 See my 'CounterfactualDependenceand Time'sArrow', Nofts 13 (1979)pp. 455-476; Jonathan

supervenience thesis: no difference without physical difference. Or,

mind? - but a thesis that says so is more than just Materialism.

But M3 is not a formulation of Materialism, for the distinction between

2~ Thisobjection against M4 as a formulation of 'the ontologicalprimacy of the microphysical'

In either case there are extra natural properties, properties instantiated in

Now we can proceed at last to formulate Materialism as a restricted and

We took Materialism to uphold the comprehensiveness of 'something not

Laws and Causation. The observation that physics discovers natural

one merit would be lost. F o r A r m s t r o n g , the lawful necessitation o f Ga by

be absolutely impossible to have N ( F , G ) a n d Fa w i t h o u t Ga. (Unless N just

So an adequate regularity analysis must be selective. Also, an adequate

If the analysis of lawhood requires natural properties, then so does the

28 See my'Causation',Journal of Philosophy 70 (1973)pp. 556-567; reprintedin Ernest Sosa,

worlds, hence in terms of duplicate initial world-segments, hence in terms

Heat is that phenomenon, whatever it may be, that manifests itself in

d e p e n d s n o t o n l y o n w h a t goes o n in t h e r e g i o n b u t also o n w h a t goes on

The Content o f Language and Thought. H i l a r y P u t n a m has given an

reference m u s t be established b y o u r o w n s t i p u l a t i o n in l a n g u a g e o r thought.