Mathematics as a Science of Patterns: Ontology and Reference

Author(s): Michael D. Resnik

Source: Noûs, Vol. 15, No. 4, Special Issue on Philosophy of Mathematics (Nov., 1981), pp. 529-
550
Published by: Wiley
Stable URL: http://www.jstor.org/stable/2214851 .
Accessed: 01/12/2013 11:53

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

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

Wiley is collaborating with JSTOR to digitize, preserve and extend access to Noûs.

http://www.jstor.org

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
 Mathematics
as A ScienceofPatterns:
Ontologyand Reference
MICHAEL D. RESNIK

UNiVERSITYOF NORTH CAROLINA

1. INTRODUCTION

I seek an account of mathematicsin whichthe logical formsof mathe-

maticalstatementsare taken at face value and theirsemanticsis stand-
ardlyreferential,say,in the mannerof Tarski. This togetherwithfairly
uncontested assumptions entails that mathematics is a science of
abstractentities,thatis,immaterialand nonmentalthingswhichdo not
exist in space and time. So I am a platonist.1
Many philosophersof mathematicswho would like to be platonists
are bothered by two rather deep problems. The firstis that since
platonic mathematicalobjects do not exist in space or time the very
possibilityof our acquiring knowledge and beliefsabout them comes
into question. The second arises fromthe fact that no mathematical
theorycan do more than determine its objects up to isomorphism.
Thus the platonistseems to be in the paradoxical positionof claiming
that a given mathematicaltheoryis about certain things and yet be
unable to make any definitivestatementof what these thingsare.2
I thinkthatthese problemsarise in partbecause of a fundamental
misconceptionof what mathematicsis about. If we conceive of the
numbers,say,as objectseach one of whichcan be givento us in isolation
from the others as we thinkof, say, chairs or automobiles,then it is
difficultto.avoid conceivingof knowledge of a number as dependent
upon some sortof interactionbetweenus and thatnumber. The same
line of thoughtleads us to thinkthatthe identityof a numbervisa vis
any other object should be completelydetermined. For some time
mathematicianshave emphasized thatmathematicsis concerned with
structuresinvolvingmathematicalobjects and not withthe "internal"
natureof the objectsthemselves.They have recognizedthatwe are not
givenmathematicalobjectsin isolationbutratherin structures.That 13
is a prime number is not determinedby some internalpropertyof 13
but ratherby its place in the structureof the natural numbers. Some

529

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
 530 NOO S

philosophershave already taken note of thismovementin mathemat-

ical thought.3But I propose to take thisidea more seriouslythan they
have. In mathematics,I claim,we do not have objectswithan "internal"
compositon arranged in structures,we have only structures.The
objects of mathematics,that is, the entitieswhich our mathematical
constantsand quantifiersdenote, are structurelesspointsor positions
in structures.As positionsin structures,theyhave no identityor fea-
turesoutside of a structure.Furthermore,the various resultsof math-
ematics which seem to show that mathematicalobjects such as the
numbersdo have internalstructures,e.g., theiridentificationwithsets,
are in fact interstructuralrelationships.
I findit more suggestiveforepistemologicalpurposes to speak of
mathematicalpatternsand theirpositionsratherthan of structures.I
viewpatternsand theirpositionsas abstractentities.Most of thispaper
willbe devoted to developingan extensionalaccount of patternsand to
explaining how viewingmathematicalobjects as positionsin patterns
leads to a reconception of mathematicalobjects which defuses the
objection to platonismbased upon our inabilityto completelyfixtheir
identity.
I also think that viewing mathematicsas a science of patterns
promises to solve the platonist'sepistemologicalproblems as well-or
at least to make them less urgent-by showing that mathematical
knowledge has a fairlycentral place in our general epistemological
picture. I expect to develop a more detailed epistemologyelsewhere,
but before proceeding to the main tasksof the paper I want to sketch
my currentview.4
When we are exposed to several instancesof a patternof certain
kinds(and are in the rightpsychologicalset and have the approrpriate
purposes) we are bynaturestruckbythesimilarity of theinstances-we
see that theyfita pattern.5Thus the detectivesees several strangula-
tionsin a month'stimeand thinks"thereis a patternto this".Of course,
thisis not to knowwhatthe patternis nor to be able to make any use of
it. However, after furtherexposure to instances of the pattern (or
furtherthoughton the previous instances)we are likelyto attemptto
describe the patternitself.This is oftendone via role talkor positional
talk. For example, the detectivewill say: the victimis strangledwitha
scarf,the murderer approaches the victimfrombehind, the crime is
committedon a Wednesday,etc. (Reflectionon thismay lead us to an
accountof how referencein mathematicsis possible.) Such descriptions
of a patterncan then be checked against the data whichgenerated the
descriptionand which are supposed to fitthe pattern. Describing a
patternis thus closelyakin to developing a theoryof it and of the data
fittingit. Once we have a partial descriptionof a patternwe can also
inferotherfeaturesof it.Thus thedetectivededuces thatthemurderer

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
MATHEMATICS AS A SCIENCE OF PATTERNS 531

has the size and strengthto stranglethe victimwitha scarf.Moreover,

once we have representeda patternto ourselveswe can easilythinkof
variationson it. Thus murder on Fridayratherthan Wednesday, two
victimssimultaneouslystrangledwithone long scarf,etc. This route to
the knowledgeof new patternsis not onlyquite familiarto mathemat-
iciansbut absolutelycrucialto myaccount of mathematicalknowledge.
For it is not implausible to thinkof our coming to know small finite
patternsthroughexposure to concreteinstancesof them,but one has
to strain quite hard to believe that we arrive at our knowledge of
infiniteor complicated patternsin thisway. Infinitepatternsare first
thoughtof, I would suggest,by thinkingof finitepatternsas indefi-
nitelyextended. There is more than this to knowledge of an infinite
pattern,of course; our representationof the infinitepatternmust be
consistent;it mustrelate to the finitepatternsit extends in the appro-
priatewayand so on. But I thinkthatalong theselines an epistemology
for number theoryand the higher branches of mathematicscan be
developed. If so, this would place the epistemologyof mathematics
along side epistemologiesforlinguisticand musicalknowledge.For no
doubt we performanalogous operationsin those fields.We internalize
patternsof linguisticutterancesor muscial performances;we are able
to generate new patternswhich have never before been uttered or
performed,and we are able to classifyand recognize general features
of these patterns.If we could not, we could not writenew novels and
songs or recognize nonsense and dissonance. Knowing thatadding is
commutativemay be like knowingthatthe active-passivetransforma-
tion generallypreservescognitivemeaning; knowingthat there is no
greatest natural number may be like knowing that given a song in
which a measure is repeated, say, twice,there is (could be) another in
which it is repeated three times.
Much more needs to be said about an epistemologyof mathematics
based upon our knowledge of patterns.But I hope that my sketch
makes it plausible thatthe epistemologyof mathematicsis no more (or
not much more) mysteriousthan the epistemologiesof linguisticsand
music. Like mathematicstheybegin withexperience, abstractfromit
and arriveat the unexperienced (and, perhaps, like mathematics,even
the unexperientable).

2. PATTERNS

Although I know of no developed philosophicalaccount of patternsI

have encountered several suggestionsas to what theymightbe.6 They
all take offfromthe relationshipbetweena patternand itsinstances.I
focus instead upon how patternsare related to each other and get
instantiationas a special case. The approach seems quite naturalto me,

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
532 NOUS

but myexperience has been thatintuitionsvaryconsiderablyon these

matters.It may help you understandmyapproach ifyou bear in mind
thatithas been generatedbyreflectingon twomathematicaltheoriesof
structure,namely,geometryand model theory.My view makes model
theoryinto a geometry.
On myview,a patternis a complex entityconsistingof one or more
objects, which I call positions,standing in various relationships(and
having various characteristics,distinguished positions and oper-
ations.)7A positionis like a geometricalpoint in that it has no distin-
guishingfeaturesotherthan those ithas in virtueof beingthatposition
in the pattern to which it belongs. Thus relative to the equilateral
triangle ABC the three points A, B, C can be differentiated,but
considered in isolationtheyare indistinguishablefromeach otherand
the verticiesof any trianglecongruentto ABC. Indeed, considered as
an isolatedtriangle, ABC cannot be differentiatedfrom any other
equilateral triangle. In geometry structuralrelationships-such as
congruence and similarity-are paramount while claims of identity
and distinctnessof points,lines, planes, etc. are reserved forcontexts
where these entitiesare related to several other geometricalobjects. I
transferthis geometricalanalogy to the various structuresstudied by
mathematics.Withina structureor patternpositionsmaybe identified
or distinguished,since the structureor patterncontainingthem pro-
vides the contextforso doing. However, the paramount relationships
among patterns are those of structuralsimilarity(congruence and
equivalence introduced below) and structuralcontainment (occur-
rence and subpattern).Let us apply the analogy to the naturalnumber
sequence (N, S). I take thisto be a patternwitha singlebinaryrelation
(successor) and the natural numbers to be its positions.Viewed thus
thereis no more perplexityin the factthatthe naturalnumbershave no
identifyingfeaturesbeyond those definable in terms of the pattern
than there is to the corresponding fact about the points in triangle
ABC.
Patternsare relatedto each otherbybeingcongruent or structurally
isomorphic.Congruence is an equivalence relationwhose fieldI taketo
include bothabstractstructuresand arrangementsof concreteobjects.
Thinkingof patternsas models of formalsystems,itis the relationship
whichholds betweenisomorphic modelsof formalsystems.Consider, for
example, a firstorder systemS witha single two-placepredicate "R"
and axioms statingthat R is a total ordering. This systemhas many
models-all the totalorderings-but theyare not all congruentto each
other.Onlythosewhose domains have thesame cardinalityare. The set
of numbers from one to ten taken in their natural order and ten
puppies in a littertaken in order of theirbirthare isomorphicmodels
of the systemS; so I count the abstractnumerical structureand the
arrangementof puppies as congruent.

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
MATHEMATICS AS A SCIENCE OF PATTERNS 533

When a patternand an arrangementof so-called concreteobjects

such as the puppies are congruent then I say that the arrangement
instantiates the pattern.Instantiationthen is a special case of congru-
ence in whichthe objects "occupyingthe positions"of a patternhave
identifyingfeatures over and above those conferred by the ar-
rangementsto which theybelong. The puppies thus instantiatethe
one-to-tenpattern.
The occurrence of patternswithineach other has played a more
fundamentalrole in mythinkingthan has instantiation.If you thinkof
the sentence "A cat bit the cat who ate the cat" as a linguisticpattern,
then the word "cat" is a patternwhich occurs three times in it. The
natural number sequence occurs withinthe patternof the real num-
bers taken in their natural order; it also occurs withinthe iterative
hierarchyof setsand withinitself.Five pointedsolid starpatternsoccur
in both the patternsof the United States flag and the Californiaflag.
The fieldof the occurrencerelationis the same as thatof congruence;
hence concretearrangementscan occur withinpatternsor other con-
crete arrangements. My interest,however, is mainly with patterns
occurringwithinpatterns.More formally,occurrenceis a reflexiveand
transitiverelation which holds between structuresP and Q when P is
isomorphicto a structuredefinablein Q. Thus (N, S) occurs within(N,
<) and withinmodels of varioussettheories.The rationalnumbersqua
field(Rat, +, ) contain an occurrenceof (N, S) but the rationalnum-
bers qua countable dense ordering (Rat, <) do not because 0 and
successor are not definable in such structures.
A special case of patternoccurrence is the subpatternrelation.
PatternA is a subpattern of B just in case everypositionof A is a position
of B and A occurs in B. It followsthatpatternA occursin patternB just
in case A is congruentto a subpatternof B.
Take note thatthe subpatternrelationis more complicated than
the submodel relationof model theory,althoughit includes an analog
to that relation in its special cases. My reasons for the more complex
approach turnupon both myintuitionsabout patternsand myaim to
give an account of reduction in mathematics.
I will take up the intuitiveconsiderationsnow and leave the dis-
cussion-of reduction for later. Suppose that we are given a char-
acterizationof a patternas havingninedistinctpositionsa, b, c, d, e, f,g,
h, i, and tworelationsA and L such thata L b, b L c, d L e, e L f,g L h, h L
i, a A d, d A g, b A e, e A h, c A f,fA i. To tryto geta bettergrasp of the
patternwe draw the followingdiagram where we let L correspond to
therelationoneunitoftheleftofand A correspondto therelationoneunit
above:

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
 534 NOUS

a b c
O 0 0
d ,' ,e %% f
o ~~~0-(a%
g ,00 h "i
? ---------lo---------1?o

Now givenmyunderstandingof thediagram(and I admitthatyou may

not see thingsas I do) the rectangleof dots containsa triangleof dots
gbi. (It containsotherdot trianglesas well,of course.) Thinkingnow of
my drawingas also a representationof a spatial pattern(one whichis
rathermore complicate'dthantheoriginalpattern),I conclude thatthe
spatialtriangleis a subpatternof thewhole spatialarray.I thentakethis
as evidence thatcorrespondingto the dot triangleare subpatternsof
the original patterninvolvingthe positionsg, b, and i. These patterns
can be characterizedby relationsdefinablein termsof A and Ljust as
the spatial dot trianglecan be describedin termsof relationsdefinable
in termsofaboveand left.Indeed the formerdefinitionscan be read off
thelatterquite straightforwardly. Nevertheless,therelationsof neither
the spatial nor the non-spatialsubpatternare simplerestrictions of the
relationsof the original pattern.Thus if we count the spatial triangle
and itscorrespondentas subpatternsof theoriginalpattern-as I think
we should-then the subpatternrelation must be defined along the
lines which I have suggested.
Myvisual intuitionswithrespectto subpatternsand patternoccur-
rences can be transferredto some of the model theoreticcases as well.
Let us thinkof the natural number sequence as representedby means
of an unending linear sequence of dots.
000000000000000 ....
Clearly a sequence of dots which we obtain from this sequence by
startingwiththe nth dot fromthe leftis a subpatternof the original
pattern.These correspond to all those progressionswhichare obtain-
able from(N, S) byrestricingS to a subsetof N. But thereare infinitely
many models of number theoryobtainable from(N, S) whichare left
out bec'ause their relations of succession are not subrelationsof the
successorrelation.The even numbersequence is one, the odd number
sequence is another,and the primenumbersequence a third.These all
correspond to selectingsome progression of dots from the original
sequence of dots. These sequences occur within the original
sequence-at least as I see it-and to account forthatboth the subpat-
ternand patternoccurrencerelationsmustbe characterizedin termsof
definability.

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
MATHEMATICS AS A SCIENCE OF PATTERNS 535

The natural number sequence (N, S) occurs withinthe natural

numbers ordered under less than, (N, <), since

Sxy x < y. (3z) (x < z . z < y);

and, allowing second order or set theoreticdefinability,the converse

relation holds too since:
X _ . Suv D VEZ)DyEz).8
y- (Z)(XEZ.. (U)(V)(UZ

Most mathematicians and logicians would regard number theory

developed in a language in whichthe successor symbolis primitiveas
essentiallythe same as a development takingthe less than symbolas
primitive.Since I am viewingnumbertheoryas the science of a certain
patternor patterns,thiswould suggest that (N, S) and (N, <) should
count as the same or essentiallythe same pattern.The problemthen is
to finda satisfactory characterizationof thisrelation.Patterncongru-
ence is too stricta condition since these two structuresare not isomor-
phic. Moreover, they are not isolated examples of non-isomorphic
structureswhichmathematiciansview as essentiallythe same: we have
Boolean Algebras in the formof rings,but also in the formof lattices,
alternativedefinitionsof groups and topologies, and so on.
While itis true thatin these cases we have theorieswhose patterns
occur withineach other, mutual patternoccurrence seems to be too
weak a conditionforpatternequivalence. One reason is that(N, S) and
(Rat, +, ) occur withineach other although theyare not regarded as
essentiallythe same.9 A second reason is that the patternoccurrence
relationcorresponds to the notion of theinterpretability of theories(by
means of definitions)and thereare theorieswhichare mutuallyinter-
pretable but not essentiallythe same fromthe mathematicalpoint of
view.I0
If we look at well known examples of "essentiallythe same"
theorieswe see thattheymeeta strongerconditionthan thatof mutual
interpretability. They are definitionallyequivalent.This means thatthere
is a set of definitionsDST + DTS such thatS + DTS yieldsboth T and
DST and T + DST yieldsboth S and DTS. In otherwords,the theories
together with interpreting definitons yieldnot onlyeach otherbut also
each other'sinterpretingdefinitions.Think of definitionsas axioms of
a special kind and thinkof the language of a theoryas includingthe
symbols introduced by definition. Then definitionallyequivalent
theoriesare a special kindof equivalentaxiomatizationsof the same set
of theorems."
To capture the notion of patternswhich are essentiallythe same
we need a relationshiplike that of definitionalequivalence. We can
move to complete theories of patternsusing techniques from model

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
536 NOUS

theorybut thiswillhelp onlywhen we can forma categoricaltheoryof a

patternas well. For patternsare specificmodels of theories-specific at
least up to isomrorphism, and complete theories withnonisomorphic
models abound. Despite this,the conception of definitionallyequiva-
lent theories as distinctencapsulations of a common set of theorems
suggestsa similarapproach for patterns.
Intuitively,twopatternsare like "essentiallythe same" iftheyboth
encapsulate some "bigger"patternfromwhichtheycan be obtained by
deletingsome of itsrelationships.More formally,let us call a patternP
a truncation of a patternQ ifeveryposition,characteristicand relation
of P is also one of Q. Then a patternP willbe said to be equivalentto a
patternQjust in case thereis a patternR whichis a subpatternof bothP
and Q and of which P and Q are respectivetruncations.To illustrate
thisconcept let us returnto (N, S) and (N, <). These are truncationsof
(N, S, <) and it occurs withinboth of them. So they are equivalent
according to the definition.On the other hand, although (N, S) and
(Rat, +, ) occur withineach other they cannot be obtained from a
common extension by means of truncationsince paring (Rat, S, +, )
down to (N, S) requires deleting some of its positions as well as its
relations.
Of theequivalence relationshipswhichcan occur betweenpatterns
congruence is the strongest,equivalence the next and mutual occur-
ence the weakest. I will not pursue the mathematicsof these relation-
ships furtherhere.
I have discussed several equivalence relationsbetween patterns-
congruence, mutual occurrence,equivalence-but have failed to pro-
pose any identityconditionsfor patterns.12 I willnot; and thatbrings
me to whatI findthe mostdifficultpointin mytheory-the restriction
of identityto withinpatterns.Recall thatpatternsare like geometrical
pointsand have no identifyingfeaturesbeyond those theyacquire by
being in a pattern.Now suppose thatwe considera pointA in a triangle
and a point A' in a rectangleusing the diagram given below:

A
A

B.LI B.

Are A and A' the same? One might respond that they cannot be
because theybelong to separate figures.Well, how do we know that
theyare separate? Because theyare drawn as being spatiallydistinct.

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
MATHEMATICS AS A SCIENCE OF PATTERNS 537

Suppose then that we ignore thisaspect of the drawing and consider

themas isolatedfigureswithno informationsupplied as to theirspatial
relationships.Then the figures could be spatiallyseparate, or they
could be contiguousand A could be A'-as would be the case ifthe line
segmentsAB and A'B' are identical. Of course, this is no reason to
conclude thatthereis no factof thematterhere concerningtheidentity
of A and A'. They are, afterall, points in space and have adequate
identityconditions-such as being on the same lines. We simplydo not
have enough informationto determinethe factsin this example.
This is quite true. But it is part and parcel of our developmentof
geometryas a theoryof space. Had geometrydeveloped instead as a
theoryor collectionof theoriesoffiguresor shapes,then points might
onlyplaytheirrole of markinglocationswithinfigureswithoutplaying
their role of markinglocations in the containing space. This would
require us to viewthe triangleand rectanglepresentedabove (withthe
spatial relationshipsas indicatedby the drawing)as a singlecomposite
figureand to take a similarview of other complexes of geometrical
figures,but I see no obstacle to doing thisor to interpretingmuch of
elementarygeometryalong these lines. Geometrydeveloped in this
way would attributeno being to points independent of the figures
containingthem. Identitymightbe relativizedto elementsof the same
figure or each figure (figure kind) mightbe treated by means of a
separate theorywithan unrelativizedidentitypredicate rangingover
all elementsof its universe.
The latter alternativeis the one which mathematicstakes with
respect to the structuresstudied by its theories. Natural number
theory,for example, deals witha certainpattern;it has the means to
raise and answerquestionsconcerningtheidentityof variousnumbers,
butitcannotevenformulate thequestionas to whetherthe numberone is iT.
On the otherhand, withinnumbertheoryidentityis absolute: "(x)(y)(x
= y V x / y)" is one of its theorems.Similarremarkshold for the real
numbersand real number theory,functionsand analysis,sets and set
theory.In each case the theoryspeaks only of elements of a certain
structureand has no means to identifyor distinguishthese from
elementsof another structure.Like the variantof geometrywithdif-
ferenttheoriesof differentshapes, mathematicsis a conglomerationof
theorieseach dealing withitsown structureor patternand each forego-
ing identitiesleading outside of its pattern.
However, if we exclude identitiesinvolvingpositionsof different
patterns(thisexclusion does not apply to subpatterns)then we cannot
use any of the equivalence relations introduced earlier as identity
conditionsforpatterns.To see whyconsidercongruence.Since itis the
strongestof these relations,we can restrictour attentionto it. The
natural number sequence is congruentto the even number sequence.

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
 538 NOUS

If we take theircongruence as sufficentfor theiridentity,then their

corresponding positions should count as identical too. But then 1
would be 2! We mighttryto avoid thisbycapitalizingon one's being the
second natural number and two'sbeing the second even number and
relativizeidentitypredicateslike this:2 7NW1,2 =E1. The troubleis that
4 is the even number correspondingto the natural number 2 but qua
even numberstheyare distinct,so we would have: 4 =E2, 4 #E2. I know
of no other means for identifyingdistinctcongruent,equivalent or
co-occurringpatternswhich will not fall prey to similardifficulties.
This suggeststhat we identifypatternsjust in case theyhave the
same positions,the same relations,etc. We could (and I would) use
extensional identityconditionsfor relations and characteristicsif we
took thisroute. But even thisis not open to me. For it would require
somethinglike:
P1 = P2 (x) (x is a position of P1 x is a position of P2) & etc.
and thatwould require admittingpositionsof differentpatternsintoa
singleuniverseof discoursecontraryto myprevious moves. So neither
identitybetweenpatternsnor betweenpositionsof differentpatternsis
(everywhere)defined.
The exclusion of patternsthemselvesfromthe fieldof identityis
consistentwith the practice of mathematicsitself. Number theory
quantifiesover just the numbers but not over the number theoretic
structure,set theoryquantifiesover sets but not over the set theoretic
hierarchy.Even model theory,which purports to treat of arbitrary
mathematicalstructures,does so by positionalizingthem, that is, by
identifyingstructureswithsets or ordered n-tuples.Nonetheless,the
reasoning which began with the restrictionof identityto positions
withinthe same patternis now in danger of underminingmy entire
theory.For it appears that even the various equivalences among pat-
ternsmustbe abandoned ifwe cannotspeak of positionsfromthe same
patternwithinthe same breath. I began by takingstructuralor model
theoreticisomorphismas criterialfor pattern congruence, but such
isomorphismsare functionsfromthe positionsof one patternto those
of another. This seems to require a universewhichcontains positions
fromdifferentpatterns.
To speak of relationshipsamong variouspatternswe mustevolvea
comprehensivetheoryin which positionsfromdifferentpatternscan
be dealt within the same breath. There are several options for such a
comprehensivetheory.First,one could use a many-sortedtheorywith
separate universesforeach patternand unreduced functionsbetween
these universesto serve as isomorphisms.This would be closestto the
spiritof my informalexposition.
A second (ontologicallycleaner) approach would constructpat-
terntheoryalong the lines of geometrybypositinga space of positions

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
MATHEMATICS AS A SCIENCE OF PATTERNS 539

fromwhichpatternscould be constructed.This theorywouldestablish

facts of the matterfor identitiesinvolvingpositions of its different
patternssince identitywould apply to all positionsin itsspace. How-
ever, like geometry,we would not expect the theoryto have theorems
dealing with particularpositions.A thirdapproach would refinethe
second one step furtherbyreducingpatternsto setsof positionsjustas
model theoryconstruesmodels as setsbuiltup fromthe domain of the
model. One could push thiseven one step furtherand construeposi-
tionsthemselvesas pure sets,therebyreducingso-calledpatterntheory
to set theory.
The lastthreemoves do not do awaywithpatterns.Instead forthe
motleyof patternsdealt withby the various branches of mathematics
and by myinitialapproach to patternstheysubstituteone big pattern
and its positionsand ask mathematicsto make do withit. For on my
view,the set theoretichierarchy,for instance,is a patternalong with
others.Pure setsare positionsin it whose apparent internalstructures
are a fabricationof theirrelationshipsto other positionsin the hierar-
chy.
Moving to one big patternin which all positions and "internal"
patternsfall withinthe scope of identity(patternsbecome sets on the
third and fourth approaches) seems to involve a retractionof my
earlier claims about identity.I think,however,thatI can explain why
thisis not so. I did not mean to suggestin discussingthe second, third
and fourthapproaches that the various patternsof mathematicsare
identicalwiththe patternsdeveloped withinone of the comprehensive
accounts; I only meant to suggestthattheycould be reduced to them.
To appreciate the differencewe must turn now to the subject of
reduction in mathematics.

3. REDUCTIONS

According to the definitionsI have given, the natural number se-

quence has multipleoccurrenceswithina patternwhichI willreferto
as the iterativeset theoretic hierarchy. It follows that even categorical
versionsof number theory can be interpretedas dealing withany of
these patterns and the set of itstrue sentenceswillremainintact.This is
it
as should be when we speak of patterns. If I completelydescribethe
relationships of a set of points in a geometrical figure,thatdescription
continues to hold if I think of that figure as embedded in a largerone.
The differenceis that my description will not cover the relationships
these points have to new points of the containing figureand willfailto
characterizethat figure. But why should that be demanded of the
original description?
The settheoretichierarchy,byvirtueof itscontainingoccurrences
of the natural number sequence, can be viewed as an extentionof the

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
540 NOUS

latterpattern.This interplaybetween occurrence and extentionhas

been usefulin guidingthe developmentof set theoryitself.Borrowing
the notationand axioms of numbertheoryto describea portionof the
set theoreticpatternwillintroduceno falsehoodsinto set theoryand it
willcall attentionto the occurrenceof the naturalnumbersequence in
question. That there is such an occurrence is surely as interesting
mathematicallyas it is philosophically.Bringing number theory in
suggests developments in set theory itself too-transfinite gen-
eralizationsof number theory.Thus the practicenow standard in set
theoryof usingthe numeralsto designatecertainsetsisjustifiedfroma
pragmaticpoint of view.
Although only one of the many occurrences of the natural se-
quence is referredto in the developmentof number theorywithinset
theory,there is no embarrassmentfor mathematicsin this. Choosing
one occurrence, rather than another, will certainlyaffectwhich set
theoretictheorems get reported in our monographs but it does not
change the set of theoremsof set theoryitself.It is importantto show
thatnumbertheorycan be done withinsettheory.Given thatthereare
severalwaysin whichitcan be done, the simplest,mostelegantor most
novel ways are what deserve attention.Doing mathematicswell re-
quires selectivity.
The developmentof numbertheorywithinsettheoryis knownas a
reduction of number theoryto set theory.More generallywhen the
patternor patternsof one theoryoccur withinthe patternor patterns
of another theory,the formeris reducible to that of the latter. Al-
though the existenceof reductionsis mathematicallyinteresting,they
have had littleeffecton the practiceof general mathematicsitself.Most
books on set theorydevelop the natural numbersin termsof sets,but
books on numbertheory,topology,algebra or analysisbegin withboth
numbers and sets. Nor do set theoristshave an atypical interestin
ontological economy, although probably some has rubbed off from
their philosophical colleagues. Instead we find numbers reduced to
sets in set theorybecause, as mentionedearlier,doing so promotesset
theoryitself.
There is another phenomenon whichhas greatlychanged math-
ematicsand whichcould be called a reduction.This is the settheorizing
of mathematics.I have in mind the use of the language of settheoryas
the background language of workingmathematicsand the attendant
objectification(or, in my terms, positionalization)of mathematical
structures.This is reflected in the differentemphases of pre-set
theoreticand post-settheoreticmathematics.Courant and Robbins'
famous book, What is Mathematics?, illustratesthe pre-set theoretic
approach to mathematics(althoughitdoes containdiscussionsof sets).
The discussionin the book is almostexclusivelyabout the elementsof

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
 MATHEMATICS AS A SCIENCE OF PATTERNS 541

traditional mathematicalstructures-the various numbers, geomet-

rical objects and functions.Most of its theoremscan be formalizedin
the firstorder theoryof these objects.Compare thiswithrecentbooks
on topologyor algebra. Here the emphasis is upon the various struc-
tures themselves,what their substructuresare, how they can be ex-
tended, and to whattheyare isomorphic.To formalizesuch theorems
we must use theories in which structuresthemselvescan be taken as
individuals,and thishas been greatlyfacilitatedby the introductionof
set theoreticmethods. Nonetheless,every mathematicaltheorydeals
with one or more structureswhich it describes by showing how the
elementsof each are internallyrelated.Post-settheoreticmathematicsis
no exception.It speaks of structuresonlybytakingthemas elementsof
a "larger" structure-the hierarchyof sets.
We now have two kinds of reduction. With the firsta pattern
occurs in another and the theoryof the formeris reiteratedin the
theoryof the latter.Withthe second patternsthemselvesare identified
withelementsof anotherpattern,therebyenablingus to achieveresults
about these patternswhichwere not even stateablepreviously.It is the
second sort of reduction which has had a genuine effectupon the
practiceof mathematics.
Philosophershave been fascinatedby the firstkind of reduction.
They wanted to knowwhethersome of theseestablishedthatnumbers
are sets. Now we see thattheymightalso wonder whetherthe natural
number sequence is a set.
Both questions are ambiguous. Taken one way they simplyask
whethercertainsetsare sets.To be interestingtheyshould be taken as
asking whethercertain things known to us before the advent of set
theoryare sets.Since,on myview,setsare positionsin a certainpattern,
thequestionsask whetherthingswhichare notknownto be positionsof
that patternare positions in it. The question make primafacie sense
since it appears that the entitiesin question mightbe positionsof the
patternin question. But how could we findout? Not by mathematical
means. No new theoremswillsettlethem,nor willnew axioms. These
questions are not like the question of the generalized continuumhy-
pothesis,"Is each aleph the cardinalityof the power set of itspredeces-
sor?", despite their grammaticalsimilarity.Taken in the non-trivial
sense they are not in the language of a single mathematicaltheory.
They can, to be sure, be "resolved" by extending the notion of set
theoryand adding new identitiesor non-identitiesto thislanguage as
axioms.13 However,even theseresolutionscannotbe decided by math-
ematicalevidence, since it makes no differenceto the contentof math-
ematicswhetherwe do settheorywithnumbertheoryon top of it,so to
speak, or whetherwe do it withnumber theoryreduced to it.

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
 542 NOUS

Of course, it does make a differenceto the elegance of our mathe-

matical theories. Elegance is highlyprized in mathematicsbut it has
nothingto do withmathematicaltruth;afterwe have settleda given
mathematicalquestion it is fittingto ask whetherwe can do it more
elegantly.From the mathematicalpoint of view the questions about
numbers, the natural number sequence and sets are not questions
about factsbut ratherquestions about how we should presentmathe-
matical theories.14 This, in my opinion, is the only non-trivialsense
whichcan be givento thesequestions.The viewthatthereis a factofthe
matterhere is equivalent to assuming that given any object x and a
positionof in a patternyeitherx = yor x 7 y This assumptionfailswith
respect to my conception of patterns.
Mathematicalreductionscannot tell us what numbers,functions,
sets,etc. are in any factualse'nse,but theycan lead us froma defective
conception or theoryof mathematicalobjects of a given kind to more
adequate ones. For example, to someone who objects to functionson
the grounds that they involve intensional notions such as rules, we
mightrespond that functionsare nothingbut sets of ordered pairs.
There is nothingwrongwiththissortof response,so long as we realize
thatwe are not claimingthatfunctionsconceived as rules are the same
as functionsconceivedas sets.Whatwe are doing is pointingout thatwe
can discard worriesabout the intensionalityof functionsby using an
extensionaltheorywhichcan serve the mathematicalends of the origi-
nal one.
Something like this can be said about my proposals for pattern
theory-although I do not thinkof myviewas a mathematicaltheory.
Those who findpatternsand positionsmysteriousmightfindreducing
themto pure setsreassuring.Of course, givenmyviewthatmathemat-
ical objects are positionsin patterns,thiswould be another face to the
reductionof mathematicsto set theory.In both cases we would switch
theorieswiththerebeing no factof the matteras to whethertheold and
new theorieshave the same ontology.On the other hand, I would still
maintainthatthereductionof patternsto pure setsis bothunnecessary
and undesirable. Unnecessary,because myconceptionof positionsand
patternsis clear enough as is; undesirable,because thereductionto one
"big'"patternwould tend to push therelationshipsamong patternsand
theirpositionsout of focus. And the discomfortinginterpatternques-
tionscannotbe avoided bythe move to one patternanyway,because we
willstillwonder how itrelatesto the old patternsor how itwillrelate to
patternsdiscovered in the future.If I am rightabout patterns,these
worries are misguided. If I rightabout mathematicalobject's being
positionsin patterns,then so too are worriesabout whethernumbers
are really sets.

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
MAIHEMATI(CS AS A SCIENCE OF PATTERNS 543

4. REFERENCE

We describe patternsin a numberof ways.One wayconsistsin stating

how manypositionsa patternhas and how theyare related. Anotheris
to presentan instanceof a patternand add thatthe patterninstantiated
is the patternwe have in mind.A thirdconsistsin introducinglabels for
positionsand statinghow the positionsare relatedin termsof these. No
descriptionof a pattern-not even one by means of a categoricalset of
axioms--willdifferentiate the patternfromitsoccurrenceswithinother
patternsor itsotheroccurrencesin isolation;Similarlya descriptionof
a square will not differentiateit from other squares to which it is
congruent.This is entirelyin accord withmypointof view.For itis not
a question of whether one pattern or many patterns are being de-
scribed, although we cannot tell which. There is just one factof the
matterconcerning how many patternsare there.15
*Mathematicaltheories describe various mathematicalpatterns.16
Their truthconsistsin doing so accurately.But if patternsand their
positionscannot be fullyidentifiedhow is a correspondence sufficient
for truthestablished between a theoryand its pattern?17To fix our
ideas let us ask how number theory describes the natural number
sequence.
The standard approach to referenceand truthassumes that it is
possible to assign exactlyone position (number) to each numeral and
do so withoutqualification.Truth is then characterizedrecursivelyin
the manner of Tarski. This approach failsforpatterns,forit presup-
poses thatwe can make unqualified identificationsof positionswhile I
claim that identificationscan only be made withina pattern.
Nicholas White in [13] developed a view which in effecttreats
numerals as functionaltermswithargument places for progressions.
Instead of "3" we have "3(P)" which denotes the fourth object in
whatever progression is chosen as a value of P. The view can be
extended to otherkindsof structuresas well withthe result,forexam-
ple, that there are progressionsP and set theoretichierarchiesS for
which 1 (P) = { 0 } (S) whileforother P and S thisfails.This approach
requires a common universe for all mathematicalstructures-or at
leastrequiresa universeof all progressions-and functionsdefinedon
it. It thus sees factsof the matterwhere I see none.
HartryField's [4] comes much closer to myview. He allows num-
erals to refer multiplyby using the notion of partial predicate. The
numeral "0", forexample, partiallyrefersto 0, to 0, to { 0 }, and to the
firstelementof everyprogression.The other numerals partiallyrefer
to the correspondingelementsof progressionswithitsbeing required
thattheybe linkedin theirpartialreferenceassignmentsso thatdistinct
numeralspartiallyreferto differentelementsof the same progression.

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
 544 NOUS

Thus although "2" partiallydenotes both { {0}} of (0, {0}, {{ 0}}, . . )

and {I{{0} } } of({ 0}, { {0}}, I{ {0} } },. . .),theobviouscontraditions
are
avoided. The relation symbols of number theory are assigned the
correspondingrelationsin a progressionas theirpartialdenotations.A
structurefor which this obtains is said to partiallyaccord with the
semanticsof number theory.A sentenceof numbertheoryis said to be
true (respectively,false) ifand onlyifit is true (false) in everystructure
whichpartiallyaccords withthe semanticsof numbertheory.Truth in
a structureis understood in the Tarski sense. It thusturnsout thatthe
truthsof the language of number theoryare exactlythose whichhold
in everyprogression.Furthermore,in a mixed number theoretic-set
theoreticlanguage, thesemanticswillcountsentencessuch as "1 = {0}"
as neithertruenor false-so long as both the numbertheoreticand the
set theoretictermsare taken as primitive.Thus Field sees truth,falsity
and lack of truth-valueexactlywhere I do.
Despite this Field sees factsof the matterwhere I see none. The
partial reference relation is one which holds absolutely between a
numeral and elementsof progressions.Now suppose that"2" partially
denotes an elementx of a progressionP1while"3" partiallydenotes an
element y of a progression P2. Field's semantics requires that x be
distinctfrom6 if P1 and P2 are the same progression.It thus requires
thattherebe a factto the matterconcerningwhetherx = y even when
P1and P2are notincluded in one and thesame structure.In thisrespect
our views differ.'8
The onlyalternativeleftthatI knowof is some versionof referen-
tial relativity.I thinka variantof Quine's viewsfitswell withmine,but
before I explain how I wantto address the question of whya theoryof
referenceis needed to complete my account. The answer, I thinkis
fairlystraightforward. I espouse realism,thusI believe thatmathemat-
ical statementsare trueor falsein virtueof the featuresof an objective
reality.So far I have presented an account of what that realityis. A
"theory"of reference is needed to explain the connection between
mathematicalstatementsand that reality.'9
Let us suppose that we have fixed an occurrenceof the natural
number sequence. Then, relativeto takingthisoccurrence as fixed,a
standard Tarskian approach to truthand referenceis possible. The
symbol"O" is assigned the firstpositionin the sequence, the successor
symbolis assigned the successorrelation,the quantifiersassigned the
totalityof the positionsin the sequence as theirrange,and so on. Truth
is defined recursivelyin the familiarway.The set of numbertheoretic
statementstherebyverifiedwillbe the same no matterwhatoccurrence
of the natural number sequence is fixed. Thus determiningnumber
theoretictruthvia one occurrencedeterminesit forall occurrences-
except thereis no factof the matterconcerningwhetherthereare one
or many natural number sequences in question.

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
MATHEMATICS AS A SCIENCE OF PATTERNS 545

Next consider a language in which number theoreticlanguage is

embedded. To be specificlet thisbe the language of set theory.There
are two cases to distinguish.In the firstthe number theoreticsymbols
are primitivesymbolsof the larger language. Then we must fix an
occurrence of a set theoretichierarchyin which the natural number
sequence occurs "at the bottom" so that the natural numbers can be
taken as ur-elements.We then assign these positionsto numerals and
othersto set symbolsand proceed as usual. This givesrise to combined
truthsof numbertheoryand settheorysuch as "34EI2,3, 4}" but creates
no new ones denotingthe numbersalone such as "2E3". No one has any
complaintsabout this.
In the second case the numbertheoreticsymbolsare defined. The
numerals thus come to referto a particularoccurrence of the natural
numbersequence within(not "at the bottom")of the occurrenceof the
set theoretichierarchyfixed.20Some truthsapparentlyinvolvingjust
the numbers-such as "2E3"-are therebycreated. My positionhere is
that this is an edifice of a reduction of number theoryto set theory.
Thus, we should not conclude thatthese are new truthsinvolvingthe
numbers alone. They are rather truthsabout positions in an occur-
rence of the natural number sequence in an occurrence of the set
theoretichierarchyand thus ultimatelytruthsabout sets.
Until now I have spoken ratherstraightforwardly about carrying
out a semanticsfornumber theoryonce an occurrence of the natural
number sequence has been fixed. But no matterhow thorough our
effortsto fixan occurrenceof a pattern,thereis, in general,no factof
the matterconcerningwhethertheoccurrencesupposedlyfixedis or is
not the same as some otheroccurrence. If we fixan occurrenceof the
natural number sequence by identifyingit as a certain subpatternof
some more inclusivepattern,then we will be able to distinguishthis
occurrence from many others withinthe larger pattern.That is why
giving a formal semantics for number theorywithina set theoretic
frameworkappears to make the numerals referabsolutely.We must
not forgetthat this is relativeto takinga fixed occurrence of the set
theoretichierarchyfor granted.
We attemptto fix an occurrence of a patternby using a repre-
sentationalscheme-a diagram,description,or theory-to pick it out.
We may even point to instancesof the patternas well. But referential
relativityentersat thispoint,foreven our best effortsdo not forestall
questionsconcerningwhethertheoccurrencefixedis the same as other
occurrence (in a perhaps more inclusivepattern).2'Thus referenceis
relativetotakingan occurrence asfixed. This simplymeans that
ofa pattern
our semanticsfora theoryof a patternis predicatedupon takingat face
value the representationalscheme throughwhich we have attempted
to fix an occurrence of the pattern.22The demand that we achieve

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
 546 NO(US

absolute referenceby fixingoccurrencesof patternsmore fullymisses

mypoint thatthereis no more to do. For thereis no factof the matter
whetheran occurrence of a patternis or is not the same as another
except when theyare both subpatternsof the same pattern.23
Since part of the "theory"of reference for a language involves
fixingthe range of itsquantifiers,a corollaryof referentialrelativity
is
ontological relativity.In my case thiscomes to the doctrinethat only
relativeto takingan occurrenceof a patternas fixeddoes itmake sense
to speak of the ontology (positions) of a theoryof a pattern.Where
Quine proclaimsthatwhatmakes sense are statementsconcerninghow
one theorymaybe interpretedin anotherI have the parallel claim that
what makes sense are statementsthatone patternoccurs in another.24
It mayseem I have done away withthe worldor withthe viewthat
truthis a matterof the way in the world is. But I do not deny the
existence of patterns,nor do I claim that how a pattern is fails to
determinewhat is true in itstheories.My claim is thatthereis enough
slippage betweenour theoriesof patternsand the patternsthemselves
to affectreference.But itdoes notaffecttruth:thetruthsof a theoryof
a pattern are invariantunder all reinterpretationsin patternscon-
gruent in it.

5. CONCLUSION

When Frege-and Dedekind too-asked what numbers were, mathe-

maticshad neitheran adequate characterizationof the naturalnumber
sequence nor an axiomaticbasis fornumbertheory.Their questionwas
in facta demand thatmathematicsdevelop an adequate conceptionof
thenumbers.One wayto do thisis to clarifythe foundationsof number
theoryitself,and that clarificationwas a direct result of their work.
Anotherway is to identifynumberswithentitiesforwhichwe already
have an adequate conception. Frege triedthistoo by identifyingthem
withsets,but enough doubt has been cast upon our conceptionof sets
thatsuch an identificationis not regarded as conceptuallyclarifyingin
itself.
This paper has takenyetanotherapproach to the questionof what
mathematicalobjects are. The problem is no longer one of clarifying
our mathematicalconceptions. Rather it is a problem of finding a
philosophicalinterpretationof them. My suggestionis thatmathemat-
ical objects are positionsin a patternis not intended as an ontological
reduction.(It could lead to one, of course, but ifwe reduce patternsto
sets it would be the old one with a detour added.) My intentionwas
instead to offeranother way of viewingnumbers and number theory
whichwould put the phenomena of multiplereductionsand ontologi-
cal and referentialrelativityin a clearerlight.Myhope is thatwhenthey

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
MATHEMATICS AS A SCIENCE OF PATTERNS 547

are seen to arise in a fairlyobvious way withrespectto patterns,then

they will seem more comprehensible with regard to mathematical
structures.25

REFERENCES

[1] Benacerraf, P., "What Numbers Could Not Be", PhilosophicalReview,74(1965):

47-73.
[2] , "Mathematical Truth",JournalofPhilosophy, 70(1973): 661-79.
[3] Corcoran,J., "On DefinitionalEquivalence and Related Topics", Historyand Phi-
losophy ofLogic, 1(1980): 231-34.
[4] Field, H., "Quine and the Correspondence Theory", PhilosophicalReview,
84(1975): 23-50.
[5] Jubien, M., "Ontology and MathematicalTruth", Noius,11(1977): 133-50.
[6] Kitcher,P., "The Plightof the Platonist",Noius,12(1978): 119-36.
[7] Parsons,C., "Ontologyand Mathematics",Philosophical Review,80(1971): 151-76.
[8] Quine, W. V., "OntologicalRelativity"inOntological and OtherEssays
Relativity (New
York: Columbia UniversityPress, 1969).
[9] Resnik, M., "More on Skolem's Paradox", Noius,3(1969): 185-96.
[10] "Mathematical Knowledge and PatternCognition",CanadianJournalof
Philosophy, 5(1975): 25-39.
[1-1] , Fregeand thePhilosophy (Ithaca: Cornell UniversityPress,
of Mathematics
1980).
[12] Steiner, M., Mathematical Knowledge(Ithaca: Cornell UniversityPres, 1975).
[13] White, N., "What Numbers Are", Synthese, 27(1974): 111-24.
[14] Wilson,M., "The Double Standard in Ontology",Philosophical Studies,forthcoming
in 1981.

NOTES

'I am using the term"platonist"as it is commonlyused today in the philosophyof

mathematicsto stand forrealismwithrespectto abstractmathematicalobjects; I do not
mean to commit myselftherebyto Plato's theoryof ideas. My reasons for espousing
platonismare varied and complicated; some of them are found in Resnik [9], [10], and
[11]. In [2] Paul Benacerrafpresentsa strongargumentforgivinga uniformsemantics
formathematicsand therestof science: ifmathematicaltermshave one kindof semantics
whileotherscientifictermshave anotherthena special semanticswillstillbe required for
sentences containing both kinds of term. This argument can be strengthened,as I
observed in discussionwithBenacerrafand othersat the Greensboro conference.For a
uniformsemanticsformathematicsand science willbe required to obtain an account of
inferencesinvolvingboth mathematicaland other scientificsentences. In [11] p. 62 I
trace thisto Frege. Once the need fora uniformsemanticsand a face-valueaccount of
logical form is granted, the weight of the Tarski approach quickly carries one to
platonism.
2Benacerraf [2] and [1] contain statementsof the firstand second problems,
respectively.Steinerdiscusses the formerat lengthin [ 12] whileJubien[5] and Kitcher
[6] elaborate on the latter.See also Maddy's paper in this issue.
3Cf.Jubien [5], Kitcher[6], Parsons [7], Benacerraf [1], Steiner [12].
4A fullerand perhaps differingaccount willappear in a paper I am writingfor a
1982 AmericanPhilosophicalAssociationWesternDivisionsymposiumon mathematical
knowledge.It willappear in Nous XV, 1 (March 1982). A previousattemptappeared in
my [10].
5Not all patternsare knownthroughexposure to theirinstances-thus the qualifi-
cations.
6One of these was given by GilbertHarman in his commentson the Greensboro
Conference version of this paper.

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
 548

7I have added the parentheticalclause to parallel the model theoreticdefinitionof

structure.Strictlyspeaking,distinguishedpositions,operationsand characteristics which
cannot be picked out using the relationsof the patternitselfmust be viewed as distin-
guished byseeing the patternas embedded in anotherpattern.Thus (N, S, 0) containsa
distinguishedposition,viz. 0, which can be picked out using the successor relation S.
However, distinguished0 in (Rat, <, 0) is dependent upon viewingthat patternin a
background patternin which 0 can be distinguished"lnternally".
81f second order or set theoreticdefinability
is used, thena theoryof typesmustbe
observedso thatno settheoreticor higherorder entitiesare added to theoriginalpattern
as new positions.Otherwise,as Mark Wilson has pointed out to me, infinitestructures
would occur withinfiniteones.
91n (Rat, +, ) we can define 0 and 1 by

(y)(O = y y + y = Y)
(y)(l = y y * y = y),

and then define N and S. To obtain (Rat, +, ) within (N, S) we use second order
definitionsto define addition,multiplicationand exponentiationover the naturalnum-
bers and defineordered pairs (a, b) as 2a. 3b. The rationalscan thenbe developed in the
usual way as ordered pairs of natural numbers.
' ?A theoryS is interpretablein a theoryT withthe same underlyinglogicjust in case
thereis a setof definitionsof the primitivesofS in T call this,DST, whichwhenadded toT
yieldsthetheoremsofS (as theoremsofT + DST). Consider twosecond order theoriesN'
and N + whose primitivesare respectively"0", ""' and "19,9 "+". The axioms of N' are:

(x)(x'
W 0)
(x)(y)(x' = y'D x = y)
(F) [FO . (x) (Fx D Fx') D (x) Fx],

whilethose of N + are exactlythe same except ""' is replaced everywhereby "+ 1". N' is
interpretablein N+ via the definition:x' = x + 1. N+ is interpretablein N' via the
definitions:

1) 1 = 0'
2) x + y = z (F) [(x) FxOx . (u)(v)(w)(FuvwDFuv'w'DFxyz].

This implies that every model of N' has a model of N+ occurringwithinit and con-
versely.Yet these two theoriesare not essentiallythe same: N' is complete as to conse-
quences and categorical,N+ is neither;N' + DN + N' entails"0 + 0 = 0", but thisis a
theoremof neitherN+ nor N+ + DN'N+. These points are due to John Corcoran.
"Our two theories N' and N+ fail to be definitionallyequivalent, by the way,
because (2) cannot be proved as a theoremof N +. A counter model can be obtained by
interpreting"x + y" as the functionf such thatf(x, 1) = x + 1, f(x,y) = 1 if y : 1. See
Corcoran [3] for furtherdiscussion.
12Mark Wilsongrapples in [ 14] withthe problemof determiningwhen twotheories
have the same ontologyand proposes definitionalequivalence as the solution.The line I
willtake eschewsall but verygeneral intertheoretic comparisonsof ontology.For exam-
ple, itmakes no sense to ask whetherthe positionsof twopatternsare thesame butitdoes
make sense to ask whetherone or both contain infinitelymany positions.
13Forexample, the question about the numberscan be "answered" affirmatively by
adding "0" and "'" to the language of set theoryand the axioms "0 = Q,', "(x)(x' = x U { x
})", or negativelybyadding thesame symbolsand theaxiom "(x) (xEND-(3y) (yEx.xAL/i)"
where "N" is given its usual set theoreticdefinitionin termsof "0" and "'".
'4Since there is no factof the matterhere, no factsare made by stipulationeither.
150f course,withina largerpatternwe can pickout manydifferent occurrencesof a
patternjust as we can pick out many stars in the American Flag Pattern. We do so,
however,by distinguishingpositionswithinthe larger pattern.
16J have in mind "concrete"mathematicaltheoriessuch as number theory,analysis
and settheory."Abstract"theoriessuch as group theory,topologyand, perhaps modern

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
 MATHEMATICS AS A SCIENCE OF PATTERNS 549

geometrydo notattemptto describea singlestructurebut deal withclasses of structures.

An "if . .. then" or deductivistapproach is appropriate for these theories.
17Non-categoricaltheories present a number of problems. First order number
theoryseemsto be a case wherewe had a clear idea of the patternand came to realize that
the theorydid not completelydescribe it. We then used other devices to rule out the
unintendedpatterns.In thecase of settheorywe also knowthatwe lack a fulldescription
of the set theoretichierarchybut remain unsure of which of the various structures
describablebyset theoryis the "intended"one. Note thatbothintendedand unintended
models of number theoryand set theorycontain congruent cores. I view our initial
versionof numbertheoryand our currentsettheoriesas attemptsto describethesecores.
That theyweresatisfiedbydivergingextensionsofthecores initiallyescaped our notice.
18Both Field's and White'sapproach are criticizedin Kitcher [6].
191have put scare quotes around "theory"because on myview thereare no factsof
the matterconcerning the referencesof mathematicalterms. I will not address the
epistemologicalquestionsabout reference-such as, "How is it possible forus to referto
positionsin patterns."I hope to deal withthese in the paper mentioned in note 4.
20MarkWilsonin [ 14] countssettheorywiththenaturalnumbersas ur-elementsand
pure set theory as ontologicallyequivalent. My view is that these are examples of
equivalentbut noncongruentpatterns.As maybe evident,traditionalquestions about a
theory'sontologydo not find a place in my account.
ilIn Quinean terms,even a complete and categoricaltheoryis reinterpretablein
other theories.
22CfQuine: "In practice of course we end the regress of coordinate systemsby
somethinglike pointing.And in practicewe end the regressof backgroundlanguages, in
discussionsof reference,byacquiescingin our mothertongue and takingitswordsat face
value" [8] p. 49.
23HartryField offersan objection to Quine's referentialrelativitywhich is predi-
cated upon reading Quine's formulation"reference is relative to some background
language" as meaningthatreferenceinsteadof being a dyadicrelation"x refersto y" is a
triadic relation "x refers to y relative to BL". (Actually Field reads Quine as saying
referencetakes the form"x refersto y relativeto translationmanual M". See [4] pp.
206-208.) The objection is based upon a misunderstanding,I believe. Quine and I are
arguingthatreferencefailsto make sense when an occurrenceof a pattern(background
language) is not accepted as fixed (taken at face value). This is like pointingout that a
personwho is notof a certainage cannotbe legallybound bya contract.In eithercase one
mightsay "relativeto . . . ", for example, "relativeto your being of age you are legally
bound to pay Jones $100" or "relative to taking English at face value the French
expression 'deux' refers to two". In neither case is an increase in the degree of the
relationshipintended. Making explicita suppressed degree in a predicate such as " is
president" can be clarifying,but it would be pointless to demand that Quine revise
"'deux' refersto two" to read "'deux' refersto two in English". If you did not already
understand the expression "'deux' refersto two" as an English sentence, adding the -
words "in English" would not clarifyit for you.
24Whenone recalls that for Quine a theoryis already interpreted,the difference
betweenour viewsis not as greatas mytalkof patternsmightlead one to believe. Indeed
in conversationand commentson an earlier draftof this paper, Quine has noted that
similarityof my positionsin patternsto his objects of a theory.It is possible to adduce
considerationsto show thatnon-mathematicalobjectsachieve theiridentityonlythrough
theirrelationshipsto otherobjectsand thenargue thatthesetoo are positionsin patterns.
This would lead to an extensionof myviewsto include a complete theoryof referential
relativity,although the route to thisconclusion would be ratherdifferentthan Quine's.
For worriesabout referencesto rabbitsvs. referencesto rabbitstages are not worries
about multipleoccurrencesof a pattern.Rabbitparts,rabbitstagesand rabbitsmightbe
positionsin non-congruentpatterns,whichoccur withineach otheror are equivalentand
therebyproduce relativityphenomena.
25Thispaper has developed in spurtsover a numberof years;thus,I owe a large debt
of thanks.In oral commentson Benacerrafs [2] Oswaldo Chateaubriand remarkedthat
we can perceivecertainabstractentitiesthroughtheirinstances.AlthoughI was familiar

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions
550 NOUS

withthe view that mathematicsis about structurethrougha number of sources, these

remarkssuggestedto me a footholdforan epistemologyforplatonism.Benacerrafhad
hintedhimselfthat mathematicsmightbe about structurein discussingthe problem of
multiplereductions.I believe thatI was the firstto propose a structuralist
approach to
both the epistemologicaland ontological difficultiesof platonismin [10]. Since then I
have benefitedfromcomments,encouragementand discussionwithmanyindividuals:
James Anderson, Catherine Anderson, Charles Chihara, James Fetzer,Jane Fleener,
Nicholas Goodman, Richard E. Grandy, Gilbert Harman, Penelope Maddy, Vann
McGee, Richard Nunan, W. V. 0. Quine, JayF. Rosenberg,FrederickSchmitt,LaVerne
Shelton, Brian Skyrms,Mark Steiner,Mark Wilson,Paul Ziffand the membersof my
National Endowmentforthe HumanitiesSummerSeminar.I wantto givespecial thanks
toJohn Corcoran and Philip Kitcherwhose manylettersand conversationshave been of
crucialassistance.I should add that,to myknowledge,fewof thoselistedthinkthatI have
gottenthingsstraightor right.Earlier versionsof theseideas have been presentedto the
following:The BuffaloLogic Colloquium, Davidson College, the NorthCarolina Philo-
sophical Society,the Universitiesof North Carolina at Greensboro and Chapel Hill
Conference in the Philosophy of Mathematics,and the philosophy colloquia of the
Universitiesof North Carolina at Chapel Hill and California at San Diego.

This content downloaded from 198.105.44.150 on Sun, 1 Dec 2013 11:53:52 AM

All use subject to JSTOR Terms and Conditions