You are on page 1of 258

Scientific / thought,

Some underlying concepts


methods L
and procedures,
Mouton/ Unesco
Paris The Hague
mcmlxxii
Scientific thought
Library of Congress Catalog Card Number: 72-79987
0 Unesco 1972, Place de Fontenoy, Paris
Printed in the Netherlands
Preface
We have al ways f ound that whenever a postul ated symmet ry pri nci pl e was
appeari ng to fail in natural phenomena, thi s must be due to some sti l l
deeper symmet ry, withwhi ch i t must be in confl i ct. We may, at a gi ven
time, fail to comprehend the aestheti cs of nature. When, however, the full
and fi nal pi cture emerges one has i nvari abl y f ound that the symmetri es
thi s exhi bi ts are prof ounder still.
Thi s extract f romProf essor Sal am s contri buti on to thi s book neatl y
sums up the hi story and one of the purposes of sci ence and phi l osophy.
There is al ways an ul ti mate beyond the ul ti mate: perhaps the ul ti mate will
at last be reached when man gets to i nfi ni ty.. .
Moreover, sci ence i s many- si ded, and each tends toregard i t f r omthe
standpoi nt of hi s own parti cul ar experi ence and i nterests. The Bronze
Age and the I ron Age, cl earl y, werenamed by technol ogy. But, asked i n
1886 whether the ni neteenth century woul d be known as the century of
i ron, of st eamor of el ectri ci ty, the emi nent physi ci st Bol t zmann had no
hesi tati on i n decl ari ng that i t woul d be known as the century of Dar wi n.
The phi l osophers, on thei r si de, have al ways tried to thi nk beyond
sci ence, and to provi de a f ramework whi ch i nvi tes the human mi nd to
thi nk about sci ence and, thereby, about itself. Ari stotl es categori es
provi ded an earl y systemati zati on whi ch Kant , di gesti ng new knowl edge
acqui red meant i me in mathemati cs and many other di sci pl i nes, trans-
f ormed. Hegel attempted a synthesi s in whi ch all the parti al and
contradi ctory phi l osophi es of his predecessors woul d be contai ned and
transcended. .. one thi ng i s certai n: as thi nkers, sci enti sts and phi l osophers
conti nue to add thei r contri buti ons, the sumof human knowl edge
i ncreases, and so, l i kewi se, does the l ayman s difficulty of keepi ng apace,
Not al one have there been tremendous devel opments in physi cs,
chemi stry, bi ol ogy and mathemati cs, but whol l y new di sci pl i nes, such
as cyberneti cs, have appeared. Mass communi cat i on techni ques have
VI Preface
transf ormed our concepts of i nf ormati on. Economi c pl anni ng is the
rul e innearl y all countri es. The moon and other pl anets are bei ng expl ored.
But, accordi ng to the Chi nese sayi ng, there is al ways a fi rst stepi na
j ourney of ten thousand miles.. .
Thi s book i s a col l ecti on of arti cl es by emi nent sci enti sts who wer e i nvi ted
by Unesco to expl ai n i deas, met hods or procedures whi ch underl i e many
of the extraordi nary di scoveri es and devel opments of the twenti eth
century. The vi ews expressed are those of the authors, and not necessari l y
those of Unesco. The arti cl es are obvi ousl y not exhausti ve, and inmost
cases do not deal wi th other vi ews that may be hel d on the same subj ect.
But each contai ns a bi bl i ography so that readers speci al l y i nterested i n
any parti cul ar arti cl e can see where i ts subj ect matter can be studi ed in
greater detai l .
In publ i shi ng thi s book, the Unesco Di vi si on of Phi l osophy bel i eves
that readers will wel come i t as a seri es of authori tati ve statements,
conveni entl y brought together between t wo covers, on di verse subj ects
whi ch have in common thei r i mmense i mportance i n moder n scientific
thought.
Contents
Preface
Not es on the authors
1. ANDRZEJ MOSTOWSKI
2. J EAN PI AGET
3. MI HAJ LO D. MESAROVI C
4. ABDUSSALAM
5. LADI SLAV TONDL
6. YEHOSHUA BAR- HI LLEL
7. SATOSI WATANABE
8. HERBERT STACHOWI AK
9. STAFFORDBEER
10. ARNOLD KAUFMANN
11. NI COLAI N. VOROBYEV
12. MARI OBUNGE
Sets
The concept of structure
Systems concepts
Symmet r y concepts and the
f undament al theory of matter
Si gn
Language
I nf ormati on
Model s
Management incyberneti c t er ms
Concerni ng the noti ons of operati on
and opti mati on
Game theory
V
IX
1
35
57
69
89
107
129
145
167
187
205
Metatheory 227
Notes on the authors
YEHOSHUA BAR- HI LLEL. Bor n Vi enna 1915. Doctorate (1949) from the
Hebr ewUni versi ty of J erusal em. Research and teachi ng in the Uni ted
States (Massachusetts I nsti tute of Technol ogy; Cal i forni a and Ann
Arbor), Europe (Konstanz and Berlin), and I srael (teachi ng l ogi c and the
phi l osophy of sci ence at the Hebr ew Uni versi ty).
Past Presi dent of the I nternati onal Uni on of the Hi story and Phi l osophy
of Sci ence; member of the I srael Academy of Arts and Sci ence.
Publ i cati ons i ncl ude: Language and information; Aspects of language:
essays in philosophy of language, linguistic philosophy and methodology of
linguistics; withA. A, Fraenkel and A. Levy, Foundations of set theory.
STAFFORDBEER. Vi si ti ng Prof essor of Cyberneti cs at Manchester Uni ver-
si ty, and of General Syst ems at the Open Uni versi ty, London. Presi dent
of the Soci ety for General Syst ems Research, Washi ngt on; member of t he
Uni ted Ki ngdomAut omat i on Counci l . Consul tant to government and
i ndustry, and a regul ar broadcaster on radi o and tel evi si on.
Publ i cati ons i ncl ude: Cybernetics and management; Decision and control
(Wi l ey- Lanchester Prize); Managenient science; Brain of the $rm;
P1ntfor.m for change (inpress).
MARI O EUNGE. Born Buenos Ai res 1919. Doctorate i n theoreti cal physi cs.
Research in atomi c and nucl ear physi cs. Has taught physi cs and
phi l osophy i n Argenti na and the Uni ted States. Now Professor of
Phi l osophy and Head of the Foundati ons and Phi l osophy of Sci ence Uni t,
McGi l l uni versi ty, Montreal .
Publ i cati ons i ncl ude: Causality; Metascientijic queries; Cineincitica del
electrn relativista; Intuition and science; The myth of simplicity; Foun-
dations of physics; Scientijic research.
x Notes on the authors
ARNOLD KAUFMANN. Professor at the I nsti tut Nati onal Pol ytechni que
of Grenobl e, France. Has taught at the Ecol e Nati onal e Supri eure des
Mi nes, Pari s, and at the Ecol e de l'Air, Sal on- de- Provence. Consul tant
on vari ous Unesco proj ects.
Publ i cati ons i ncl ude: Les cadres et la rvolution informatique (Centi
Pri ze, 1970); Des sous-hommes et des super-machines (wi th J . Pze);
L'inventique - nouvelles mthodes de crativit (wi th M. Fusti er and
A. Drevet); Mathmatiques nouvelles pour le recyclage des parents (wi th
G. Cul l mann); Introduction la combinatorque; L'homme d'action et la
science; Introduction la thorie des sous-ensembles flous.
MIHAJLO D. MESAROVI C. Professor of Engi neeri ng and Di rector of
Systems Research Center at Case West ern Reser ve Uni versi ty, Cl evel and,
Ohi o, Uni ted States. Conti nui ng to wor k on the mathemati cal theory
of general systems and the appl i cati on of the mul ti -l evel , hi erarchi cal
approach in envi ronmental and soci o- economi c probl ems (in parti cul ar
inthe ' Predi cament of Manki nd' proj ect of the Cl ub of Rome, of whi ch he
is a member).
Publ i cati ons i ncl ude: Theory of multilevel hierarchical systems (wi th
D. Macko and Y. Takahara); Foundations foi a mathematical general
systems theory (wi th Y. Takahara); Systems approach and the city (ed.,
wi th A. Rei sman); Theory of non-numerical problem solving (ed., wi th
R. Banerji); Systems theory and biology (ed.).
ANDRZEJ MOSTOWSKI . Teaches mathemati cs, l ogi c and f oundati ons of
mathemati cs at War saw Uni versi ty (si nce 1946). Now engaged i n
research on model s of extended systems of set theory and of systems of
ari thmeti c.
Publ i cati ons i ncl ude : Constructible sets, with applications.
J EAN PI AGET. Bor n Neuchtel , Swi tzerl and, 1896. Taught at Uni versi ti es
of Neuchtel , Lausanne and Geneva, inSwi tzerl and, and at the Sorbonne.
Pi oneer of research on chi l d psychol ogy. For mer Di rector of the I nter-
nati onal Bureau of Educati on, inGeneva, and vari ous research agenci es.
Honorary doctorates f r omvari ous European and Amer i can uni versi ti es
i ncl udi ng Harvard, Yal e and the Sorbonne. Reci pi ent of several pri zes
i ncl udi ng the Amer i can Psychol ogi cal Associ ati on Prize (1969).
Publ i cati ons i ncl ude: Mmoire et intelligence (wi th H. Si ncl ai r-de
Notes on the authors XI
Zwaart); Le structuralisme; Lpistmologie gntique; Psychologie
et pistmologie. Contri buted I ntroducti on and chapter on Psychol ogy to
Main trends of research in the social and human sciences ( Unesco, 1970).
ABDUS SALAM. Di rector of the I nternati onal Centre for Theoreti cal
Physi cs, Tri este, I tal y. Professor of Theoreti cal Physi cs at the I mperi al
Col l ege of Sci ence and Technol ogy, London. Speci al i st i n parti cl e
physi cs and gravi tati on theory.
Publ i cati ons i ncl ude contri buti ons on the fol l owi ng subj ects: Theory
of groups and the symmet r y physi ci st ( London Mathemati cal Soci ety);
The rol e of symmet ry physi cs; Weak and el ectromagneti c i nteracti ons
(Nobel Symposi um8); Fundament al theory of matter: a survey of resul ts
and methods; Gravi tati onal ward i denti ti es (withJ. Strathdee); Com-
putati on of renormal i zati on constants; Fi ni te field theori es (Rochester
Conference).
HERBERT STACHOWI AK. Professor of the Phi l osophy of Sci ence, Free
Uni versi ty of Berl i n.
Publ i cati ons i ncl ude: Denken und Erkennen im kybernetischen Modell;
Rationalismus im Ursprng. Die Genesis des axiomatischen Denkens.
Edi tor of Cyberneti cs paperback seri es. Co-edi tor of Schriften zur
wissenschaftlichen Weltorientierung (vol s. I-VII); Lexikon der kyberne-
tischen Pdagogik; Teorma (Uni versi dad de Val enci a); Grundlagenstu-
dien aus Kybernetik und Geisteswissenschaft.
LADI SLAV TONDL. Bor n 1924. Research and teachi ng at the Czechosl ovak
Academy of Sci ences on the methodol ogy of sci ence, and l ogi cal seman-
ti cs.
Publ i cati ons i ncl ude: Methodology of experimental sciences: Con-
temporary western philosophy; Problems of semantics; M a n and science;
Scientijc procedures.
NI COLAI N. VOROBYEV. Bor n 1925. Doctorate (1961) i n Physi cs and
Mathemati cs. Professor of Theoreti cal Cyberneti cs; Chai r man of the
Depart ment of Operati ons Research, Central I nsti tute for Economi cs
and Mathemati cs (Leni ngrad Branch) of the USSR Academy of Sci ences;
Professor at the Leni ngrad State Uni versi ty. Speci al i st ingame theory.
Publ i cati ons i ncl ude arti cl es on coal i ti onal games, extrema1 al gebra
XII Notes on the authors
of posi ti ve matri ces, moder n state of the game theory; Fibonacci numbers;
Theory of series; Developinent of game theory (all inRussi an); Grundlagen
der Spieltheorie und ihre praktische Bedeutung; Grundfragen der Spiel-
theorie und ihre praktische Bedeutung.
SATOSI WATANABE. Bor n 1910. Doctorates f r omthe Sorbonne and
Tokyo Uni versi ti es. Taught phi l osophy, el ectri cal engi neeri ng, physi cs
and i nf ormati on sci ences at vari ous uni versi ti es i ncl udi ng Hawai i ,
For dham, Yal e, Col umbi a and Tokyo. Presi dent of the I nternati onal
Soci ety for the Study of Ti me; member of the I nternati onal Academy of
the Phi l osophy of Sci ence.
Publ i cati ons i ncl ude: Knowing and guessing - a quantitative study;
Methodologies of pattern recognition (editor); Frontiers of pattern
recognition (editor).
1 Sets
ANDRZEJ MOSTOWSKI
We shal l deal , i nthis paper, wi th the concept of a set. Our chi ef concern
will be wi th sets occurri ng inmathemati cs, because i t i s i nmathemati cs
that the concept of a set has most appl i cati ons and the moder n phi l -
osophi cal controversi es about the nature of sets have ori gi nated. Whi l e i t
i s true that many sci enti fi c concepts based on the concept of a set do not
bel ong to mathemati cs, the met hods of usi ng sets- i n thei r constructi on
deri ve f r ommathemati cs, and so i t can be safel y stated that the concept
of a set used i n sci ence is not essenti al l y di fferent f r omthat used in
mathemati cs.
It will be seen bel owthat many subtl e di scussi ons concerni ng very
abstract mathemati cal probl ems have thei r ori gi n inthe way we speak of
sets in ordi nary l anguage.
1. SETS I NEVERYDAY LANGUAGE
1. 1 Col l ecti ons
We very often speak of col l ecti ons of obj ects as of enti ti es consi sti ng of
t hese obj ects. Vari ous wor ds are used to denote these enti ti es, e.g.
col l ecti on, group, set, cl ass, aggregate. Somet i mes very speci fi c words are
used to denote certai n col l ecti ons (e.g. constel l ati on), whereas other wor ds
can be appl i ed in vari ous si tuati ons (e.g. a group of demonstrators, a
group of bui l di ngs). We cannot treat all these wor ds as synonymous:
group of demonstrators is correct, but cl ass of demonstrators is hardl y
acceptabl e. Neverthel ess, they all obvi ousl y have a cl earl y rel ated meani ng.
To avoi d l i ngui sti c difficulties weshal l use the neutral wor d col l ecti on to
denote any enti ty (sets, groups, cl asses, teams) consi sti ng of i ndi vi dual
obj ects, These obj ects are cal l ed the el ements of the col l ecti on. Thus we
2 Andrzej Mostowski
have a col l ecti on of footbal l pl ayers (a team), a col l ecti on of stars
(a constel l ati on), a col l ecti on of the i nhabi tants of a gi ven citywhose age
i s bel ow20 (a group), and so on.
1.2 Operati ons on collections
The above use of words is fully sancti oned ineveryday l anguage and there
i s not the sl i ghtest difficulty i n understandi ng i t. It is very hel pful in
cl assi fyi ng obj ects and i nhandl i ng el ementary probl ems i n statistics. Thus,
for exampl e, when prepari ng statistics on city dwel l ers we di vi de the whol e
group i nto vari ous col l ecti ons by age, sex, occupati on, i ncome, etc. Even
at thi s very el ementary stage, vari ous useful operati ons (see Tabl e)
Table
Collection A = a, c, 8, 4, z, y.
Collection B = 6, e, 9,4, x, y.
UnionA,B = a,c,8,4,z,y,b,e,9,4,x,y.
Intersection A, B = 4, y.
Difference A, B = a, c, 8, z.
Difference B, A = 6, e, 9, x.
can be made wi th col l ecti ons, e.g. the union of t wo col l ecti ons (defi ned
as a col l ecti on consi sti ng excl usi vel y of el ements whi ch bel ong to ei ther
of the gi ven col l ecti ons or to both of them), or f ormi ng an i ntersecti on
of t wo col l ecti ons whi ch consi sts of el ements common to both. The di ffer-
ence of t wo col l ecti ons A, B, consi sts of el ement s whi chbel ong to the f ormer
but do not bel ong to the latter col l ecti on. The general l aws whi ch govern
these operati ons f or mwhat is usual l y cal l ed Bool ean al gebra. One wor d
of cauti on i s necessary here: if we use the word col l ecti on or si mi l ar
words ineveryday l anguage, we al ways assume that each col l ecti on has
el ements. Al so we are i ncl i ned to excl ude col l ecti ons whi ch woul d have
j ust one el ement, the reason bei ng si mpl y that the basi c i dea of a col l ecti on
is that of an enti ty consi sti ng of several el ements. I f there is j ust one
el ement there is no need to appl y the process of f ormi ng a col l ecti on.
However , i t is cl ear that such restri cti ons are very i nconveni ent: even in
the si mpl est appl i cati ons to statistics we must extend the concept of a
col l ecti on by al l owi ng col l ecti ons wi thout el ements and col l ecti ons
consi sti ng of j ust one el ement. For i nstance, if we di vi de a popul ati on i nto
Sets 3
groups accordi ng towhether they have certai n properti es, i t may very well
happen that there i s j ust one member of the popul ati on who has a
property or that there are no such member s at all. The l aws of Bool ean
al gebra speci fi cal l y assert the exi stence of an empt y set.
1.3 The principle of abstraction
A theoreti cal l y i mportant met hod of constructi ng col l ecti ons is provi ded
by the so-cal l ed pri nci pl e of abstracti on. Let us consi der a col l ecti on and
a t wo- t ermed rel ati on whi ch may or may not hol d between any t wo gi ven
el ements. We assume that the rel ati on is refl exi ve (i.e. each el ement of the
col l ecti on bears the rel ati on to itself), symmetri c (i.e. if x bears the
rel ati on to y, then y bears i t to x) and transi ti ve (i.e. if x bears the rel a-
ti on to y and y to z, then x bears the rel ati on to z). Such rel ati ons are
cal l edequi val ence rel ati ons. The pri nci pl e of abstracti on says that under
the above assumpti ons the gi ven col l ecti on can be decomposed i nto
subcol l ecti ons, no t wo of whi ch have common el ement s, and so that t wo
el ements are inthe same subcol l ecti on if, and onl y if, they bear the gi ven
rel ati on to each other. The subcol l ecti ons i nto whi ch the whol e col l ecti on
decomposes are cal l edthe equi val ence cl asses of the gi ven rel ati on.
As an i l l ustrati on l et us consi der the col l ecti on of all shi ps and the
rel ati on whi ch one shi p bears to another if they have the same mode of
propul si on. The equi val ence cl asses are then the cl ass of all steam- shi ps,
the cl ass of all sai l i ng-shi ps, etc.
1.4Abstract character of collections
We shal l now di scuss the probl emwhether col l ecti ons whose el ements are
physi cal obj ects are t hemsel ves to be consi dered as such obj ects. The
probl emi s not very preci se because the meani ng of the words physi cal
obj ect is far f r omcl ear, but we shal l not anal yse thi s concept here and
shal l try to use i t as i t is used ineveryday l anguage. It is customary to
requi re that a physi cal obj ect f i l l a well defi ned porti on of space and that
i t be i nsome way connected. I n thi s sense most col l ecti ons are certai nl y
not physi cal obj ects; for i nstance, a constel l ati on is not a si ngl e connected
enti ty. But, even if we admi t that physi cal obj ects need not be connected,
we sti l l cannot consi der col l ecti ons as physi cal obj ects. Otherwi se the
col l ecti on of el ementary parti cl es whi ch are present inthe stars f ormi ng a
4 Andrzej Mostowslci
gi ven constel l ati on woul d be i denti cal withthis constel l ati on, whi ch is
fal se si nce one col l ecti on consi sts of el ementary parti cl es and the other of
stars.
The probl emwasdi scussed by Frege (1884) who showed concl usi vel y
that no col l ecti on i s a physi cal obj ect. Even a col l ecti on consi sti ng of
j ust one el ement is not the same as thi s el ement: this i s parti cul arl y
obvi ous inthe case when the uni que el ement of a col l ecti on i s not a
col l ecti on.
Stati ng the thesi s that col l ecti ons are abstract enti ti es and not physi cal
obj ects we do not take any phi l osophi cal stand concerni ng the real
exi stence of abstract enti ti es. All we want to say i s that all names of
col l ecti ons bel ong to the fami l y of words whi ch do not name physi cal
obj ects. Of course, there are many such words ineveryday l anguage e.g.
all adj ecti ves. The mer e fact that wor ds denoti ng col l ecti ons are nouns
likethe names of physi cal obj ects does not i mpl y that col l ecti ons exist in
the same sense as physi cal obj ects. On the contrary, col l ecti ons are not
physi cal obj ects and exi st onl y int he same sense as do properti es.
1.5 Collections and properties
There are cl ose connecti ons between col l ecti ons and properti es. In
everyday l anguage we can al ways defi ne the f ormer i n f avour of the l atter.
Each col l ecti on i s compl etel y determi ned by a property: the col l ecti on of
Pol es who wer e 51 in 1969 is compl etel y determi ned by the property of
bei ng a Pol e born in1918. We shoul d note, however, that di fferent proper-
ti es may somet i mes determi ne the same col l ecti on: inthe exampl e above
we may consi der the property of bei ng a Pol e born i nthe year inwhi ch the
fi rst worl d war ended.
Let us cal l t wo properti es extensi onal l y equal if any obj ect whi ch
possesses one of t hemal so possesses the other. Two properti es determi ne
the same col l ecti on if and onl y if they are extensi onal l y equal . Thus
rememberi ng the pri nci pl e of abstracti on we may state that col l ecti ons are
(or can be uni quel y correl ated wi th) equi val ence cl asses of the rel ati on of
the extensi onal equal i ty.
The cri teri on for the i denti ty of col l ecti ons al so hel ps to make cl ear
why col l ecti ons are not physi cal obj ects. There mi ght be some temptati on
to thi nk, for exampl e, that a shi p is i denti cal witht he col l ecti on of pl anks
of wood used ini ts constructi on. That this suggesti on is i ncorrect can
Sets 5
be seen by consi deri ng the fact that if a rotten pl ank i s repl aced by a new
one, the resul ti ng col l ecti on of planks i s, by the f oregoi ng cri teri on, a
di fferent col l ecti on f r omthe ori gi nal one. The i denti ty of the shi p,
however, remai ns the same. The battl eshi p Vi ctory whi ch can be seen
today by touri sts is the same shi p as the one whi ch was engaged i nthe
Battl e of Trafal gar, al though doubtl ess i t i s a di fferent col l ecti on of
pl anks. Si mi l arl y, a bui l di ng i s not the same as the col l ecti on of bri cks
and stones whi ch wer e used tobui l d i t. Thi s observati on, t hough i t does
not prove concl usi vel y that a col l ecti on i s not a physi cal obj ect, does
di spense wi th the most obvi ous reason for thi nki ng that i t mi ght be (see
D. Wi ggi ns, 1967, for further di scussi on of thi s point).
1.6 Collections of collections
The abstract character of col l ecti ons mani fests i tsel f i nthe possi bi l i ty of
f ormi ng col l ecti ons whose el ements are t hemsel ves col l ecti ons. Thei r use
is perfectl y sancti oned i n everyday l anguage. Thus we speak e.g. of a
col l ecti on of footbal l t eams or of a col l ecti on of all constel l ati ons or of a
col l ecti on of all sets of si l verware.
Obj ects whi ch are not col l ecti ons are sai d to be of type O; col l ecti ons
whose el ements have type O are sai d to be of type 1; col l ecti ons whose
el ements have type 1 are sai d tobe of type 2 etc. In everyday l anguage we
use only excepti onal l y words denoti ng col l ecti ons of type hi gher than 2.
Also we never consi der col l ecti ons wi th mi xed types, i.e. ones some of
whose el ements have types di fferent f r omthe others. However , there are
no compel l i ng reasons why we shoul d excl ude such col l ecti ons on pri n-
ci pl e, They j ust do not seemto be needed i n si tuati ons descri bed i n
everyday speech.
1.7 Remarks
The thesi s stated above about the abstract character of col l ecti ons is
accepted wi del y but not uni versal l y. One very ardent proponent of the
opposi te thesi s was LeSni ewski who mai ntai ned that col l ecti ons are
physi cal bodi es and that the rel ati on of an el ement to a col l ecti on has
properti es si mi l ar to that whi ch a part of a body bears to the whol e body.
See Luschei (1962) for a detai l ed di scussi on of LeSni ewski s theory; see
al so Goodman (1951).
6 Andrzej Mostowski
2. DEVELOPMENT OF THE CONCEPT OF A SET IN MODERN
MATHEMATICS
Hencef orth we shal l constantl y use the wor d set i nstead of col l ecti ons.
It became common to use this word inmathemati cs and no l i ngui sti c
awkwardness can resul t f r omi ts use. Moreover, vari ous words whi ch in
ordi nary l anguage are used to denote certai n col l ecti ons acqui red i n
mathemati cs very speci fi c meani ngs and someti mes do not denote
any sets at all. Thi s i s true e.g. for such common words as group
and cl ass.
2.1 The set-theoretic foundation of mathematics
Sets were used i n mathemati cs l ong ago as i s wi tnessed by the noti on of
geometri c l oci used si nce ti mes i mmemori al inthe teachi ng of geometry.
They di d not pl ay an essenti al rol e, however, unti l the ni neteenth century
when mathemati ci ans began to use t hemextensi vel y i nthei r wor k on the
f oundati ons of cal cul us.
Let us recapi tul ate bri efl y some sal i ent hi stori cal facts. The great
mathemati cal di scoveri es made i nthe seventeenth century were f ol l owed
duri ng the next century by a stormy devel opment of the newi deas and of
thei r appl i cati ons. The newmathemati cs created i n thi s way was very
di fferent f r omthe ol d not onl y because the content was changed but al so
because the new mathemati cs was i ncomparabl y l ess exact. The mat hema-
ti ci ans of the ei ghteenth century strove towards enri chi ng mathemati cs by
new i nteresti ng facts and fi ndi ng new appl i cati ons. It is well known that
the f ounders of cal cul us based thei r theori es on utterl y uncl ear noti ons.
When one reads some of the works of the best mathemati ci ans of the
ei ghteenth century one often wonders how i t was possi bl e that they were
abl e to fi nd correct resul ts, so uncl ear and unconvi nci ng (by our present
standards) were thei r met hods (g. Pol ya, 1954, 20-21). The wonderf ul
preci si on of the ol d Gr eek mathemati cs fell i nto obl i vi on. Duri ng the
ni neteenth century a reacti on set i n. We see many outstandi ng mat hemat i -
ci ans worki ng on re-f ormul ati ng and correcti ng the f oundati ons of
anal ysi s. The essenti al task was to f ormul ate exact defi ni ti ons of such
basi c concepts as functi on (or mappi ng), sequence of numbers, compl ex
number , real number and then toderi ve thei r properti es f r omaxi oms and
general l y accepted l aws of l ogi c.
Sets 7
We cannot present here the detai l s of the l ong and pai nful wor k done
by the l eadi ng mathemati ci ans of the ni neteenth century starti ng wi th
Cauchy and endi ng wi th Frege and Cantor. Thei r wor k resul ted i n what i s
usual l y cal l ed the set-theoreti c f oundati ons of mathemati cs: all concepts
necessary for a ri gourous exposi ti on of mathemati cs proved to be
defi nabl e by means of very f ewbasi c concepts provi ded that a rather
unrestri cted use of sets was al l owed in the defi ni ti ons. All t heorems
formerl y accepted on the basi s of i ntui ti on al one or establ i shed in an
i ncompl ete and often obj ecti onabl e way proved to be strictly deri vabl e
f roma smal l number of very natural axi oms whi ch express the f undamen-
tal properti es of the basi c concepts. The wor d set occurred i nmany of
these axi oms even though at thi s earl y stage no axi oms for sets exi sted. I f
we coul d ask a ni neteenth century mathemati ci an why i t was so, he woul d
probabl y answer that properti es of sets are establ i shed inl ogi c and are of
no concern to mathemati ci ans.
We shal l show a f ewexampl es of reducti on of mathemati cal noti ons to
mor e si mpl e ones. The i ntui ti ve noti on of a mappi ng, or functi on, f r oma
set A i nto a set B, whi ch may or may not be di fferent f r omA, i s that of a
fi xed rul e whi ch associ ates witheach el ement x of A one, and one onl y,
el ement y of B. Thus, for exampl e, the functi on defi ned by +d x of the
set of i ntegers i nto the set of real numbers associ ates wi th each i nteger x
i ts posi ti ve square root. The i mage of an el ement x of A under a mappi ngf
f romA i nto B i s the el ement y of B whi ch i s associ ated wi th x by the
mappi ng f. Thus, i nthe above exampl e, 2 is the i mage of 4 under +l/x.
However , the noti ons of fi xed rul e and associ ate are themsel ves vague.
A preci se defi ni ti on of a mappi ng can be gi ven int er ms of sets, usi ng the
concept of an ordered pai r, that i s, of a set constructed f romt wo el ements
x and y such that the ordered pai r (x, y) is the same as the ordered pai r
(w, z) if and onl y if x = w and y = z. Thus a mappi ng f of A i nto B
i s defi ned as a set of ordered pai rs (x, y), wi th x inA and y i n B, such
that y i s the i mage of x under J: Thus i nstead of mappi ngs (functi ons)
we can speak of certai n sets of ordered pai rs. Ordered pai rs can al so be
defi ned by means of sets, but we shal l not di scuss thi s reducti on here.
Si mi l arl y, a sequence a(O), a(l), . . . of real number s i s defi ned as the
functi on whi ch maps the i ntegers O, 1 ... i nto real s and correl ates wi th
the i nteger y1 the val ue a(n). Thus agai n a sequence can be repl aced by
a set of pai rs, Compl ex number s a + bi can be defi ned as ordered pai rs
(a, b) of t wo real numbers.
8 Andrzej Mostowski
Most difficult proved the anal ysi s of real numbers. Thi s was, of course,
to be expected because, as i s well known to everybody even moderatel y
acquai nted withthe hi story of mathemati cs, these number s had al ready
caused a great deal of troubl e i npre- Eucl i dean ti mes. I n moder n mat he-
mati cs one has t wo ways of deal i ng withthese numbers. One way i s to
accept the noti on of a real number as pri mi ti ve and characteri ze i t by
means of sui tabl e axi oms; the other is to defi ne real number s by means of
rati onal number s (i.e. fractions). The fi rst met hod is cal l ed axi omati c, the
second geneti c. I n both methods the concept of a set pl ays an essenti al
rol e. Thus if we use the geneti c met hod, we defi ne a real number a as an
ordered pai r consi sti ng of t wo sets of rati onal numbers. I ntui ti vel y
speaki ng the fi rst set consi sts of rati onal s whi ch are smal l er than a and
the second of those whi ch are greater than or equal to a. I f we use the
axi omati c met hod we adopt among others the axi omof conti nui ty
whi ch says that i f A and B are t wo sets of real number s such that each
el ement of A is smal l er than each el ement of By then there i s a real
number x whi ch separates these sets, i.e. has the property that no el ement
of A is greater than x and no el ement of B i s smal l er than x.
Rati onal number s are easi l y defi nabl e by means of i ntegers but an
anal ysi s of thi s l atter noti on agai n proved to be rather difficult. Usual l y
one admi ts thi s noti on as pri mi ti ve and characteri zes i t by axi oms. Agai n
sets are of par amount i mportance because they are used in the basi c
axi omof i nducti on. (The si mpl est axi omequi val ent to the axi omof
i nducti on states that every non- voi d set of i ntegers contai ns a l east
el ement. ) The geneti c met hod of i ntroduci ng i ntegers was attempted by
Frege (1884; 1893-1093) (cf. Secti on 3.3 bel ow). Axi oms for i ntegers wer e
fi rst f ormul ated by Peano (1889).
Before we l eave the subj ect di scussed inthis secti on we want to make
three observati ons:
A. The use of sets inl ayi ng down correct f oundati ons for anal ysi s is an
hi stori cal fact but does not seemto be a necessi ty. Other general noti ons
such as functi ons or rel ati ons coul d be equal l y well used i nstead of sets to
achi eve thi s purpose. It i s not i mpossi bl e that sti l l other noti ons can be
i nvented whi ch coul d repl ace the noti on of sets.
B. In the f ormul ati ons of the axi oms and i nthe descri pti ons of the
vari ous defi ni ti ons gi ven above we di d not speak of i ndi vi dual sets but of
thei r totalities. Thi s is wi tnessed by the fact that we use quanti fi ers for
every set inthe axi oms of conti nui ty and of i nducti on. Thus we behave as
Sets 9
if the total i ty of all sets (of real s or of i ntegers) wer e a wel l -defi ned
col l ecti on of obj ects.
C. If the geneti c met hod of constructi ng real number s is used, then sets
of whi ch we i mpl i ci tl y speak inmathemati cal theorems have much hi gher
types than the col l ecti ons about whi ch we are used to speak i neveryday
l anguage. E. g. a mappi ng of real s i nto real s becomes a set of pai rs each
member of whi ch i s an ordered pai r consi sti ng of t wo sets of rati onal s.
2.2 The role of sets in further development of mathematics
We sawabove that the noti on of an arbi trary subset of a gi ven set became
a l egi ti mate mathemati cal noti on (cf. remark B above). Very soon
mathemati ci ans began to use i t extensi vel y and f ormed wi th i ts hel p new
chapters of mathemati cs whi ch were desi gned partl y to deepen the study
of the ol d 'classical' domai ns of mathemati cs and partl y to deepen the
study of the newnoti ons per se.
We cannot, of course, present here the detai l s of the new theori es whi ch
wer e created inthi s way because thei r subj ect matter is obvi ousl y very
' techni cal ' , We may, however, tryto poi nt out certai n rel evant facts.
In ol der mathemati cs the mai n i nterest was focused on speci al functi ons
whi ch mapped ei ther the set of real number s i nto the set of real number s
or the set of compl ex number s i nto the set of compl ex number s and
whi ch had some i nteresti ng properti es or wer e useful in mechani cs or
other parts of physi cs. The i mmedi ate effect of al l owi ng arbi trary
subsets of a gi ven set was the ri se of i nterest i nqui te arbi trary functi ons
of real s i nto real s; each functi on i s as we menti oned above, nothi ng el se
but a set of ordered pai rs. I n this way was created the newtheory of real
functi ons. It turned out that properti es whi ch i nol d theory wer e deemed
essenti al for the noti on of functi on (likeconti nui ty and deri vabi l i ty) were
onl y excepti onal l y sati sfi edby arbi trary functi ons: most real functi ons are
di sconti nuous and most conti nuous functi ons cannot be di fferenti ated.
Thus new met hods had to be created inorder to study such functi ons.
The theory of arbi trary sets was agai n the basi c tool inthi s theory.
The moder n theori es of measure and i ntegrati on were a natural
out come of thi s devel opment. These theori es proved much superi or to the
ol der theori es of measure and i ntegral created i n the ni neteenth century.
Many cl assi cal parts of anal ysi s coul d devel op onl y because the ol d theori es
of i ntegral wererepl aced by the newone. Today not onl y mathemati ci ans
10 Andrzej Most owski
but even physi ci sts are taught abstract general measure theory.
Si mi l ar devel opment t ook pl ace in geometry where one started to
di scuss properti es of very general sets of poi nts i nstead of anal ysi ng as
before onl y very regul ar curves and surfaces.
The stri ve towards general i ty di d not stop here. Havi ng been so
successful inthe study of arbi trary mappi ngs of real s i nto real s mat hema-
ti ci ans began to consi der functi onal s, i.e. functi ons whose arguments wer e
themsel ves functi ons or sets. An i ntegral or a measure are the si mpl est
such functi onal s. The systemati c study of functi onal s resul ted in the
creati on of a new mathemati cal theory cal l ed functi onal anal ysi s. Thus we
see that the step whi ch has l ed f romthe study of certai n speci fi c real
functi ons to the general theory of arbi trary such functi ons has been
repeated but on a hi gher l evel . Usi ng the noti on of types we can say that
the functi onal anal ysi s di scusses obj ects whi ch inthe hi erarchy of types
lieone step above the obj ects di scussed inthe cl assi cal anal ysi s.
Processes si mi l ar to that outl i ned above t ook pl ace not onl y in the
theory of functi ons but al so in geometry and especi al l y in topol ogy.
Everywhere the i nterest shi fted, from the study of si mpl e obj ects to the
study of obj ects of hi gher types. I f we l ook at thi s devel opment f r omthe
poi nt of viewwhi ch i nterests us here, we can express as fol l ows the general
pri nci pl e whi ch underl i es these processes: whenever a set i s acknowl edged
as a l egi ti mate obj ect of mathemati cs, so i s the set of all i ts subsets.
We shal l now descri be the ul ti mate step inthe stri ve towards general i ty
whi ch i s so characteri sti c of moder n mathemati cs.
Moder n theori es l i ke functi onal anal ysi s, topol ogy and some others
wer e concerned inthei r earl y stages al most excl usi vel y with parti cul ar
exampl es. Thus e.g. intopol ogy very general sets wer e di scussed but the
underl yi ng space inwhi ch the sets under di scussi on werecontai ned was
very speci al . Si mi l arl y functi onal anal ysi s deal t with functi onal s whi ch
were qui te abstract and general but whi ch operated i n some preci sel y
defi ned speci al sets, e.g. the set of real functi ons or the set of all real
sequences etc. The ul ti mate step consi sts of repl aci ng these underl yi ng
sets by arbi trary ones. Thus inthe moder n theory one starts withdefi ni ng
axi omati cal l y certai n spaces , i.e. sets i nwhi ch some addi ti onal structures
may be gi ven. The obj ect of the study are subsets of the space or mappi ngs
of one space i nto another. Thi s abstract approach is very economi cal
because i t usual l y al l ows us to obtai n many of the previ ousl y devel oped
theori es as parti cul ar i nstances of the general theory.
Sets 11
That such an abstract is possi bl e i s, of course, due to the devel opment
of the theory of abstract sets. We shal l outl i ne thi s devel opment inthe
next secti on but before doi ng thi s we shal l add some mor e remarks about
the i mportance of the abstract set-theoreti c approach inappl i cati ons of
mathemati cs.
2.3 Remarks about applicability of the abstract theories
These remarks must of necessi ty be very bri ef because the author does not
f eel compet ent enough to di scuss the probl emat l ength. Let us therefore
onl y ment i on that the abstract functi onal anal ysi s fi nds appl i cati ons in
quant umtheory where i t was di scovered that physi cal magni tudes such as
energy, moment um, etc. can best be descri bed by operators (functi onal s)
acti ng inan abstract space.
On a l ogi cal l y mor e el ementary l evel the study of functi ons defi ned on
abstract sets has made i t possi bl e to create, for exampl e, the theory of
games whose practi cal i mportance is today general l y recogni zed. I f one
reads the descri pti on of games e.g. invon Neumann and Morgenstern
(1947), one sees that no mathemati cal descri pti on of thi s concept woul d be
possi bl e wi thout abstract mathemati cs wi thi ts emphasi s on the study of
arbi trary sets and functi ons. The same can be sai d of the moder n theori es
of probabi l i ty withthei r many appl i cati ons as well as of the comput er
sci ence.
3. THE DEVELOPMENT OF ABSTRACT SET THEORY DURING THE
NINETEENTH CENTURY
Abstract set theory was created by Cantor, and thi s secti on is al most
enti rel y devoted to hi s work. However , we ment i on briefly the wor k of
Bol zano who had al ready di scovered many essenti al facts before Cant or,
and Frege whose studi es on the l ogi cal f oundati ons of ari thmeti c had a
prof ound al though i ndi rect i nfl uence on the theory of sets.
3.1 Bolzano (1781-1848)
Col l ecti ons about whi ch we speak i n everyday l anguage necessari l y
consi st of finitely many el ements. We sawin the previ ous secti on that
sets whi ch are real l y i mportant for mathemati cs are al ways i nfi ni te. The
12 Andrzej Mostowski
Czech phi l osopher B. Bol zano was the fi rst who systemati cal l y di scussed
i nfi ni te sets. ( Some remarks about thi s subj ect can al so be f ound in
earl i er wri ti ngs starti ng f r oa Anti qui ty: see Becker, 1964, 272.) Bol zano
di scovered that the properti es of i nfi ni te sets are very di fferent f rom
fi ni te sets, and he was so puzzl ed by thi s di fference that he named the
book inwhi ch he descri bed his di scoveri es Paradoxes of hzjinity (1851).
Of course, the name is very mi sl eadi ng because there i s nothi ng paradoxi -
cal in the theory of i nfi ni te sets; they j ust happen to behave di fferentl y
than the finite sets and upon refl ecti on one must admi t that i t woul d be
strange if i t wer e not so.
Let us gi ve an exampl e of a paradox di scovered by Bol zano. Two sets
A, B are cal l edequi pol l ent if there i s a one- to- one functi on F whi ch maps
A onto the whol e of B. The expressi on one- to- one means that t wo
di fferent el ements of A are al ways mapped onto di fferent el ements of B.
Thus e.g. i f A consi sts of all l etters whi ch occur inthe wor d spi ri t (there
are j ust 5 such l etters because i occurs twice) and B consi sts of all l etters
occurri ng inthe wor d Catul l a then we obtai n a one- to- one mappi ng of
A onto B if we correl ate C wi th s, a withp, t wi th i, u wi th r and
1 witht. I n other words, the functi on whi ch establ i shes equi pol l ence of
the t wo sets A and Bis the set of ordered pai rs (s,C), (pya), ( i $ ) , (r,u), (t,l).
For fi ni te sets equi pol I ence means si mpl y that they have the same
number of el ements, f r omwhi ch i t fol l ows that no finiteset i s equi pol l ent
with any of i ts proper subsets ( proper means that the subset is not
i denti cal withthe whol e set.)
For an i nfi ni te set A, one can easi l y exhi bi t a functi on whi ch establ i shes
the equi pol l ence of A wi th one of i ts proper subsets. Thus if A i s the set
of posi ti ve i ntegers 1 , 2, 3, . . , then the mappi ng J defi ned by the f ormul a
f(x) = 2x establ i shes the equi pol l ence of A and the set of all even
i ntegers.
3.2 Cantor (1845-1918)
Cantor was the true f ounder of the abstract theory of sets. Very earl y he
redi scovered most of Bol zano s paradoxes and what is mor e i mportant
showed that the abstract noti ons of set theory can be used to obtai n
amazi ngl y si mpl e sol uti ons of some mathemati cal probl ems (cJ hi s well-
known proof that there exist transcendental numbers).
The basi c noti on around whi ch centered the whol e theory i s that of the
Sets 13
equi pol l ence of t wo sets. I nstead of sayi ng that A and B are equi pol l ent
Cantor used to say that A and B have the same cardi nal i ty. I f A is equi pol -
l ent to a part of B but not conversel y, then we say that the cardi nal i ty
of A is smal l er than that of B.
The fi rst and most i mportant di scovery of Cantor was that there exi st
infinite sets of di fferent cardi nal i ti es; e.g. the cardi nal i ty of i ntegers is
smal l er than that of real s. Mor e general l y he showed that the cardi nal i ty
of any set A is smal l er than the cardi nal i ty of the set of all subsets of A.
Deeper resul ts concerni ng cardi nal i ti es wer e establ i shed by Cantor by
means of a newand most ori gi nal noti on - that of a wel l -ordered set. Let
us briefly expl ai n it.
A set as such has no parti cul ar structure. Of each obj ect one can say
that i t ei ther bel ongs to the set or not but f r oma mer e defi ni ti on of a set
one cannot drawany i nferences about how i ts el ements are rel ated to each
other. A set together wi th certai n rel ati ons defi ned for the el ements of
thi s set is cal l ed a structure. Thus e. g. the set of i ntegers together wi th
the l ess-than rel ati on is an exampl e of a structure as i s the same set
together withthe rel ati on x i s di vi si bl e by y . Of course these structures
are di fferent because the rel ati ons are di fferent.
A rel ati on R is cal l ed connected ina set A if for any t wo (di fferent)
el ements a, b of A ei ther a bears R to b or b bears R to a. ( Thus the
l ess than rel ati on is connected inthe set of i ntegers but not the rel ati on
of divisibility.) A rel ati on R is well f ounded in A if there exi sts no
infinite sequence a, by c, . . . of el ements of A such that b bears R to a, c
bears R to by etc. A structure consi sti ng of a set A and a rel ati on R
whi ch is connected and well f ounded i nA i s sai d tobe well ordered by R.
I n cases when there i s no doubt what is the orderi ng rel ati on R we si mpl y
speak of well ordered sets. As an exampl e of a wel l -ordered set we may
quote the set of i ntegers withthe l ess than rel ati on. But the same set wi th
the greater than rel ati on i s not well ordered.
The noti on of wel l -orderi ng whi ch may seemstrange at fi rst is as Cantor
has shown very i mportant i ndeed. The wel l -orderi ng of a set repl aces in
the theory of i nfi ni te sets the el ementary process of counti ng as we know
i t f romthe everyday practi ce: if we know a wel l -orderi ng of A, then
we can say whi ch el ement of A is the fi rst (inthe gi ven wel l -orderi ng),
whi ch i s the second, whi ch i s the fi rst whi ch fol l ows all el ements of a
gi ven subset of A, etc.
Usi ng the theory of wel l -orderi ngs Cant or was abl e to show that any
14 Andrzej Mostowski
t wo sets whi ch can be well ordered are comparabl e wi th respect to thei r
cardi nal i ti es. I ndeed, the cardi nal i ti es of any fami l y of wel l -orderabl e sets
f or mt hemsel ves a set wel l ordered by the l ess-than rel ati on between
cardi nal i ti es. I n parti cul ar, i t fol l ows that for each wel l -ordered set A, there
i s another such set whose cardi nal i ty i mmedi atel y fol l ows the cardi nal i ty
of A.
Other resul ts obtai ned by Cantor withthe hel p of wel l -orderi ngs wer e
concerned with proofs and defi ni ti ons by means of the so-cal l ed trans-
finitei nducti on. We shal l not pursue thi s matter here.
The theory of wel l -orderi ngs has i mmedi atel y rai sed several probl ems.
We ment i on onl y a fewof t hem: can every set be well ordered? I s the
cardi nal i ty of real number s next to the cardi nal i ty of i ntegers? I s the
cardi nal i ty of the set of all subsets of any infiniteset A next to the cardi -
nal i ty of A? These probl ems are cal l edrespecti vel y: the probl emof well-
orderi ng, the cont i nuum probl em and the general i zed cont i nuum
probl em. Thi l e we know a great deal about the fi rst, we know next to
nothi ng about the second and thi rd.
We shal l gi ve no mor e detai l s about Cantor s wor k al though of course
the above remarks contai n onl y a smal l porti on of resul ts reached by hi m
duri ng the many years whi ch he devoted to set theory. ( Moder n exposi ti on
of Cantor s resul ts are contai ned inall textbooks of set theory, see e.g.
Fraenkel , 1956. The ori gi nal wri ti ngs of Cant or are easi l y accessi bl e in
Cant or, 1932.)
3.3 Frege (1848-1925)
Frege s contri buti ons to the theory of i nfi ni te sets were onl y i ndi rect. He
was mai nl y occupi ed by what we woul d cal l today the geneti c theory of
i ntegers (cf. Secti on 2.1). He wanted to f ound the theory of i ntegers on the
set-theoreti c noti ons al one but treated the theory of sets as part of l ogi c.
He expressed the l aws of l ogi c on whi ch he based his constructi ons
wi th an unusual preci si on and cl ari ty whi ch al l owed thei r thorough
i nspecti on and di scussi on. I n thi s way a fi rst seri ous anal ysi s of the basi c
assumpti ons of what essenti al l y was set theory was made possi bl e. The
resul t of this anal ysi s was devastati ng: Frege s theory proved to be
i nconsi stent and he had to abandon his grandi ose plan.
The basi c i dea of Frege was very si mpl e. He defi ned si mi l arl y as Cantor
and Bol zano t he rel ati on of equi pol l ence, noti ced that i t i s an equi val ence
Sets 15
rel ati on and appl i ed to i t the pri nci pl e of abstracti on whi ch he hi msel f
di scovered (cf. Secti on 1.1). The equi val ence cl asses can be cal l ed cardi nal -
ti es; those whi ch consi sts of finitesets are accordi ng to Frege the usual
i ntegers.
Frege went beyond Cant or i nsofar as he expl i ci tl y used the set of all
sets inhi s constructi on. The rel ati on of equi pol l ence i s defi ned for all sets
and therefore the pri nci pl e of abstracti on was appl i ed by Frege to the set
of all sets. Whi l e Cantor woul d certai nl y not oppose such a set, yet he
never used i t expl i ci tl y. Hi s excel l ent mathemati cal i ntui ti on must have
warned hi mthat set is somewhat dubi ous and moreover of no i mportance
for mathemati cs. But Frege who was a great l ogi ci an had l ess i nterest in
pure mathemati cs and therefore fell easily i nto the trap. The i nconsi sten-
cies di scovered inhis axi oms brought about a l ong crisis inthe f oundati on
of set theory.
4. ANTINOMIES OF SET THEORY AND A REVISION OF ITS
FOUNDATIONS
4.1 Examples of antinomies
The weakest si de of the theory created by Cant or were i ts f oundati ons.
Cantor expl ai ned sets as col l ecti ons of obj ects whi ch together f or ma
new enti ty. I n Cantor s own words: Unter ei ner Menge verstehen wir j ede
Zusammenf assung M von besti mmten wohl unterschi edenen Obj ekten m
unserer Anschauung oder unseres Denkens (wel che di e El emente von M
genannt werden) zu ei nemGanzen (Mathematische Annalen, 1895).*
If this l oose expl anati on is accepted, then evi dentl y the set of all sets
* For the benefi t of readers who l i ke the compl i cated ni neteenth century Ger man
we quote sti l l Cantor s defi ni ti on dati ng back to 1879 ( ober unendl i che Li neare
Punktmanni gfal ti gkei ten, 0 3, Mathematische Annalen, 1879): Ei ne Manni gf al ti gkei t
(einI nbegri ff, ei ne Menge) von El ement en, di e i rgendwel cher Begri ffssphre angehren,
nenne i ch wohl defi ni ert, wenn auf Gr und i hrer Defi ni ti on und i nfol ge des l ogi schen
Pri nzi ps vom ausgeschl ossenen dri tten es als i ntern best i mmt angesehen werden muss,
sowohl ob i rgendei n dersel ben Begri ffssphre angehri ges Obj ekt zu der gedacht en
Manni gfal ti gkei t als El ement gehrt oder ni cht, wi e auch ob zwei zur Menge gehri ge
Obj ekte, trotz f ormal er Unterschi ede i n der Art des Gegebensei ns ei nander gl ei ch si nd
oder ni cht.
16 Andrzej Most owski
exi sts. But then we i mmedi atel y obtai n a contradi cti on. I f we denote the
set of all sets by A then accordi ng to Cantor s t heoremmenti oned in
Secti on 3. 2 the set B of all subsets of A woul d have the cardi nal i ty l arger
than the cardi nal i ty of A, whi ch i s i mpossi bl e because Bis a subset of A.
A mor e el ementary and better known anti nomy was i nvented by Russel l
who consi dered sets whi ch are not thei r own el ements and showed that
if X is the set of all such sets then the statements X is an el ement of X
and Xis not an el ement of X are equi val ent to each other.
Russel l s ant i nomy was publ i shed i n the appendi x to Frege s book
(1893-1903) (cf. al so Hei j enoort, 1967, 127 and 126). The fi rst anti nomy
ever to be publ i shed was that of Bural i -Forti (cf. Hei j enoort, 1967, 104).
It deal t withsl i ghtl y l ess el ementary concepts and we shal l not reproduce
i t here.
Cantor s fi rst reacti on tothe anti nomi es was that sets shoul d be di vi ded
i n t wo cl asses : the consi stent and i nconsi stent ones. Wher eas the f ormer
can be thought of as an obj ect , thi s i s not true of the latter (cf. Cantor,
1932, 443, al so Hei j enoort, 1967, 113). Thus presumabl y the usual l aws
of set theory are val i d onl y for consi stent sets.
The fol l owi ng anti nomy di scovered by Ri chard (cf. Hei j enoort, 1967,
142) shows that i nconsi stenci es may appear even if one restri cts onesel f to
sets of i ntegers. We present thi s ant i nomy ina sl i ghtl y modi fi ed versi on.
Let us consi der i nfi ni te sets of i ntegers such that there are infinitely many
i ntegers whi ch do not bel ong to the set. I f such a set can be defi ned by
means of an expressi on consi sti ng of finitely many Engl i sh words, then
we cal l i t a Ri chardi an set.
We observe now that all Engl i sh expressi ons consi sti ng of finitelymany
words can be arranged as they are i na di cti onary and one obtai ns thus an
infinite sequence contai ni ng all such expressi ons. We r emove f romthis
sequence expressi ons whi ch do not defi ne infinite sets of i ntegers or
defi ne i nfi ni te sets of i ntegers whi ch contai n all but fi ni tel y many i ntegers
so that we are l eft witha subsequence of expressi ons each of whi ch defi nes
a set contai ni ng infinitely i ntegers and such that i nfi ni tel y many i ntegers
l i eoutsi de. Let the terms of this sequence be D(O), D( l), . . . Now we sel ect
a pai r (ao, bo) of i ntegers such that a0 i s the l east i nteger in the set
defi ned by D(0) and bo is the l east i nteger whi ch does not bel ong to thi s
set. Let (al , bl) be a pai r such that al i s the l east el ement of the set
defi ned by D(1) whi ch i s di fferent f r oma0 and f r ombo; si mi l arl y bl i s
the l east i nteger not inthe set defi ned by D(1) whi ch i s di fferent f roma0
Sets 17
and f rombo. The next pai r (a2, bz) consi sts of the l east el ement a2 of the
set defi ned by D(2) whi ch is di fferent f r omao, bo, al, bl and of the l east
i nteger bz whi ch i s not an el ement of the set defi ned by D(2) and i s
di fferent f romao, bo, al, 61. Conti nui ng in thi s way we obtai n an
i nfi ni te number of pai rs and we see easi l y that the set R consi sti ng of the
number s bo, bl, ... is i nfi ni te, and that i nfi ni tel y many i ntegers do not
bel ong to i t (e. g. none of the numbers aj bel ongs to this set). The set R
has been defi ned above by means of an Engl i sh expressi on because the
whol e of the present paragraph can be consi dered as a defi ni ti on of R.
Hence R i s a Ri chardi an set and thus is defi ned by one of the expressi ons
occurri ng inthe sequence D(O), D(1), . . ., say, by the expressi on D(r). But
thi s is a contradi cti on because br is not an el ement of the set defi ned by
Evi dentl y the onl y possi bl e expl anati on of these anti nomi es is that the
sets A, X di scussed above do not exi st. It is not cl ear however how to
pi npoi nt the pl ace where we vi ol ated the rul es for f ormi ng sets.
The di sturbi ng i dea whi ch must have crept i nto the minds of some
mathemati ci ans after they l earnt about the anti nomi es is that perhaps
sets used in mathemati cal theori es and defi ned by means of purel y
mathemati cal concepts al so l ead tocontradi cti ons. They have not yet been
di scovered, but there is no assurance that they will not be di scovered in
future.
It must be sai d that most mathemati ci ans di d not share these doubts.
They devel oped set-theoreti cal l y ori entated mathemati cal theori es whi l e
remai ni ng unconcerned with what happened in the f oundati ons of set
theory itself. They fel t confi dent that if we fol l owcertai n rul es of con-
structi ng sets, no contradi cti ons will appear. The rul es accepted by the
maj ori ty if mathemati ci ans wer e eventual l y codi fi ed as axi oms of set
theory. These axi oms will be presented bel ow. Yet there remai ned a smal l
but i mportant group of mathemati ci ans who were not convi nced that the
mer e formul ati on of the axi oms represents the final sol uti on of the prob-
l emposed by the anti nomi es. We ment i on some of these di ssenti ng
vi ews inSecti on 6.2.
D(r)-
4.2 Zermelo-Fraenkel axioms of set theory
The fi rst axi omati c systemfor set theory was f ormul ated by E. Zermel o
in1908 (cf. Hei j enoort, 1967, 199); essenti al amendment s wer e i ntroduced
18 Andrzej Mostowski
by Fraenkel (1922) and Skol em(see Hei j enoort, 1967, 290) inthe earl y
twenti es. Vari ous modi fi cati ons of the systemwer e al so proposed and we
shal l ment i on t heml ater but fi rst we want to descri be the Zermel o-
Fraenkel syst em(abbrevi ated Z- F) itself, whi ch i s by far the most
wi del y- used axi omati c systemof set theory.
The pri mi ti ve concepts of Z- F are sets and the rel ati on of bei ng an
el ement of a set. These concepts are not defi ned but characteri zed by
axi oms. The fi rst axi om(axi omof extensi onal i ty) states that sets are
determi ned by their el ements or more preci sel y: if A, B are sets and each
el ement of A i s an el ement of B and conversel y, then A = B. The next
three axi oms descri be operati ons whi ch yi el d newsets f romsets al ready
gi ven. These are: the axi omof pai rs, whi ch says that for every t wo sets
there exi sts a newset whose uni que el ements are the gi ven sets; the axi om
of power sets says that for each set A there exi sts a set whose el ements are
all the subsets of A; the axi omof uni ons says that for each set A, there
exi sts a set whi ch contai ns as a subset each set bel ongi ng toA.
All t hese axi oms, however, do not yet guarantee that sets exi st in the
fi rst pl ace and even if they exi st, whether there are infinitesets. In order
to settl e these probl ems we assume the exi stence of at l east one i nl i ni te set.
The f ormul ati on of this axi omof i nfi ni ty by means of the pri mi ti ve
concepts i s rather i nvol ved and we shal l omi t i t.
There remai n sti l l three axi oms to be enumerated. They are mor e
compl i cated than the ones we menti oned thus far and requi re a mor e
detai l ed descri pti on.
The fi rst of t hemis cal l ed axi omof choi ce and says that for each set
of sets A no t wo di fferent el ements of whi ch have common el ements,
there exi sts a set C (the choi ce set) whi ch contai ns exactl y one el ement
f r omeach non- voi d el ement of A.
The axi omof choi ce has a di sti nctl y di fferent character t han the
previ ous ones. Unl i ke the axi oms of constructi on gi ven above, i t does not
descri be an operati on whi ch yi el ds a uni quel y determi ned resul t when
appl i ed to a set; i t onl y states the exi stence of a choi ce set. I n general a
set A sati sfyi ng the assumpti ons of the axi omwill have many choi ce sets.
For thi s reason the axi omwas met witha degree of mi strust especi al l y by
those mathemati ci ans who wanted to consi der onl y sets defi ned by
properti es of their el ements. We now know that; in general , no such
defi ni ti on i s possi bl e: for i nstance i f A is the set of all unordered pai rs
(f, -f} where f is a real functi on defi ned on the set of all Teal s, then no
Sets 19
choi ce set for A can be defi ned inthe f orm: the set of all those x whi ch
satisfy a condi ti on F. Here F may be any condi ti on expressi bl e in the
(fi rst-order) l anguage of set theory and not contai ni ng parameters. (Thi s
exampl e has been communi cat ed to me by Dr W. Marek. )
In spi te of thi s however, much can be sai d for admi tti ng the axi omof
choi ce. I n several speci al cases i t can be proved, e.g. when the set A is
finite. Many i mportant resul ts cannot be proved wi thout i ts hel p. For
i nstance, we need this axi omin the proof that each set can be wel l
ordered (cf. Secti on 3.2). The choi ce axi omi s equi val ent to many
theorems whi ch seemto be i ntui ti vel y obvi ous. Thus al though some
consequences of the axi omare very strange, i t i s al most general l y
accepted. Most mathemati ci ans liketo vi sual i ze a set as a bunch of obj ects
and mai ntai n the exi stence of the choi ce set C by argui ng that if several
bunches of obj ects are gi ven, we can sel ect an el ement f r omeach bunch
and col l ect these el ements together to f orma set.
Before we f ormul ate the l ast axi om(of repl acement) we di scuss a
weaker axi omof comprehensi on (al so cal l ed axi omof subsets). Thi s was
the axi omori gi nal l y accepted by Zermel o. Onl y l ater Skol emand
Fraenkel noti ced that a stronger axi omwas needed.
Zermel o s f ormul ati on of the comprehensi on axi omwas the fol l owi ng:
whenever the proposi ti onal functi on E(x) is defi ni te for all el ements of a
set M, M possesses a subset ME contai ni ng as el ements preci sel y those
el ements x of M for whi ch E(x) is true (cf. Hei j enoort, 1967, 262).
The obvi ous weakness inthe f ormul ati on of thi s axi omi s that i t uses
the concept of a defi ni te proposi ti onal functi on whi ch must be made
preci se if one wants to avoi d the recurrence of anti nomi es of the Ri chard
type. Zermel o s own expl anati on of thi s concept was not very clear.
We know at present t wo ways whi ch al l owus to express preci sel y the
axi om of comprehensi on. One of t hemproposed by Skol em(cf:
Hei j enoort, 1967, 290) will be presented bel ow. The second met hod will
be deal t wi th inthe next secti on.
Skol em s i dea was to speci fy the l anguage of set theory and i denti fy
defi ni te proposi ti onal functi ons wi th those whi ch are expressi bl e inthi s
l anguage. Thus i nstead of a si ngl e axi omof comprehensi on we obtai n
a scheme of axi oms, each parti cul ar case of the scheme correspondi ng
to a f ormul a of the l anguage.
The f ormul ae of the l anguage are defi ned by i nducti on: we sel ect a cl ass
of initial (or atomi c ) f ormul ae and construct the compound ones by
20 Andrzej Mostowski
appl yi ng arbi trari l y many times the operati ons of j oi ni ng t wo f ormul ae
together by a l ogi cal connecti ve and of prefi xi ng a f ormul a by a quanti fi er
rangi ng over obj ects whi ch are ei ther sets or el ements of sets. For the
purpose of the general set theory i t is suffi ci ent to admi t as atomi c
f ormul ae the fol l owi ng ones: x i s a set , x i s an el ement of y , where
l etters x and y may be repl aced by any other l etters.
The axi omscheme of comprehensi on takes now the f or m:
For every set A there is a set B such that an arbi trary obj ect x is an
el ement of B if and onl y if x i s an el ement of A and F(x).
We obtai n parti cul ar cases of the axi omof comprehensi on repl aci ng
i nthi s scheme F(x) by any f ormul a inwhi ch the vari abl e By does not
occur.
We now come to the last axi omof the Zermel o- Fraenkel system, the
so-cal l ed axi omof repl acement. Speaki ng l oosel y, thi s axi omstates that
for an arbi trary operati on defi ned on el ements of a set, the i mages of
all the el ement s of thi s set f ormagai n a set.
Agai n the concept of an operati on requi res cl ari fi cati on and we
achi eve thi s by al l owi ng onl y operati ons defi ned inthe l anguage. Thus
we assume the fol l owi ng scheme:
I f A i s a set and for every el ement x of A there is exactl y one obj ect y
such that F(x,y), then there i s a set B such that an arbi trary obj ect y i s an
el ement of B if and onl y if F(x,y) for some el ement x of A .
We obtai n parti cul ar cases of the axi omof repl acement repl aci ng i n
thi s scheme F(x,y) by any f ormul a inwhi ch the vari abl e 23 does not
occur.
I n order to make the content of the scheme mor e accessi bl e to the
reader we noti ce that the assumpti on states the exi stence of an operati on
defi ned by the f ormul a F(x,y) for all el ements of A: to each x in A
corresponds exactl y one y such that F(x,y). The concl usi on then says that
there exi sts a set B whose el ements are all the i mages of the el ements of
A under thi s operati on.
The descri pti on of the axi oms of Z F i s now compl ete and the axi oms
of comprehensi on and of repl acement are preci se. However , they are not
si ngl e axi oms but i nfi ni te col l ecti ons of axi oms determi ned by schemes.
We noti ce still that each parti cul ar case of the axi om(scheme) of
comprehensi on can be deri ved f r oma sui tabl e parti cul ar case of the
axi om(scheme) of repl acement so that the f ormer scheme can be di spen-
sed wi th. We preferred to keep thi s scheme, however, because i t i s much
Sets 21
si mpl er than the axi omscheme of repl acement and very often used i n
set-theoreti c proofs.
Skol em s i dea of a correct f ormul ati on of Z F axi oms is very neat
because i t remi nds us that sets whi ch we mostl y use are defi ned by means
of properti es of thei r el ements, and properti es are expressed ina l anguage.
The drawback of hi s met hod is that we do not obtai n a fi ni te number of
axi oms but an i nfi ni te col l ecti on of axi oms defi ned by t wo schemes.
Of all the met hods for f oundi ng abstract set theory on a secure basi s
the systemZ- F is most wi del y used and seems to approxi mate best the
i ntui ti ve mathemati cal i dea of sets. A group of French mathemati ci ans
who publ i sh under a j oi nt pseudonym Bourbaki were ri ght incal l i ng
a sl i ght vari ant of Z F a systemof f oundati ons of mathemati cs for
worki ng mathemati ci ans (cf. Bourbaki , 1949).
Mor e references to sources quoted above can be f ound in Fraenkel
(1956); this book contai ns a detai l ed exposi ti on of topi cs deal t wi th inthi s
and in the subsequent secti ons.
4. 3 Von Neumann-Bernuys-G~e~ axioms of set theory
In thi s secti on, we descri be briefly another met hod of maki ng preci se the
axi oms of comprehensi on and of repl acement.
The basi c i dea was f ormul ated by von Neumann (cf. Hei j enoort,
1967, 393) who i ntroduced the concept of an arbi trary operati on as a new
pri mi ti ve concept of set theory and gave axi oms whi ch characteri ze it.
His i dea was taken up by Bernays and Gdel (cf. 1939); i nstead of
operati ons these authors used a concept of a proposi ti onal functi on. I t is
customary to use the wor d cl ass i nstead of proposi ti onal functi on . Thus
the voli Neumann- Ber nays- Gdel set theory (abbrevi ated B- G) has now
t hree pri mi ti ve concepts: set, cl ass and el ementhood . All obj ects deal t
wi th i n the theory are di vi ded i nto el ement s, i.e. obj ects whi ch bel ong
to at l east one cl ass, and non- el ements (al so cal l ed proper cl asses) whi ch
are not el ements of any cl ass. One assumes that each set is a cl ass: mor e
preci sel y a set is a cl ass whi ch i s an el ement; proper cl asses are therefore
not sets. A cl assfi s an operati on if i ts el ements are excl usi vel y ordered
pai rs and moreover i f i t contai ns no pai rs wi th the same fi rst member but
wi th di fferent second member s: if (xJ) and (x,y) are el ements of f ,
then y = y. The domai n of f i s defi ned as a cl ass whi ch consi sts of all
el ements x such that, for some el ement y, the ordered pai r @,y) i s an
22 Andrzej Mostowski
el ement of f. The range of f is defi ned si mi l arl y: i t i s the cl ass consi sti ng
of all el ements y such that, for some el ement x, the ordered pai r (x,y) i s
an el ement of f.
Wi th these defi ni ti ons we can express the axi oms of comprehensi on
and of repl acement very bri efl y: the axi omof comprehensi on says that
the i ntersecti on of a set and of a cl ass i s a set, whereas the axi omof
repl acement says that if the domai n of an operati on is a set, then so is
i ts range.
However , we need sti l l several axi oms whi ch al l owus to prove the
exi stence of vari ous cl asses. They are cal l ed axi oms of cl ass f ormati on.
We descri be t hemonl y bri efl y and do not f ormul ate all of t hem. Fi rst,
we assume the exi stence of t wo cl asses: V whose el ements are all sets and
E whose el ements are all pai rs (x,y) such that y is a set and x is an el ement
of y. These cl asses correspond tothe atomi c f ormul ae used i nthe met hod
of Skol em. In the remai ni ng axi oms we assumed that certai n operati ons
correspondi ng to l ogi cal connecti ves and quanti fi ers perf ormed on cl asses
agai n yi el d cl asses. We al so assume that t wo cl asses wi th the same
el ement s are equal . The reader is referred to Fraenkel (1956) for a
compl ete di scussi on of these axi oms.
In most moder n exposi ti ons one assumes that all el ements are sets, but
thi s assumpti on is not real l y necessary and is onl y made i n order to
si mpl i fy certai n arguments.
The axi omati c systemB-G has many advantages. Unl i ke Z F i t i s
axi omati zed by means of finitely axi oms. Moreover, the use of cl asses
al l ows us tosi mpl i fy our mode of speech; we can e. g. state and prove the
t heoremsayi ng that the cl ass of all sets i s not a set and have thus a rati onal
way of expressi ng a t heoremwhi ch l i es at the bot t omof vari ous anti no-
mi es. Proper cl asses woul d of course be counted by Cantor to i ncon-
si stent sets (cf. Secti on 4.1). Thus the systemE-G al l ows us to di scuss
some at l east of these sets.
In spi te of these advantages, we bel i eve that i t i s the systemZ F rather
than the B-G whi ch occupi es the central pl ace i n f oundati ons of set
theory. The new pri mi ti ve concepts (that of a cl ass or of an operati on)
pl ay inB-G or i nthe ori gi nal systemof von Neumann a subordi nate rol e.
The sol e purpose for whi ch these concepts wer e i ntroduced was to
si mpl i fy the f ormul ati on of the axi oms of repl acement and comprehensi on
of Z- F. Thi s i s wi tnessed by the rather artificial and compl i cated f or mof the
axi oms of cl ass f ormati on and was the reason why we omi tted t hemabove.
Sets 23
Let us fi nal l y note that the systems Z F and B-G are rel ated to each
other so that the consi stency of one of t hemi mpl i es the consi stency of the
other (see Fraenkel , 1956).
4.4 Type theory
Mor e or l ess si mul taneousl y with Zermel o, B. Russel l f ormul ated
another systemwhi ch al l owed hi mto reconstruct the essenti al parts of the
Cantori an set theory. Thi s system, cal l ed the si mpl e theory of types, is
i nteresti ng because of i ts cl ose connecti ons wi th the concepti on of sets
whi ch we admi t i neveryday l anguage. As the adj ecti ve si mpl e i ndi cates,
Russel l al so consi dered another theory of types (cal l ed the rami fi ed
theory of types) and osci l l ated for several years bet ween these t wo theori es
as is evi dent f r omthe i ntroducti on to the second edi ti on of hi s (and
Whi tehead s) Principia Mathematica (1910).
Accordi ng to the si mpl e theory of types all obj ects are di vi ded i nto
types. Obj ects whi ch are not sets have type O, sets of these obj ects have
type 1, sets consi sti ng of sets of type 1 have type 2, etc. Thi s hi erarchy of
sets i s in compl ete agreement wi th our i ntui ti ve concepts of sets: we
noti ced al ready in Secti on 1.6 that we al ways di sti ngui sh bet ween
col l ecti ons whose el ements are obj ects and col l ecti ons whose el ements are
themsel ves col l ecti ons.
The theory of types was expressed by Russel l in a very el aborate
l anguage inwhi ch vari abl es wer e di vi ded i nto types and coul d occur onl y
i n certai n combi nati ons. Thus anti nomi es wer e avoi ded because they
si mpl y coul d not be f ormul ated i n the l anguage of type theory. The
theory was based on axi oms the most i mportant of whi ch bore cl ose
rel ati on to the axi omof comprehensi on. Most of current mathemati c can
be expressed al ready in the theory of types, al though not wi thout a
certai n degree of awkwardness.
We want to di scuss rel ati ons between the type theory and Z-F. We
can easi l y show that the type theory i s i nterpretabl e i n Z F . In order to
achi eve this we sel ect an arbi trary set Ro of obj ects whi ch are not sets and
defi ne i nZ-F an infinitesequence Roy RI, R2, ... such that Rn+l i s the set
of all the subsets of R,. We can now defi ne an i nterpretati on of the type
theory inZF: obj ects of type O are i nterpreted as el ements of Ro, sets of
type 1 as el ements of RI, sets of type 2 as el ements of R2 whose el ements
have type 1, etc. Sets obtai ned inthis way are defi nabl e inZ-F in terms of
24 Andrzej Mostowski
80. All axi oms of the type theory become theorems of Z-F wi th the
possi bl e excepti on of the axi omof i nfi ni ty whi ch i s sati sfi edif and onl y if
the Ro i s i nfi ni te.
We see f r omthi s i nterpretati on that Z-F i s a much ri cher theory than
the theory of types. Whi l e each set consi dered in the type theory must
have a type, the theory Z-F does not excl ude sets consi sti ng of el ements
of vari ous types. For i nstance R3 contai ns sets some of whose el ements
have type O and some type 1. We say that such sets have mi xed types.
The sequence of the sets R, can be further extended: we denote the
uni on of all the sets R, by Rw and start agai n f ormi ng successi ve power
sets: Rw+l, Rw+2, ... where each t er mconsi sts of all the subsets of the
previ ous one. Thi s process can be extended further: for each well-
orderi ng a we obtai n a set Ru. Sets Roy RI, ... correspond to the case of
wcl l -orderi ngs of a finiteset, but if o! is a wel l -orderi ng of an infiniteset we
obtai n new sets whose types we call transfi ni te. The el ements of Ru have,
general l y speaki ng, mi xed types.
Assumi ng inZ F some addi ti onal axi oms whi ch we shal l not f ormul ate
here, one can prove that every set bel ongs to one of the sets Ru. Thus i n
thi s extended theory, one can prove that each set has a (in general mi xed
and transfi ni te) type. Thi s shows that the theory Z F whilevery di fferent
f r omthe theory of types is ineffect not so unrel ated to thi s latter theory
as mi ght be t hought at the begi nni ng. It rather represents a far-reachi ng
extensi on of the theory of types obtai ned by al l owi ng sets of mi xed types
and extendi ng the hi erarchy of types far beyond the finitetypes.
4.5 Other methods of avoiding antinomies
In thi s secti on we menti on briefly sti l l t wo theori es whi ch, l i ke 2-F,
B- G and the type theory, attempt to sal vage set theory f r omanti nomi es
whi l e preservi ng as many cl assi cal resul ts as possi bl e.
A. P. Mor se (1965) and J. L. Kel l ey (1955) extended the systemB- G by
admi tti ng that every f ormul a defi nes a cl ass consi sti ng of all sets whi ch
sati sfy thi s f ormul a. In the Bernays- Gdel systemthe exi stence of this
cl ass can onl y be proved for f ormul ae whose bound vari abl es are restri cted
to sets (cf. Secti on 4.3).
W. V. Qui ne (1969) has publ i shed several versi ons of a compl etel y
di fferent system(see esp. Chap. XIII). Hi s concepti ons devi ate most f rom
those of Z-F: e.g. one can prove inhi s theory that for every set A there
Sets 25
i s a set whose el ement s are all those sets whi ch do not bel ong to A (!). Yet
no anti nomy has appeared inhis syst embecause of restri cti ons i mposed
on f ormul ae whi ch can be used to construct a set. Al t hough they are
effi ci ent, these restri cti ons seemto l ack suffi ci ent i ntui ti ve j usti fi cati on,
and i t seems doubtf ul to the present author whether any part of Qui ne' s
theory will ever gai n popul ari ty among mathemati ci ans.
5. INCOMPLETENESS OF SET THEORY '
5.1 Theory and metatheory
Each mathemati cal theory is expressed ina l anguage whi ch must not be
conf used withthe l anguage inwhi ch we speak about the theory. I n order
to expl ai n why thi s di sti ncti on i s so i mportant we return to anti nomi es
and di scuss once agai n the anti nomy of Ri chard (see 4.1). The concept of
defi nabi l i ty used inthe defi ni ti on of a Ri chardi an set is cl earl y a met a-
theoreti c one and we saw inSecti on 4.1that the use of thi s concept inset
theory resul ts in a contradi cti on. Onl y the l i mi tati on i mposed on the
l anguage of Z-F (Secti on 4.2) or of the theory of types el i mi nates
contradi cti ons f romthese theori es.
Once a mathemati cal theory i s f ormul ated with a suffi ci ent degree of
preci si on we can ask (in the metatheory) vari ous general questi ons
concerni ng thi s theory. Thus we may ask: i s the theory consi stent? Is i t
consi stent rel ati ve to another better known theory? I s the theory com-
pl ete? Ar e i ts axi oms i ndependent?
5.2 Problem of consistency
Not very much can be sai d about thi s i mportant probl em. As is well
known Hi l bert reacted to the di scovery of anti nomi es by i ni ti ati ng
. extensi ve metamathemati cal studi es whose purpose was to show that he
whol e of mathemati cs (provi ded that i t be correctl y descri bed, e.g. as an
axi omati c system) is consi stent. However , Gdel has shown (1931), see
al so (Hei j enoort, 1967, 592) that every such proof must use means whi ch
go beyond the axi oms whose consi stency we want to prove. It fol l ows
thus f romGdel ' s resul t that no consi stency proof for set theory exi sts
whi ch woul d use excl usi vel y methods f ormal i zabl e inthis theory. Now set
26 Andrzej Mostowski
theory is such a strong systemthat practi cal l y all of the present day
mathemati cs can be i nterpreted ini t (cf. Secti on 3.2). These observati ons
expl ai n why no strictlymathemati cal consi stency proof for set theory has
been f ound. For the same reason there can be no proof of consi stency
rel ati ve to theori es whi ch can be i nterpreted i n set theory, l i ke e. g.
ari thmeti c of i ntegers or ari thmeti c of real nuni bers.
All mathemati ci ans bel i eve, however, that the systemZ-F is consi stent
and they argue, not wi thout good reason, that if there werecontradi cti ons
i nthe systemthey woul d certai nl y have al ready been f ound. After all, the
systemhas al ready exi sted for al most 60 years and was extensi vel y
studi ed by the best mathemati ci ans of the present century.
5. 3 Incompleteness
The general resul ts of Gdel (1931) showthat if 2-F i s consi stent, then i t
i s i ncompl ete and that the same i s true for other axi omati c systems
proposed for set theory. The di squi eti ng fact i s that inthe case of Z- F as
well as inother systems proposed as a possi bl e f oundati on for set theory
i t i s rather easy to f ormul ate probl ems whi ch cannot be deci ded on the
basi s of the axi oms. Roughl y speaki ng, all probl ems whi ch have not been
easi l y sol ved wi thi n a short timeafter bei ng f ormul ated turned out to be
undeci dabl e. We shal l descri be bel owsome undeci dabl e probl ems.
5.4 Large cardinals
The axi omof infinity secures the exi stence of at l east one inIiniteset; usi ng
the power set axi omand the axi omof uni ons we can obtai n f r omthi s
set other sets of surpri si ngl y hi gh cardi nal i ti es.
Now we ask about properti es of these cardi nal s. Let us call a cardi nal
accessi bl e if i t is the cardi nal of a set Ru and is ei ther the power set of a
precedi ng Rp or is representabl e as a uni on of a fami l y of sets A whose
cardi nal i ty i s smal l er than Ru and all whose el ements have the cardi nal i ty
smal l er than Ru. The i dea behi nd this defi ni ti on is that a set Rp withan
accessi bl e cardi nal i ty can be obtai ned f r omsets of smal l er cardi nal i ti es by
the use of t wo operati ons: f ormi ng uni ons and sets of subsets. The
questi on whether there exi st i nacessi bl e cardi nal s has proved to be
undeci dabl e inZ- F. Mor e exactl y if Z-F i s consi stent, then i t remai ns so
after adj uncti on of an axi omstati ng that all cardi nal s are accessi bl e. The
Sets 27
consi stency of an axi omstati ng that there are i naccessi bl e cardi nal s
cannot be proved but i t seems extremel y unl i kel y that this axi omwoul d be
i nconsi stent wi th Z-F.
I nvesti gati on of i naccessi bl e cardi nal s was started al ready in 1912 by
Mahl o (1912-1913). The present state of the theory of these cardi nal s is
due mostl y to Tarski (1963- 1964 and the extensi ve l i terature quoted
therein). We cannot di scuss thi s hi ghl y techni cal subj ect here and r emar k
onl y that inthe course of these studi es a great many axi oms wereproposed
whi ch were all cal l ed axi oms of i nfi ni ty because they postul ate the
exi stence of extremel y hi gh cardi nal i ti es. All these axi oms are i ndependent
of Z-F and probabl y consi stent wi th Z-F. They are i nterrel ated to each
other someti mes ina very compl i cated way and much wor k was done in
order to establ i sh these rel ati ons and to study the effects whi ch the
admi ssi on of some axi oms of i nfi ni ty woul d have on the theory of smal l er
cardi nal s. Set-theori sts do not know at present whether they shoul d
admi t or rej ect these axi oms.
5.5 The continuum hypothesis
Thi s hypothesi s states that the answer tothe cont i nuumprobl emwhi ch we
f ormul ated i n Secti on 3. 2 is afl i rmati ve. The general i zed cont i nuum
hypothesi s states l i kewi se that the answer to the general i zed cont i nuum
probl emi s affi rmati ve. Bot h these hypotheses bel ong to a l arge f ami l y of
sentences whi ch nei ther can be proved nor refuted in ZF. I n addi ti on,
i t has been shown that the strengtheni ng of Z-F by the axi oms of i nfi ni ty
securi ng the exi stence of i naccessi bl e cardi nal s does not change the
si tuati on. I n 1939, Gdel (see al so Secti on 6.2) showed that if Z-F i s
consi stent, then i t remai ns so if we add to i t the general i zed cont i nuum
hypothesi s. Gdel ' s di scovery represented the fi rst great breakthrough in
the meta- mathemati cal study of set theory. Anot her such breakthrough
is due to Cohen (1966) who proved i n1964 that cont i nuumhypothesi s i s
i ndependent of ZF. Heachi eved this remarkabl e resul t by showi ng that
if Z-F is consi stent, then there exi sts a model of Z F i nwhi ch there are
sets whose cardi nal i ti es (cal cul ated i nthe model ) are i ntermedi ate bet ween
the cardi nal i ty of i ntegers and that of real s, i.e. are greater than the f ormer
and smal l er than the l atter. What is sti l l mor e stri ki ng i s that Cohen
showed that there are many such model s. In some of t hemthere is j ust
one i ntermedi ate cardi nal i ty, inothers t wo or three, or f our, etc. , sti l l in
28 Andrzej Mostowski
others there are i nfi ni tel y many such cardi nal i ti es. Thus Cohen showed
that axi oms of Z-F gi ve us al most no i nf ormati on about the cardi nal i ty
of the set of real s. We know that thi s cardi nal i ty i s di fferent f romthat of
the i ntegers and we al so know that the set of real s i s not the uni on of a
sequence of sets wi th i ncreasi ng cardi nal i ti es. But no hypothesi s consi stent
wi th these t wo resul t contradi cts the axi oms (Sol ovay, 1965). Si mi l ar
phenomena are val i d for the general i zed cont i nuumhypothesi s. (See
Rosser (1969) and references made there to the wor k of Easton. )
These resul ts show that, contrary to earl i er expectati ons, Z F is an
utterl y i ncompl ete system. How i ts axi oms shoul d be strengthened in
order to deci de the cont i nuumhypothesi s i s agai n an open probl em,
5.6 Other undecidable problems; independence of axioms
There are many other probl ems of set theory whi ch proved to be
undeci dabl e inZ-F. They are very techni cal and i t woul d not be possi bl e
to present t hemhere. Met hods of provi ng undeci dabi l i ty can al so be
appl i ed to establ i sh the mutual i ndependence of vari ous axi oms. Thus
Cohen (see 1966) was abl e to sol ve compl etel y the vexi ng probl emof the
i ndependence of the axi omof choi ce, and many mor e refi ned probl ems
concerni ng the mut ual dependence of several consequences of this axi om
wer e sol ved subsequentl y by hi mand hi s fol l owers.
6. OTHER TRENDS IN FOUNDATIONS OF SET-THEORY
6.1 Philosophical importance of the independence results
Do the undeci dabi l i ty resul ts descri bed in the previ ous secti on prove
that there is anythi ng wrong wi th abstract set theory? The probl emi s an
ol d one and was di scussed l ong before the recent undeci dabi l i ty resul ts
were strictly establ i shed. Si nce the earl y twenti es, nobody has real l y
bel i eved that the cont i nuumhypothesi s can be sol ved on the basi s of
Z-F and there wer e di sti ngui shed mathemati ci ans likee.g. Lusi n (1927)
who f romthe start bel i eved that even very strange hypotheses concerni ng
the cardi nal i ti es of the cont i nuumare compati bl e wi th ZF.
The general resul ts of Gdel (1931) have shown that no axi omati c
theory contai ni ng ari thmeti c can be compl ete. Thus the mer e i ncompl ete-
Sets 29
ness of Z- F i s not an al armi ng sympt omby itself. What is di sturbi ng i s our
i gnorance of where to l ook for addi ti onal i nf ormati on whi ch woul d
permi t us to sol ve probl ems whi ch seemvery si mpl e and natural but
whi ch are neverthel ess l eft open by the axi oms of Z-F. We come here very
cl ose to f undamental probl ems of the phi l osophy of mathemati cs whose
basi c questi on i s: what i s mathemati cs about ? A formal i st woul d say that
i t i s about nothi ng; that i t i s j ust a game pl ayed witharbi trari l y sel ected
axi oms and rul es of proof . The i ncompl eteness of Z-F is thus of no
concern for a formal i st. Pl atoni sts on the contrary bel i eve inthe obj ecti ve
exi stence of mathemati cal obj ects. A set-theoreti cal Pl atoni st bel i eves
therefore that we shoul d conti nue to thi nk mor e about sets and experi ment
wi th t hemunti l we fi nal l y di scover newaxi oms whi ch, added to Z-F,
will permi t us to sol ve all outstandi ng probl ems (cf. Gdel , 1947).
What ever the fi nal out come of the fi ght between these t wo opposi ng
trends will be, i t is obvi ous that we shoul d concentrate on the study of
concepts whi ch seemperfectl y cl ear and perspi cuous to us. I n Cantor s
timethe concept of an arbi trary set seemed to be a very cl ear concept but
the anti nomi es proved that this was not so. Today, this concept has been
repl aced by that of an arbi trary subset of a gi ven set. I n addi ti on, the
bel i ef that all subsets of a gi ven set f or ma set is al most uni versal l y
accepted. However , i t is by no means true that these vi ews are shared by
all mathemati ci ans. Even Gdel hi msel f, who as the quotati on above
i ndi cates shoul d be counted among the Pl atoni sts, has once expressed
the vi ewthat the concept of an arbi trary subset of a gi ven set is inneed
of cl ari fi cati on(cf: Gdel , 1951).
The present wri ter bel i eves (al though he cannot present convi nci ng
evi dence to support this view) that i t i s inthi s di recti on where the future
of set theory lies.
6.2 Finitism and nominalism
One of the general mathemati cal concepts whi ch i s accepted wi thout
hesi tati on by all mathemati ci ans i s that of the repeti ti on of one and the
same operati on an arbi trary number of ti mes. It i s thi s concept whi ch
l i es at the bottomof the ari thmeti c of i ntegers and al so of the theory of
computati ons.
The most radi cal trend i n the f oundati on of mathemati cs, whi ch i s
represented by a Russi an school under the l eadershi p of Mar kov and
30 Andvzej Mostowski
Shani n (cf. 1958), wants to retai n inmathemati cs excl usi vel y comput abl e
sequences a(O), a(l), ... of i ntegers and such other concepts as are i mme-
di atel y reduci bl e to t hem. All other concepts are to be si mpl y rej ected.
Let us remi nd briefly what computabl e sequences are i norder to see how
much we l ose by adheri ng to this vi ew.
A sequence a(O), a(l), . . . of i ntegers i s sai d to be computabl e if for each
i nteger y2 the t er ma(n) can be comput ed inaccordance witha fi ni te set
of absol utel y expl i ci t and mechani cal i nstructi ons requi ri ng no mat hemat -
i cal i ngenui ty for thei r appl i cati on. We can al so say that a sequence is
comput abl e if i ts t er ms can be cal cul ated on a ni achi ne pr ogr ammed wi th
certai n finiteand expl i ci t sets of i nstructi ons and capabl e of perf ormi ng
certai n si mpl e operati ons such as pri nti ng or erasi ng symbol s on an
arbi trary l ong tape, or shi fti ng the tape in accordance wi th the gi ven
i nstructi ons. Such machi nes are cal l ed Turi ng machi nes (cf. Davi s,
1958).
The fami l y of comput abl e sequences is thus very smal l when we com-
pare i t wi th the fami l y of sets, sequences and functi ons admi ssi bl e inthe
cl assi cal theory. Of course, most of the resul ts of abstract set theory
woul d be i rreparabl y l ost if the approach of thi s radi cal school was
general l y adopted.
Less radi cal proposal s were made by mathemati ci ans and phi l osophers
who insist that sets consi dered in mathemati cs shoul d correspond to
properti es expressi bl e ina l anguage. Thi s is not an unreasonabl e poi nt
of vi ew: most sets one fi rst consi dered i nmathemati cs and all col l ecti ons
accepted i neveryday l anguage are defi ned ei ther by means of properti es
of thei r el ements or by the use of the pri nci pl e of abstracti on appl i ed
to a defi nabl e equi val ence rel ati on.
Many ways were proposed in the past of how to limit the concept
of a set so as to deal only wi th sets whi ch are defi ned i n one or other
sense of the word. We can subsume themall under the name nomi nal i sti c
trends.
The earl i est attempt in this di recti on was represented by Russel l s
rami fi ed theory of types (cf. 1910). In this theory sets are not onl y di vi ded
i nto types but each type is further di vi ded i nto orders. The order of a set
is determi ned by the f or mof i ts defi ni ti on.
Vari abl es of the l anguage used to express the rami fi ed theory of types
are al so di vi ded i nto types and orders. A set of type t i s counted to the
order n if and onl y if i t can be defi ned by means of a f ormul a whi ch sati s-
Sets 31
fi es the fol l owi ng t wo requi rements: 1) i t contai ns no vari abl es whose
types are ei ther of a type hi gher than t or of the type t but of an order
hi gher than n; 2) i t contai ns no bound vari abl es of type t and of order n.
These f ormal requi rements correspond to the f amous vi ci ous ci rcl e
pri nci pl e by means of whi ch Russel l wanted to el i mi nate all anti nomi es.
The pri nci pl e says that no total i ty can be defi ned i n t erms of total i ti es
to whi ch i t bel ongs . Noti ce that accordi ng to Russel l , each set of type t
and order n is i nvol ved in the constructi on of each set of a type hi gher
than t and al so of each set of type t and order hi gher than n. Also this set
bel ongs to the range of quanti fi ers whi ch bound vari abl es of type t and
order n. We thus see that the l i mi tati ons i mposed on f ormul ae whi ch
defi ne sets of type t and order n wer e so sel ected that the vi ci ous ci rcl e
pri nci pl e can not be vi ol ated i n the rami fi ed theory of types.
Extensi ons of the rami fi ed theory of types to i ncl ude certai n transfi ni te
types were proposed by Wang (see Fraenkel (1956) for a descri pti on of
his theory).
Al so the concept of a constructabl e set whi ch Gdel i nvented (1939
and 1951) inorder toprove the consi stency of the cont i nuumhypothesi s
was obtai ned by extendi ng the concept of sets defi ned inthe rami fi ed
theory of types.
However , Gdel al l owed arbi trary wel l -orderi ngs to occur as types
and orders, and so his constructabl e sets do not sati sfy the strong requi re-
ments of nomi nal i smwhi ch insist that all sets have to be defi ned by
f ormul ae. It is thi s non-nomi nal i sti c admi xture whi ch i s responsi bl e for
the fact that all axi oms of Z- F are true of constructabl e sets. On the other
hand, the fact that many set-theoreti cal probl ems (like the general
cont i nuumhypothesi s and the Sousl i n probl em) become sol vabl e inthe
domai n of constructabl e sets i s due to the fact that these sets, whi l e not
defi ned inthe absol ute sense of the wor d, are defi nabl e by i nducti on by
means of wel l -orderi ngs about whi ch we know much mor e than about
arbi trary sets.
The axi omstati ng that all sets are constructabl e was once, perhaps not
too seri ousl y, proposed by Gdel (1939) but rej ected by hi ma5 fal se
(1947). Yet t heaxi omi s consi stent rel ati ve to Z-F (cf. 1939 and 1951).
We shoul d sti l l menti on cal cul i i nvented by Lorenzen (cf. Fraenkel ,
1956 for thei r descri pti on) whi ch al so represent an i ncreasi ng hi erarchy of
systems conf ormi ng to the nomi nal i sti c thesi s.
At the other extreme we fi nd proposal s to repl ace the rami fi ed hi erarchy
32 Andrzej Mostowski
by a mor e nar r owfami l y of sets. Thus e.g. Weyl inhi s i nfl uenti al book
(1921) argued that onl y sets of order O shoul d be admi tted and the i dea
of hi s predi cati ve anal ysi s based on the noti on of sets of order O was
taken up by Grzegorczyk (1954), Krei sel (1960) and others.
The l i mi tati ons i mposed on the concept of a set by all the programs
enumerated above go very far i ndeed. They automati cal l y excl ude sets
defi ned by a chance event (e.g. by repeated tosses of a coi n) al though
such an i dea is very cl ear and shoul d not be abandoned wi thout seri ous
reasons. Besi des cl assi cal mathemati cs onl y the i ntui ti oni sti c mathemati cs
of Brouwer i s suffi ci entl y ri ch to admi t thi s concept. On the other hand,
Brouwer s i deas on l ogi c and set theory are so radi cal l y di fferent f romtl i e
usual ones that one can wonder whether even modest parts of Cantori an
set theory can be reconstructed in i ntui ti oni sti c mathemati cs (see e.g.
Fraenkel (1956) for a descri pti on of Brouwer s ideas).
CONCLUSION
Let us briefly summari ze our di scussi on of sets and thei r theory. Sets or
whatever other name we gi ve t hemineveryday l anguage are very useful
i n descri bi ng vari ous si tuati ons of life. We use t hembecause we possess
a capaci ty to l ook at vari ous separate i ndi vi dual s as if they f ormed
together a newobj ect of a hi gher type.
The concept of sets was taken over by mathemati ci ans who used i t to
organi ze and systemati ze the basi c concepts of thei r fi el d. Duri ng this
process they consi derabl y extended the concept of sets toi ncl ude infinite
sets and assumed, fi rst subconsci ousl y and l ater consci ousl y, certai n
pri nci pl es of constructi on of new sets f r omother sets. The f ree use of sets
in mathemati cal constructi ons brought about the creati on of new
mathemati cal theori es whi ch weremor e abstract and mor e general than
the ones previ ousl y cul ti vated. The new theori es al l owed the mathemati -
ci ans to i ntroduce new concepts whi ch proved useful not onl y i npure but
al so i n appl i ed mathemati cs. I n addi ti on, studi es on f oundati ons of set
theory proved to be extremel y frui tful : they prompt ed the devel opment
of l ogi c, showed the necessi ty of a correct use of l anguage and of
di sti ngui shi ng l anguage f rommeta- l anguage. The appearance of set-
theoreti cal anti nomi es and the apparent i mpossi bi l i ty of sol vi ng many
si mpl e questi ons concerni ng sets on the basi s of axi oms of set theory
Sets 33
rai se doubts whether in spi te of the t remendous successes of the set-
theoreti c approach to mathemati cs the very concept of set i s real l y as
cl ear and perspi cuous as is general l y bel i eved. Neverthel ess, abstract set
theory i s bei ng constantl y devel oped and the obj ects of set-theoreti cal
studi es become mor e and mor e abstract.
The sayi ng of A.N. Whi t ehead (1933) can very appropri atel y be appl i ed
to set theory - i t is l i ke Ophel i a: very beauti ful , but a littlemad.
REFERENCES
BECKER, O. , (1964) Grundlagen der Mathematik in geschichtlicher Entwicklung, 2nd
BOLZANO, B., (1851) Parudoxien des Unendlichen, Prague.
BOURBAKI , N. , (1949) Foundat i ons of mat hemat i cs for the worki ng mathemati ci an ,
CANTOR, G. , (1932) Gesammelte Abhandlungen mathematischen und philosophisches
COHEN, P., (1966) Set Theory and the Continuum Hypothesis, New Yor k- Amst er dam.
DAVI S, M. , (1958) Computability and Unsolvability, NewYor k.
FRAENKEL, A. , (1922) Zu den Grundl agen der Cant or- Zermel oschen Mengenl ehre , in
FRAENKEL, A. , (1953) Abstract Set Theory, Amst er dam.
FRAENKEL, A. and BAR- HI LLEL, Y. , (1956) Foundations of Set Theory, Amst er dam.
FREGE, G. , (1884) Die Grundlagen cler Arithmetik. Eine logischmathematische Unter-
suchung ber den Begriffder Zahl, Bresl au (Engl i sh transl ati on by L. Austi n: The
Foundation of Arithmetic, 2nd edi ti on, New York, 1960).
FREGE, G. , (1 893-1903) Grundgesetze der Arithmetik, begriffsclzriflich abgeleitet, vol . I
and II, J ena.
GDEL, K., (1931) Ober f ormal unentschei dbare Stze der Pri nci pi a Mat hemat i ca und
verwandter Syst emI, Monatshefte fr Mathematik und Physik, 37, 173-198.
GDEL, K., (1939) Consi stency proof for the general i zed cont i nuumhypothesi s ,
Proceedings of the National Academy of Sciences of the U.S.A., 25, 220-224.
GDEL, K., (1951) Tlie Consistency of the Axiom of Choice and of the Generalized
Continuum Hypothesis with the Axioms of Set Theory, 2nd edi ti on, Pri nceton.
GDEL, K., (1947) What is Cant or cont i nuumprobl em? , Amer. Math. iMontlzly, 54,
GOODMAN, N. , (1951) The Structure of Appearance, Cambr i dge (Mass.).
GRZEGORCZYK, A. , (1 954) El ementari l y defi nabl e anal ysi s, Fundamenta Mathematicae,
HEI J ENOORT, J. VAN, (1967) From Frege to Gdel. A Source Book in Mathematical Logic
KELLEY, J . L., (1955) General topology, New York.
edi ti on, Frei burg- Muni ch.
i n The Journal of Symbolic Logic, 14, 1-8.
Inhalts (edi ted by E. Zermel o), Berl i n.
Mathematische Annalen, 86, 230-237.
51 5-525.
41, 311-338.
1879-1931, Cambr i dge (Mass.).
34 Andrzej Mostowski
KREI SEL, G. , (1960) La prdi cati vi t, Bulletin de la Socit Mathmatique de France, 88,
LUSCHEI , E. C., (1962) The Logical System of LeSniewski, Amst er dam.
LUSIN, N., (1927) Sur l es ensembl es anal yti ques , Fundamenta Mathematicae, 10,
MAHLO, ~. , (1912-1913) Zur Theori e und Anwendung der p- Zahl en , Berichte ber die
Verhandlungen der Schsischen Akademie der Wissenschaften zu Leipzig (math. - phys.
KI.), 64 (1912), 108-112, and 65 (1912-1913), 268-282.
MORSE, A. P., (1965) A Theory of Sets, New Yor k- London.
NEUMANN, J. VON and MORGENSTERN, O. , (1947) Theory of Games and Economic
Behavior, 2nd edi ti on, Pri nceton (N.J.), Pri nceton Uni v. Press.
PEANO, G. , (1889) Arithmetices principia, novo methodo exposifa, Turi n (cf. G. Peano,
Opere scelte, Roma, 1958 and Engl i sh transl ati on inHei j enoort, 1967).
POLYA, G. , (1954) Induction and Analogy in Mathematics, Vol. 1, Oxf ord.
QUINE, W. V., (1969) Set Theory and its Logic, 2nd edi ti on, Cambr i dge (Mass.).
ROSSER, J. B., (1969) Independence Proofs, New York.
RUSSELL, B., (1910) See Whi t ehead and Russel l (1910).
SHANI N, N. A. (ed.) (1958) Probl emy konstrukti vnogo napravl enya v mat hemat i ke
Sborni k rabot , Trudy Mafematiieskogo Instituta im. V. Steklova, 52, I zd. Ak. Nauk
SSSR, Moscow- Leni ngr ad.
SOLOVAY, R. , (1965) 2 H can be anything it ought to be. The theory of models (edi ted by
J. W. Addi son), Amst er dam, p. 435.
TARSKI , A. and KEI SLER, J. H. , (1963-1964) From accessi bl e to i naccessi bl e numbers ,
Fundainenta Mathematicae, 53,268-282.
WEYL, H., (1921) Das Kontinuum, Lei pzi g.
WHI TEHEAD, A. N. , (1933) Adventure ofldeas, New York, Macmi l l an.
WHITEHEAD, A. N. and RUSSELL, B., (1910) Principia Mathematica Z, Cambr i dge;
WIGGINS, D., (1967) Zdentity and Spatia-temporal Continuity, Oxf ord.
371-391.
1-95.
2nd edi ti on: Cambr i dge 1925.
2 The concept of structure
JEAN PIAGET
INTRODUCTION
I n cogni ti on, one of the most spontaneous tendenci es i s toproceed f r om
the si mpl e to the compl ex; i n pri nci pl e thi s i s l egi ti mate, provi ded we
agree on what we mean by si mpl i ci ty - a hi ghl y rel ati ve concept.
Di ffi cul ti es ari se f romthe fact that there are t wo ki nds of si mpl i ci ty -
obj ecti ve si mpl i ci ty (i nherent i n the obj ects) and subj ecti ve si mpl i ci ty
(i.e. deri vi ng f romour own fami l i ar actions). An exampl e of obj ecti ve
si mpl i ci ty i s si mpl e addi ti on - addi ng one -, whi ch al l ows us to move
f romI to I + I = 2, f rom2 to 2 + I = 3, and so on. On the other hand,
i t woul d be a matter of subj ecti ve si mpl i ci ty, when the subj ect, accustomed
to fashi oni ng a whol e by the assembl y of antecedent parts, general i zes
this atti tude to a case where he is cal l ed upon to i nterpret an organi zed
total i ty; inthis i nstance, a ki nd of apriori or preconcei ved i dea l eads hi m
to begi n by l ooki ng for the el ements of whi ch the totality mi ght be
composed, treati ng i t as a si mpl e combi nati on of these el ements.
Now, i t i s at once apparent that al though the + I operati on is el emen-
tary, i t real l y expresses a l awof f ormati on: the l awexpressed by H.
Poi ncar by the symbol n + I , whi ch governs the enti re cl ass of natural
number s (or of all whol e number s if, inaddi ti on, + I i s i nverted i nto - I);
i nthi s case the shift f romthe si mpl e tothe compl ex does not destroy the
l atters properti es and the el ementary operati on is i nno way at odds wi th
the bui l di ng of a structure i n the sense i nwhi ch we shal l shortl y defi ne
i t. On the other hand, when psychol ogi sts i nthe last century (and far on
i nto the present century) ori gi nal l y regarded a percepti on (a compl ex
real i ty such as the i mmedi at e apprehensi on of an obj ect, a geometri cal
fi gure, a face, etc.) as havi ng to be the product of a set of initial sensa-
ti ons whi ch were si mpl y associ ated, or agai n, when l i ngui sts before
Ferdi nand de Saussure saw l anguage pri mari l y as an assembl age of
36 Jean Piaget
words, and thought that they had exhausted thi tasks of l i ngui sti cs once
they had traced the hi story of each one, in both cases they proceeded
uncri ti cal l y f r omthe si mpl e tothe compl ex, invi rtue of a tendency mor e
subj ecti ve than obj ecti vel y substanti ated, whi ch i mpel l ed t hemto begi n
by breaki ng down the whol es they were studyi ng i nto i ndi vi dual el ements,
mor e or l ess real (the words), or even partl y artificial (the sensati ons
consi dered as psychol ogi cal rather than essenti al l y physi ol ogi cal units).
A sort of reversal of thi s natural tendency of the i ntel l ect was, therefore,
needed i n order to thi nk that a whol e mi ght have l aws of i ts own as a
total i ty; and that the met hod that had at fi rst seemed the onl y one
possi bl e, because i t started f rompre-exi sti ng el ements, i n fact tended to
conceal these l aws of totalities or to di stort t hemrather seri ousl y.
Structural i sm, whose obj ect is to get at preci sel y these l aws of systems,
emerged onl y bel atedl y, then, inthe devel opment of sci ence. For exampl e,
geometry, the source, si nce the end of the ni neteenth century, of such
spl endi d exampl es of structural i st anal ysi s, remai ned curi ousl y opaque to
this lineof thought among the Greeks: Eucl i d s Elements never even got
to establ i shi ng i nterconfi gurati onal rel ati onshi ps ( space and its co-
ordi nates, i ncontradi sti ncti on tothe i nternal properti es of the confi gura-
ti ons) and the theorems di d not approach the questi on of the trans-
f ormati on group whi ch they i mpl i ci tl y used.
In actual fact, if we take structural i sm in i ts stri ct sense, wi thout
extendi ng i t to i ncl ude all the specul ati ons about f orms i nthe stati c sense
(f romAri stotl e onwards), or about the organi smas a hi gher uni t, ani -
mat ed by a vi tal force , etc., a structural i st met hod onl y emerged i nsci ence
wi th Gal oi s s di scovery of the group structure i n al gebrai c trans-
f ormati ons. Yet thi s fi rst model al ready presents the essenti al charac-
teristics whi ch are to be f ound inthe vari ous structural i st methodol ogi es
proper to the di fferent scientific di sci pl i nes (di sregardi ng, of course, the
abuses to whi ch thi s t er mhas gi ven ri se and whi ch are di squi eti ngl y
rampant incertai n contemporary fads). We may, therefore, take Gal oi s s
group structure as our starti ng-poi nt insetti ng out to defi ne structures.
1. GENERAL FEATURES OF STRUCTURES
1.1 The fi rst property of a structure i s that i t consti tutes a total i ty, wi th
l aws of i ts own; that i s, i t has l aws as a system, i ndependentl y of t he
The concept of structure 37
parti cul ar characteri sti cs of i ts el ements. I n the case of a group these
properti es are the fol l owi ng:
a) The i ntroducti on of an operati on combi nabl e wi th itself, that we
shal l call T. To take our earl i er exampl e of obj ecti ve si mpl i ci ty, thi s
will be the operati on +n whi ch fi gures i n the addi ti ve group of whol e
numbers, and enabl es any i nteger to be added to any other.
b) To each operati on T there corresponds an i nverse T-1 whi ch i n this
speci fi c case will be the subtracti on -n.
c) The product of an operati on T and of i ts i nverse T-1 i s a neutral
el ement O, whose combi nati on with others does not modi f y t hem:
n + O = n (i denti cal operati on).
d) The combi nati ons are associ ati ve: (I + m) + n = I + (m+ n).
These four properti es thus characteri ze the whol e system, consi dered
as a total i ty, in contradi sti ncti on to the parti cul ar properti es of the
el ements (odd or even numbers, number s di vi si bl e by 3, etc.). I n other
structures than the group, the properti es of the total i ty will obvi ousl y be
di fferent, but there al ways are such properti es, di sti nct f r omthe properti es
of the el ements.
1.2The second general characteri sti c of a structure is that these l aws of
the totality operate on transformati ons, i nthe broadest sense, and not on
stati c characteri sti cs. I n other words, the structure consi sts of a system
of operati ons whose combi nati ons transf ormone t er mi nto another. I n
our earlier addi ti ve group exampl e, the fact of addi ng or subtracti ng one
number to or f r omanother l eads to a thi rd number (except as regards
+ O), etc. This characteri sti c of transf ormati on i s parti cul arl y stri ki ng in
the case of the f undamental groups of the vari ous types of geometry,
formerl y consi dered model s of purel y descri pti ve sci ences, deal i ng onl y
wi th f orms or stati c fi gures. To begi n wi th, the consti tuent group of
Eucl i di an geometry is the transf ormati on group whi ch al l ows the posi ti on
of an el ement to be changed, wi thout changi ng i ts f or mor si ze. I f we
vary di stances, so that onl y angl es, paral l el l i nes and strai ght l i nes
remai n unchanged, we get the group of transformati ons of si mi l i tude. If
we further vary angl es, we get the affi ne transf ormati ons group. By
modi f yi ng paral l el l i nes i n addi ti on we get the proj ecti ve groups and,
if we do not retai n even strai ght l i nes, we come to the group of topol og-
ical equi val ents or homeomor phi sms. Each of these groups bei ng a sub-
group of the succeedi ng group, wi th the possi bi l i ty of passi ng f r omone to
38 Jeun Piaget
the other, we end up witha general systemof transformati ons whi ch has
prof oundl y renovated geometri cal studi es.
1. 3 A thi rd characteri sti c of structures is l ess frequentl y brought to noti ce,
at l east wi th the vocabul ary we shal l be adopti ng. Thi s i s thei r sel f-
regul atory property, that is: the i nternal combi nati ons of the structure
never gi ve products outsi de the structure (a whol e number added to
another al ways gi ves a whol e number again), whi l e constantl y enri chi ng
it, and never i nvol ve any extraneous el ement. I n the case of operati onal
structures thi s sel f-regul atory property goes wi thout sayi ng and is si mpl y
one wi th the l aws of combi nati on. When we come to bi ol ogi cal , psycho-
l ogi cal or soci al structures, we shal l find sel f-regul ati on by a whol e range
of homeostati c mechani sms. But i t i s possi bl e to concei ve of all the
transi ti ons f r omcyberneti c l oop structures, wi th regul ati on proceedi ng
f r omthe resul t of the acti ons, to operati onal systems where pre-
correcti ons and the combi nati on of anti ci pati ons and retroacti ons
consti tute perfect regul ati ons. The psychogeneti c devel opment of the
cogni ti ve functi ons enabl es us to wi tness i na concrete f ormthi s passage
f r omregul ati on to operati on.
Let us further note wherei n thi s thi rd characteri sti c of structure woul d
seemto be responsi bl e for the noti on s success: whether we take the
sort of constructi ve aut onomy that i ts sel f-regul atory property gi ves to 8
f ormal structure, or whether we take the organi sati on i nherent in the
sel f-regul ati ng el ements of a l i vi ng system, we arri ve inei ther case at an
expl anatory pri nci pl e far superi or to the hypothesi s of l i nkages establ i shed
after the fact between pre-exi sti ng el ements.
2. MATHEMATICAL STRUCTURES
2.1 The fecundi ty of mathemati cs comes f r omi ts unl i mi ted power of
constructi ng operati ons upon other operati ons and, si nce any one
operati on call never be i sol ated, but al ways ties i n wi th the others,
mathemati cs can accordi ngl y be l ooked on as a general theory of struc-
tures. Gi ven t wo sets E and F, i t i s possi bl e fi rst of all, to appl y an x inE
on to one (and onl y one) y i nF, resul ti ng i na functi onal operati on that
may be one- t o- one (if there is onl y one x) or many- t o- one (if there are
several x). The product E x F of the t wo sets can be got; or al ternati vel y
The concept of structure 39
thei r quoti ent-set , by a parti ti on based on an equi val ence rel ati on
(e.g. the set manki nd di vi ded by the rel ati on Co-ci ti zens gi ves the set
nati ons). Or agai n, the set of subsets may be extracted f r omany set by a
combi natori al ; or, by repeati ng operati ons, an overal l scal e on the base
E-F may be obtai ned. Thus, i t is possi bl e to f or mstructures, regardl ess
of the nature of the basi c sets, by i denti fyi ng common properti es through
the operati ons carri ed out on these sets, and the vari ous structures can
fi nal l y be compared. This poi nt marks the start of the unrestri ctedl y
fl exi bl e range of morphi sms . I n parti cul ar, there will be sai d to be
i somorphi smbetween t wo structures where there is one- t o- one corres-
pondence both between thei r el ements and between thei r rel ati ons
(i ncl udi ng di recti ons). Li kewi se we shal l hear of uni val ent if not mul ti -
val ent theori es l i nki ng i somorphi c structures.
The Bourbaki school has di sti ngui shed i tsel f in thi s re-thi nki ng of
mathemati cs as a theory of structures, inseeki ng, i nthe rel ati ons between
the structures, to arri ve at sorts of f ormal fi l i ati ons whi ch, t hrough
successi ve constructi ons, can comprehend the enti re fieldof mathemati cs.
The pri nci pl e of thi s i s tobegi n by getti ng back, by the determi nati on of
the common properti es (i somorphi sms), to a mi ni mal number of parent
structures not reduci bl e to each other but abl e, by thei r combi nati ons,
to engender all other structures. The probl emis then to determi ne, as
t hough one wer e deal i ng wi th the growth or the evol uti on of l i vi ng
organi sms, the di fferenti ati on and co-ordi nati on processes gi vi ng ri se
to these deri vati ve and mor e parti cul ari zed structures. The di fferen-
ti ati on takes the f ormof compl i cati ng one of the f undamental structures
by i ntroduci ng successi ve requi rements or restri cti ve axi oms, thus defi n-
i ng the possi bl e substructures (by a process anal ogous to that descri bed
i nSecti on 1.2 for the f undamental groups of the vari ous geometri es); or,
al ternati vel y of reduci ng a strong structure to weaker structures
(e.g. a semi - group, whi ch woul d be associ ati ve, but woul d l ack a neutral
el ement). Co- ordi nati on on the other hand takes the f ormof i mposi ng
t wo structures at a timeon a si ngl e set of el ements, as, for i nstance, for
al gebrai c topol ogy.
Thi s vast enterpri se was crowned with compl ete success and has
prof oundl y al tered our overal l vi si on of mathemati cs, inthe sense that the
mul ti pl i ci ty of sectors or branches hi therto di sti ngui shed sol el y by thei r
content has been superseded by a si ngl e edi fi ce, inwhi ch everythi ng is
comprehended but whose archi tecture has become purel y structural , i.e.
40 Jeun Piuget
hi ngi ng sol el y on the f orms, wi thout regard to the nature of the el ements
and hence to the contents, with thei r di versi fi cati ons and artificial
di stri buti ons. I n thi s way the authors effect an admi rabl e general i zati on
of the structural i zati on whi ch had al ready gai ned acceptance i n certai n
domai ns duri ng the ni neteenth century, fol l owi ng the di scovery of the
group structure, but for whi ch i t has thus become possi bl e to adduce sti l l
other structures insupport.
The f undamental structures arri ved at by the initial i somorphi sm-
based reducti on process turned out, i n fact, to be three innumber , not
on the grounds of a priori necessi ty, but on the grounds of an i nducti vel y
ori ented i nventory. The fi rst consi sts of the al gebrai c structures, whose
prototype i s the group, and whose di sti ncti ve characteri sti c i s that they
al l owof reversi bi l i ty through i nversi on or negati on: T.T-1 = O. The
second f undamental structure i s the order structure, whose prototype
is the l atti ce, beari ng on the succeedi ng-precedi ng rel ati ons of the
el ements, any t wo of whi ch al ways have a l east upper bound (the l east
of the succeedi ng el ements or supremum) and a greatest l ower bound
(the hi ghest of the precedi ng el ements or injmum). Earl i er conj ectured by
Dedeki nd, thi s structure was studi ed mai nl y by Bi rkhoff, and i ts uni ver-
sal i ty is comparabl e to that of the group. It is thi s structure, for exampl e,
that characteri zes the rel ati ons operati ve in the sets of sub-sets (or
si mpl exes) di scussed inthe earl y part of thi s secti on. As inall systems of
rel ati ons, i ts reversi bi l i ty is based upon reci proci ty, the l awof dual i ty of
l atti ces transf ormi ng A. B precedes A+B i nto A+B succeeds A.B by
permutati on of the (.) and (+) and of the rel ati ons precedi ng and
succeedi ng . Lastl y, the thi rdtype of f undamental structure i s topol ogi cal
in nature, based on the noti ons of nei ghbourhood, conti nui ty and limit.
2.2 To arri ve at the epi stemol ogi cal si gni fi cance of the noti on of structure
inmathemati cs, the fi rst thi ng is toask onesel f if these three f undamental
structures, whose di fferenti ati ons and co-ordi nati ons can then produce
all the other structures by a quasi -organi c process, are or are not
natural , i nthe sense inwhi ch we speak for exampl e of natural numbers
to i ndi cate that they had al ready been worked out by pre-sci enti fi c
thought. Of course, the use of structures by mathemati ci ans i nvol ves
thei r formal i zati on, and the axi omati c met hod empl oyed inso doi ng does
not hel p us answer the questi on. It i s, however, i mportant to note that
thi s ki nd of met hod i s i n no way i ncompati bl e with the exi stence of
The concept of structure 41
i ntui ti ve correspondents attesti ng the entrenchment of the f undamental
structures i n the mechani sms of the i ntel l i gence itself, t hough in an
obvi ousl y el ementary, if not to say rudi mentary, f orm.
I n fact, if we study the spontaneous devel opment of the fi rst l ogi cal and
mathemati cal operati ons i nthe chi l d, we noti ce that they are organi zed
f romthe start i nstructures such as cl assi fi cati ons, seri ati ons, correl ati ons,
etc. whi ch certai nl y are not passed down by adul t teachi ng (as thei r
appreci abl e l i mi tati ons are enough to show). (This research was carri ed
out before peopl e had turned, as is now the fashi on, to tryi ng to teach
Cantor ini nfant school s, i nci dental l y wi thout the l east suspi ci on of what
the chi l d can do on hi s own and i n hi s own l anguage. ) Now an
exami nati on of these natural structures shows ri ght away that they
bel ong to three general f orms. The fi rst i s f ounded on reversi bi l i ty
through i nversi on and covers the operati onal groupi ngs * by f ormati on
of cl asses: A - A = O, etc. The second characteri zes the rel ati onshi p
groupi ngs and i s f ounded on reversi bi l i ty through reci proci ty. The thi rd
concerns parti ti ons or envel opi ngs beari ng upon the cont i nuumand
bri ngi ng i n the noti ons of nei ghbourhood and separati on, i nteri or or
exteri or, boundari es, etc. Her e, therefore, there i s an evi dent ki nshi p with
the three f undamental structures; and compari son remai ns possi bl e when,
at later stages of the subj ect s devel opment, we fi nd himbecomi ng abl e
to co- ordi nate the i nversi ons N and reci procal s R, i mpl i ci tl y usi ng for the
purpose a quaterni on or Kl ei n group, such that, for exampl e, an i mpl i ca-
ti on (whi ch we can wri tep q = I) will i ncl ude i ts i nverse (N= p.@, i ts
reci procal (R = q 3 p) and i ts correl ati ve (C= pq), l eadi ng to the
commutati ve composi ti ons NR = Cy NC = R, RC = N and NRC = I.
I f the cstructuresy thus have natural roots, then thi s means that, starti ng
f romthese el ementary f orms, the i ntel l ect, through a combi nati on of
refl ecti on-by-abstracti on and of new operati ons carri ed out upon the
previ ous ones, i s abl e to embar k on an unbroken sequence of con-
structi ons. Structural i smis thus by no means i ncompati bl e withmat he-
mati cal constructi oni sm, but, on the contrary, i mpl i es it. A further
i ndi cati on of thi s is provi ded by the fact that, wi thi n the Bourbaki school
i tsel f, there is a strengtheni ng tendency today to make preponderant use
of the category structure descri bed by McLane and Ei l enberg, whi ch
* For the notion of a grouping, the interest of which is genetic rather than formal,
see section 4,
42 Jean Piaget
bears upon cl asses of obj ects pl us thei r functi ons and morphi si ns. Thi s
structure is bui l t upon the very operati ons that gave ri se to the f oregoi ng
structures, hence by refl ecti on-by-abstracti on and by reorgani zati on
on a hi gher pl ane (it may, i nci dental l y, be noted that categori es are
represented intrivial natural f orms and at a l evel earl i er than the opera-
ti onal constructi ons, i nthe course of the devel opment of the intelligence).
A thi rd i ndi cator of thi s constructi oni smi nherent i n structural i smand
extendi ng the natural constructi ons of the human intellect is that
ari si ng f r omGdel s well known theoremof the limits of formal i zati on.
First l et us si mpl y note that formal i zati on has limits, and that outsi de
these limits there are operati onal constructi ons inacti on whi ch are not
(or not yet) formal i zed, thus i mpl yi ng some ki nd of rel ati onshi p wi th
natural structures. Secondl y, if, for the compl ete formal i zati on of a
structure (that is, for everythi ng i n i t to become deci dabl e), another
stronger structure i s requi red, thi s gradati on in the strength of the
systems i mpl i es unl i mi ted constructi on. Thi rdl y, Fef erman and Schtte
have recentl y succeeded indemonstrati ng the exi stence, wi thi n transfi ni te
sets, of a number Kappa (KO) whi ch marks the limit of predi cati vi ty.
Beyond thi s limit there i s therefore no further effecti ve constructi vi ty,
that i s, constructi ons based on a combi natori al . On the other hand, new
possi bi l i ti es may then be opened by means of what must be cal l ed rel a-
ti ve recursi vi ty and deci dabi l i ty. Suppose we have a cl ass So in whi ch
everythi ng i s deci dabl e, excepti ng a non- deci dabl e proposi ti on ND1. If,
for reasons exteri or to the system, we arri ve, by hypothesi s, at the bel i ef
that i t i s true (or false), we get the set S1 (=So + NDl), whi ch i s rel a-
ti vel y deci dabl e i n rel ati on to ND1. Suppose we then have a second
undeci dabl e proposi ti on ND2; i f we can treat thi s in the same way, we
then get the set 52 (=S1 + ND2) whi ch is rel ati vel y deci dabl e in
rel ati on to ND2, and so on, by transfi ni te repeti ti on. We thus get a
hi erarchy of degrees of sol ubi l i ty, i n whi ch each stage is i ncreasi ngl y
strongl y i rreduci bl e to that precedi ng i t. It may, therefore, be concl uded
that the set of possi bl es i s not a cl osed set and that each di scovery (beari ng
i n this i nstance upon the NDs) opens new possi bi l i ti es, whi ch further
demonstrates the uni on between structural i smand constructi oni smtaken
i nthe epi stemol ogi cal sense: i.e., operati onal constructi ons concei ved as
consti tuti ve, and not merel y as condi ti ons of accessi on.
The concept of structure 43
3. PHYSICAL AND BIOLOGICAL STRUCTURES
3.1 There are sti l l mor e compel l i ng reasons for seeki ng a rel ati onshi p
between l ogi co- mathemati cal structures and the systems of transf orma-
ti ons at wor k i nthe knowi ng subj ect, and even partl y i nobj ects. Namel y,
the exi stence of physi cal structures, and the questi on of thei r rel ati ons
wi th those of the physi ci st-subj ects structures; and the exi stence of
bi ol ogi cal structures whi ch mi ght even be consi dered the source of a
subj ects own structures, inhis psychol ogi cal functi oni ng.
The physi ci sts wor k l ong consi sted essenti al l y in seeki ng l aws, by
appl yi ng l ogi co- mathemati cal operati ons to obj ects, and thus deter-
mi ni ng functi onal equati ons whi ch express the dependence between
measurabl e vari abl es. However , partl y because of the i ncreasi ng di ffi cul -
ti es of measurement on the di fferent scal es i nvol ved, and partl y due to the
overpoweri ng need to understand and expl ai n, rather than si mpl y to note
and predi ct phenomena, physi cs has turned i ncreasi ngl y tothe el aborati on
of model s, whose deducti ve character sati sfi es the i ntel l ect, but whi ch
are al so and al ways expected to be attri butabl e to the obj ects t hemsel ves.
But the constructi on of model s l eads to the di scovery of structures, and
contemporary physi ci sts are getti ng to the poi nt of seeki ng the structure
before they have even got the measurements.
Now the structures thus brought tol i ght inphysi cal phenomena are the
same i npart as those whi ch the mathemati ci ans have mastered, and thi s rai -
ses the t wo rel ated probl ems of the adequati on of mathemati cs to materi al
real i ty on the one hand and of the degree of obj ecti vi ty or subj ecti vy of
the structures thus attri buted tothe real on the other. I n a broad seme,
physi cal real i ty i s a vast hi erarchy of nested structures, on scal es rangi ng
f r omthe most mi nutel y nucl ear tothe most sweepi ngl y cosmol ogi cal , and,
on all these scal es, the experi mental resul ts are mathei nati zabl e to a poi nt
where a great many of t hemcan be reconstructed deducti vel y by theoret-
ical model s appreci abl y surpassi ng the i ntui ti ve representati ons infi nesse
and power. However, as a resul t of the twi n facts of the i rreversi bi l i ty of
timeand the essenti al rl e of probabi l i ty consi derati ons - whi ch partl y
amount to the same thi ng infact - the mathemati zati on is not necessari l y
a matter si mpl y of reduci ng the model s to the maj or structures di scussed
inSecti on 2. Thi s makes i t all the mor e stri ki ng how rel ati vel y frequentl y
the l atter do enter the pi cture: the group structures i nparti cul ar occur
wi thsurpri si ng persi stence at all orders of magni tude, f rommi crophysi cs
44 Jean Piaget
to rel ati vi ty theori es, achi evi ng a remarkabl y accurate correspondence
wi th certai n transf ormati ons of the real .
Thi s bei ng the case, shoul d we therefore assume that the group
appertai ns tothe obj ects, expressi ng thei r properti es i ndependentl y of us,
or on the contrary shoul d i t be vi ewed si mpl y as a superi or i nstrument
of representati on, or agai n (an i ntermedi ate possi bi l i ty) as transl ati ng the
experi mental acti ons that we perf ormon the obj ects and the co- ordi na-
ti ons between observers tied to di fferent systems of reference? I n fact,
these three eventual i ti es are all real i zed, varyi ng wi th the si tuati on; and
the poi nt of i nterest is preci sel y the bond whi ch l i nks t hem, deri vi ng f rom
the nature of causal expl anati on.
I n effect, to the extent that physi cs thi nki ng seeks the reason for the
l aws observed, i t i s obl i ged to go beyond the fronti ers of the observabl e
si nce i ts i mmedi ate experi ence embraces onl y movement s or changes of
state, whereas the probl emof expl anati on boi l s down to arri vi ng at
necessary l i nks between rel ati ons observed. However , any readi ng of the
facts itself presupposes the use of operati onal recordi ng i nstruments
(i nstruments i nthe doubl e sense of l ogi co- mathemati cal operati ons and
materi al mani pul ati ons, wi th all the transi ti onal stages possi bl e between
the two); when i t comes to combi ni ng the establ i shed data i nto a system
that provi des an expl anati on of t hem, the sol uti on can onl y consi st i n
co-ordi nati ng the l aws and attempti ng to understand how the obj ects
gi ve ri se to these regul ari ti es; but si nce the l atter are i nseparabl e f r omthe
operati ons whereby they may be determi ned, i t is sel f-evi dent that the
systembei ng sought will consi st ks t and f oremost i n a co-ordi nati on of
these operati ons, that is, ina structure whose transformati ons are reck-
oned to express the acti ons of the obj ects themsel ves. I n other words, the
transi ti on f r oml aw- seeki ng to cause-seeki ng will consi st in repl aci ng
operati ons whi ch are si mpl y appl i ed to obj ects by operati ons whi ch can
be attri buted to the obj ects, that is, by the same operati ons, rei nforced
by a systemof transformati ons. All the psychogeneti c research on the
devel opment of the noti on of causabi l i ty has demonstrated t hi s rl e of
operati onal transf ormati ons attri buted to the real as soon as they are
f ramed by the subj ect: transi ti vi ty i nthe case of transmi ssi ons, addi ti ve
composi ti ons, mul ti pl i cati ve composi ti ons (proporti ons and di stri buti v-
ity), reci proci ti es (acti ons and reactions), etc.
We thus see that the group structure can i ndeed be regarded al ternatel y
as a property of the obj ects (when they l end themsel ves to it), of the ex-
The concept of structure 45
peri mental mani pul ati ons, or of the subj ect al one l i mi ti ng himself inthi s
case to anti ci patory appl i cati ons. As regards the mor e general probl em
of the adequati on of mathemati cs to the real , a key i s no doubt to be
f ound inthe anal ysi s of bi ol ogi cal structures.
3.2 A l i vi ngorgani smcompri ses l aws of a total i ty, a compl ex of con-
ti nui ng transformati ons and systems of sel f-regul ati on. It is thus
unquesti onabl y a structure, i n the sense of the defi ni ti on i n Secti on 1,
but one whose secrets are unfortunatel y sti l l far f r omenti rel y reveal ed.
Wer e they all reveal ed, that structure or those structures woul d present
the consi derabl e i nterest of havi ng some of the characteri sti cs of physi co-
chemi cal mechani sms whi l e at the same t i me bei ng at the ori gi n of the
structures of behavi our and of cogni ti ve functi ons; si nce the organi smis
thus both an obj ect and the starti ng poi nt of what consti tutes the subj ect,
i t i s probabl y in thi s uni on that we are l i kel y tofi nd the answer to the prob-
lemj ust posed, of the har mony, not pre-establ i shed but i ntri nsi cal l y
establ i shed, between mathemati cal and physi cal structures.
I n the meanwhi l e, the known data of bi ol ogy al ready offer some
l essons whi ch are hi ghl y i nstructi ve on the general theory of structures.
The fi rst is the manner i nwhi ch, at all l evel s of the organi sm, what mi ght
have seemed to be made up of aggregates (wi th si mpl e associ ati ons
between i ndependent el ements) turns out on anal ysi s to be a systemwi th
i ts l aws of the total i ty and above all i ts regul atory mechani sms. Thus, at
the begi nni ng of thi s century, the genome was regarded as a col l ecti on of
genes wi th no rel ati ons between t hem, each seemi ngl y responsi bl e for the
heredi tary transmi ssi on of a si ngl e i ndi vi dual characteri sti c. For thi s
concepti on, that Mayr humorousl y l abel l ed beanpi l e geneti cs, the facts
have l ed to the substi tuti on of a mor e structural i mage: several genes
pl ay a part i nthe transmi ssi on of any gi ven characteri sti c and each gene i s
i nvol ved in the transmi ssi on of several characteri sti cs ( pol ygeny and
pl ei otropi sm or, i n other words, several -to-one and one-to-several
correspondences) and the whol e operates as an orchestra and not as a
col l ecti on of sol oi sts as Dobzhansky has remarked. Fr omthe genome
l evel onwards we get the i nterventi on of regul atory mechani sms, and
what Lerner has cal l ed a geneti c homeostasi s . The most stri ki ng fact i n
thi s connexi on i s that the natural uni t is not so much the i ndi vi dual
genome or the genotype as the popul ati on , or mi xture of bal anced
strai ns, wi th its geneti c pool in whi ch the genotypi cal heredi ti es are
46 Jeun Piaget
bl ended. And the bl end itself, i nstead of f ormi ng a si mpl e aggregate, as
one mi ght have supposed, is i n i ts turn the seat of stabi l i zi ngregul ati ons,
as has been shown by the cl assi c experi ment by Dobzhansky and Spassky
on fourteen strai ns, mi xed artificially, where after a phase of i mbal ance
we get the recurrence of a good proporti on of the ori gi nal s survi val and
recombi nati on coeffi ci ents.
In a wor d, the facts j ust ci ted i l l ustrate cl earl y what was sai d at the
start of thi s arti cl e, namel y, that i ndeal i ng withunfami l i ar processes, we
begi n by fol l owi ng the natural bent of the i ntel l ect i n tryi ng to wor k
f romthe si mpl e to the compl ex by a sort of atomi sti c compoundi ng
process, but sooner or l ater we di scover l aws of total i ty that i mpose a
structural composi ti on, and i t is onl y then that the obj ecti ve expl anati ons
begi n.
The second structural i st l esson provi ded by moder n bi ol ogy i s the
f undament al i mportance of sel f-regul atory mechani sms and f eedback
ci rcui ts. We have j ust seen this at the l evel of the genome. Waddi ngt on
used the name homeor he~i s~ for the ki nd of dynami c equi l i brati on
whi ch, in the course of embryogenesi s, bri ngs the embryo back to i ts
necessary paths (or creods ) whenever devi ati ons occur; now thi s is not
si mpl y an effect of heredi tary programmi ng, si nce ontogenesi s deri ves
f roman epi geneti c system i n whi ch the syntheses take pl ace i ni nterac-
ti on wi th the envi ronment. On the physi ol ogi cal pl ane of the adul t
organi sm, we are pretty well acquai nted with the mul ti pl e homeostases
whose descri pti on deri ves f r omthe researches of Cl aude Bernard and
Cannon and whose del i cate sel f-regul ati ons are bei ng reveal ed by
contemporary research. As regards the nervous system, i t fol l ows u
fortiori that the si tuati on is the same: to take onl y one exampl e, i t is
known today that the so-cal l ed refl ex arc i s by no means a si mpl e associ a-
ti ve arc , bat i tsel f i ncl udes a homeostati c l oop system.
All of this i s hi ghl y i mportant for structural i sm, f romat l east t wo
poi nts of vi ew. Fi rstl y, we fi nd that, al ongsi de the f ormal structures
whose operati onal sel f-regul ati on deri ves f r omthei r cl osure and the
i ntri nsi c necessi ti es whi ch go wi th cl osure, the l i vi ng structures, whi l e
remai ni ng rel ati vel y open (si nce they compri se constant exchanges wi th
the outsi de and not onl y among thei r own substructures, as wi th cl osed
structmes) make use of temporal and dynami c systems of sel f-regul ati on
whi ch represent a ki nd of outl i ne or an i ncompl ete f ormof the opera-
ti onal systems. Secondl y, these sel f-regul atory systems, becomi ng
The concept of structure 47
i ncreasi ngl y mobi l e and acti ve withthe mul ti pl i cati on of the organi sm s
exchanges with the external worl d throughout i ts l earni ng and devel -
opment , i nfact consti tute the source of the cogni ti ve structures whi ch, on
the pl ane of the human i ntel l ect, l ead to l ogi co- mathemati cal opera-
ti onal structures.
4. PSYCHOLOGICAL STRUCTURES
Hence i t i s the psychogeneti c study of the f ormati on of these l atter
structures whi ch provi des the means of j usti fyi ng the hypothesi s of a
transi ti on f r omstructures of the sel f-regul ati on or homeostasi s type to
structures of the operati onal or l ogi co- mathemati cal type.
The Gestal t theory, born i n1912, can al ready be col rsi dered structural -
ist, its pri nci pl e bei ng to seek f romthe start l aws of total i ti es i n all areas
(percepti on, moti vi ty, memor y, intelligence), and to consi der the el ements
not as pri mary but as bei ng structured f r omthe fi rst, rather t han bei ng
structuri ng . But thi s school of psychol ogy, whi l e havi ng the great meri t
of exposi ng the mi sconcepti ons of the associ ati oni sts, wi th thei r atomi sti c
model s, made the mi stake of tryi ng toreduce everythi ng to a si ngl e type
of structure, the Gestalt, and of formul ati ng this concept on the pattern
of the fi el ds of physi cs, notabl y the el ectro-magneti c fi el ds. Fr omthe
fact that the l aws of total i ti es are other than the l aws of thei r el ements,
whi ch i s true, Werthei mer and Koehl er drewthe fal se concl usi on that a
necessary concomi tant of any psychol ogi cal or cogni ti ve structure is a
non- addi ti ve composi ti on. But el ementary operati onal structures such
as a cl assi fi cati on, a seri ati on, or the seri es of natural number s are
ri gorousl y addi ti ve, even t hough they i nvol ve l aws of total i ti es above and
beyond the parti cul ar properti es of thei r el ement s. This shows f r omthe
start that the structures of i ntel l i gence are not reduci bl e to the structures
of percepti on, these l atter bei ng i ndeed non- addi ti ve and fi tti ng the
defi ni ti on of the Gestalt (for exampl e, if A < B < C, the el ement B will
be perceptual l y overesti mated if compared to A and underesti mated if
compar ed to C, through a perceptual exaggerati on of the di fferences, a
fact whi ch excl udes addi ti vi ty as i t does quanti tati ve conservati on
itself).
I n i ts research on the devel opment of the cogni ti ve functi ons, the
Geneva school therefore set i tsel f the task of studyi ng the di verse
48 Jean Piaget
vari eti es of structure i nthei r f ormati ons and above all i nthei r fi l i ati ons,
so as to understand i nparti cul ar what the l ogi co- mathemati cal structures
spri ng f rominthe way of psycho-bi ol ogi cal sources.
An i nstructi ve initial fi ndi ng is that, as earl y inthe chi l ds devel opment
as the sensori - motor stage, i.e. before l anguage, we note the growth of
certai n structures such as the group of di spl acements, al though at this
stage i t i s a matter of co-ordi nati on between successi ve acti ons (in par-
ti cul ar, the reversi bi l i ty and the associ ati vi ty of the group, correspondi ng
to returns and detours) and not yet a matter of si mul taneous represen-
tati ons. At thi s sensori - motor stage, i nci dental l y, many other general co-
ordi nati ons are to be noted, but as yet wi thout a structurati on comparabl e
to that of the spati al group: co- ordi nati ons between acti on schemes
accordi ng to order of successi on, or by uni ons, i ncl usi ons, correspon-
dences, etc., f r omwhi ch will be deri ved at l ater stages by refl ecti ve
abstracti on the rel ati ons whi ch consti tute the representati onal structures.
When representati on becomes possi bl e, wi th the acqui si ti on of semi oti c
tool s and l anguage, we fi rst have a phase i n whi ch acti ons are i nteri ori zed,
f ol l owed at age 5 to 7 on average, by the el aborati on of what may be
cal l ed the consti tuent functi ons expressi ng the i nterconnexi ons proper
to the acti on schemes and the dependences between the vari abl e charac-
teristics of the obj ects; for exampl e, understandi ng that pushi ng on a flat
rectangul ar bl ock near one of i ts corners makes i t turn i nstead of movi ng
f orward ina strai ght l i ne, or that if l i qui d is poured f r omA i nto By the
l evel drops i nA if i t ri ses inB etc. The resul t is the f ormati on of el emen-
tary structures of categori es , inparti cul ar i nthe case of several - to- one
appl i cati ons. But nei ther i nverse correspondences nor reversi bi l i ty in
general , nor above all quanti tati ve conservati ons are yet grasped, so that
we cannot as yet tal k of operati ons.
At the fol l owi ng l evel (7-10 years of age), we get the crystal l i zati on of
the operati ons whi ch will be cal l ed concrete in that they bear di rectl y
upon obj ects and not as yet on hypotheses: cl assi fi cati on, seri ati on,
correspondence, the sequence of whol e number s by synthesi zi ng i ncl usi on
and order, etc. These nascent operati ons, sources of conservati ons due to
thei r reversi bi l i ty and to a previ ousl y unrecogni zed transi ti vi ty, thus
consti tute wel l -defi ned operati onal structures that are, however, essenti al -
ly l i mi ted by thei r step- by- step composi ti on, as yet l acki ng the combi na-
tori al and, hence, sets of subsets. These structures, whi ch we have
cal l ed groupi ngs , are hal f - way between the group proper (though
The concept of structure 49
wi thout compl ete associ ati vi ty) and the lattice (wi th suprmum but
wi thout infimum or the reverse) and present the psychogeneti c i nterest
of prepari ng the way to both; and, addi ti onal l y, of provi di ng el ementary
exampl es of al gebrai c structures wi th thei r i nversi ons, of ordered
structures wi th thei r reci proci ti es, and of topol ogi cal structures, as we
have seen i n Secti on 2.
Lastl y, at the l evel of proposi ti onal or f ormal operati ons (age 11-12)
we get operati ons carri ed out on these operati ons, whi ch then gi ve us a
combi natori al (combi nati on operati ons bei ng cl assi fi cati ons of all
possi bl e cl assi fi cati ons of TZ el ements, and permutati on operati ons bei ng
seri ati ons of seri ati ons) and the INRC group, whi ch co-ordi nates i n a
si ngl e systemi nversi ons and reci proci ti es. A seri es of substructures then
becomes possi bl e and advances i n thi s l ogi co- mathemati cal structurati on
are echoed i n the domai n of causal i ty by the ascri pti on of a seri es of
attri butes to the obj ects, expressi ng i n ki nemati c and dynami c t er ms the
newl y constructed operati onal transformati ons (e.g. acti ons and reacti ons,
vector composi ti on of forces, di stri buti vi ty inthe i nternal modi fi cati ons
of a system, etc.).
The mai n l esson of thi s psychogenesi s of structures is i ts demonstrati on
of the possi bl e and even necessary uni on between structural i smand
constructi oni sm. None of these structures whose devel opment has j ust
been very bri efl y outl i ned coul d be consi dered an i nnate i dea or an a
priori necessi ty, but each is constructed f romthe structures precedi ng i t
by a combi nati on of refl ecti ve abstracti ons, drawi ng certai n co- ordi na-
ti ons f romsi mpl er systems, and of reorgani zati ons or reconstructi ons
whi ch inthe fi nal anal ysi s consi st i n carryi ng out second power operati ons
on the precedi ng operati ons unti l the creati on of a coherent new whol e.
I n other words, each structure requi res a genesi s, but each genesi s consi sts
i na f ormati ve passage f roma si mpl er structure to a ri cher one that i nte-
grates the si mpl er ingoi ng beyond i t.
Further, the f ormati ve pri nci pl e of t hese successi ve structures consi sts
i n sel f-regul ati ons or equi l i brati ons whi ch, duri ng the phases of el ab-
orati on, mani fest themsel ves i n vari ous trial and error approaches, wi th
progressi ve co-ordi nati on of retroacti ons and anti ci pati ons, as i nsi mpl e
f eedback regul ati ons, but whi ch i n thei r fi nal f or mare operati onal
structures. I n t hi s respect, operati onal reversi bi l i ty can be taken as the
product of the equi l i brati ons, and as the l i mi t t er mof the semi -reversi bi l i -
ti es proper to the regul ati ons. The psychogenesi s of the structures thus
50 Jean Piaget
comes in between the homeostases of vari ous bi ol ogi cal l evel s and the
constructi on of the l ogi co- mathemati cal structures.
Fr omthi s poi nt of vi ew, the i ntri nsi c necessi ty whi ch characteri zes
t hese last structures is seen not as the product of antecedent condi ti ons
but as a termi nal characteri sti c produced by the cl osi ng of systems that
remai n open duri ng the phases of thei r f ormati on but whi ch, once
compl eted, become sel f-contai ned, and are dependent onl y on thei r
i nternal composi ti ons. A parti cul arl y cl ear exampl e i s the rel ati onshi p of
transi ti vi ty, of the f ormA < B, B< C therefore A < C. This i s not acknowl -
edged as a necessary rel ati on before the stage of concrete operati ons, but
i t becomes evi dent as soon as the subj ect becomes capabl e of constructi ng
a seri ati on by an exhausti ve met hod wi thout trial and error, worki ng
si mul taneousl y on the rel ati ons < and >.
5. LINGUISTIC A N D SOCIAL STRUCTURE
A l anguage is a soci al i nsti tuti on (inDurkhei m s sense), even t hough
i ndi vi dual speech may expl oi t the possi bi l i ti es of the l anguage i n con-
stantl y changi ng f orms; so an exami nati on of l i ngui sti c structures is
undoubtedl y the best and most exact met hod of graspi ng at once the
mechani smof the soci al structures and the degree of i ni ti ati ve whi ch, in
any i nstance, they l eave to or requi re of i ndi vi dual s. It i s not wi thout
reason, therefore, that the structural i smof a Lvi -Strauss shoul d rel y on
l i ngui sti cs as much as, if not mor e than, on al gebrai c structures.
Li ngui sti c structural i smdates back to F. de Saussure, who descri bed i t
i n t er ms of systems , and i ntroduced t wo f undamental i nnovati ons: first,
that a l anguage is a genui ne whol e whose el ements are i nterdependent in
thei r opposi ti ons no l ess than inthei r connexi ons; and second, that thi s
system, i nasmuch as i t i s a phenomenon of equi l i bri um(here, De Saussure
drewhi s i nspi rati on f romthe wor k of economi sts), depends essenti al l y on
the synchroni c condi ti ons, so that the hi stori cal , etymol ogi cal , etc. and
even comparati ve researches of pre-Saussure l i ngui sti cs di d not get at
any reasons behi nd l i ngui sti c phenomena.
By thei r characteri sti cs of total i ty and sel f-regul ati on, as well as by the
pl ay of thei r i nternal transformati ons, the l i ngui sti c structures di scovered
by De Saussure are ina sense comparabl e to the cogni ti ve i ntel l ectual
structures whi ch we have j ust been di scussi ng, despi te certai n obvi ous
The concept of stvuctzire 51
di fferences in the nature of the transformati ons i nvol ved (not that
l anguage has not a certai n l ogi c of i ts own, if onl y i n the rel ati onshi p
between subj ect and predi cate). But there is a f undamental di sti ncti on
between systems of si gns, f romwhi ch l anguage stei ns, and operati onal
structures, whi ch bear upon noti ons and operati ons, that i s, upon what is
si gni fi ed rather than the si gni fi ers. I ndeed, si nce si gns are conventi onal
and hence partl y arbi trary ( absol utel y or rel ati vel y so, as the case may
be, as De Saussure specifies), the synchroni c state of the systemmay be
i ndependent of i ts previ ous states invaryi ng degrees (accordi ng as needs
have changed, as i nthe hi story of markets ineconomi cs); on the contrary,
an operati onal structure i mposes itself withnecessi ty - i ts state of equi l i b-
ri umi s the end resul t of i ts hi story, that is (as was seen inSecti on 4),
of a process of progressi ve equi l i brati on. The rel ati onshi p between
di achrony and synchrony i s thus whol l y di fferent in the structures of
l anguage and of the i ntel l i gence.
Lvi -Strauss took l i ngui sti c structures as a model for structural i sm
i n ethnol ogy, and the pri macy of the synchroni c factor l ed him, in
parti cul ar, to under-appreci ate hi stori es and genesi s. As regards genesi s,
however, recent wor k inl i ngui sti cs has radi cal l y modi fi ed i ts structural i st
perspecti ves. After Harri s and Hal l e had emphasi zed the conti nuous
creati vi ty di sti ncti ve of l anguage, si nce speech consi sts i n an endl ess
constructi on of new combi nati ons, N. Chomsky worked out the noti on
of generati ve gr ammar s and provi ded the met hods of studyi ng t hem,
posi ti ng i nci dental l y, as the starti ng-poi nt of this constructi ve process, an
i nnate fi xed kernel of a rati onal ki nd. Thus, certai n posi ti ons that had
been tradi ti onal si nce De Saussure s l i ngui sti cs have been rul ed out
(Bl oomfi el d, etc.): synchrony and genesi s cease to be contradi ctory;
reason or l ogi c are no l onger products of l anguage, but the reverse, etc.
We then see the har mony between thi s constructi oni smand the construc-
ti oni smwe f ound in our research on i ntel l i gence (Secti on 4); the onl y
di fference i s the l i ngui sts hypothesi s of the i nnateness of the initial
fi xed kernel , but this hypothesi s i s unnecessary tothe extent that sensori -
mot or i ntel l i gence, ante-dati ng l anguage, suffi ces toprovi de the condi ti ons
for the devel opment of thi s initial kernel , and that thi s initial f or mof
cogni ti on i s i tsel f the product of successi ve constructi ons and equi l i bra-
ti ons.
Turni ng now tothe properl y soci al structures, a consi derabl e number of
f orms can be descri bed, wi th the same spread as in psychol ogy: f rom
52 Jeun Piaget
si mpl e homeostati c f eedback syst ems, poi nted out by the economi sts;
through quasi -operati onal f orms (likethe i nterl ocki ng of l egal l anguage
nor ms so prof oundl y anal ysed by H. Kel sen s normati vi sm ); to true
operati onal structures such as those that we have sought to bri ng to
l i ght i n the mechani smof co- operati on i n cogni ti ve exchanges (when
reci proci ty occurs wi thout the i nterventi on of constrai nts or of authori ty).
It i s i mportant, however, not to mi suse the t ermstructure - as i t is
currentl y fashi onabl e to do: i ts use is not j usti fi edsi mpl y because one is
di scussi ng totalities. The great vi rtue of Durkhei m s soci ol ogy i s to have
spotl i ghted these totalities but, by regardi ng themas si mpl y emergi ng
f r oma uni on of i ndi vi dual s, and by attri buti ng thei r regul ati ons to an
overl y general process of cc~n~t r ai nt ~ , he mi ssed achi evi ng thei r struc-
tural i st anal ysi s (actual l y i ni ti ated by hi s di sci pl e, Marcel Mauss).
Parsons structural -functi onal systemon the other hand goes further
t owards thi s ki nd of anal ysi s, parti cul arl y as regards the mechani sms of
equi l i bri umand val ori zati on, and has the vi rtue of not di ssoci ati ng
structure and functi ons, i mperati ve whenever functi oni ng takes pl ace
t hrough ti me.
The prototype of a structural i smwhose i ntenti on is qui te authenti c
i s the wor k of Lvi -Strauss, who di d not content hi msel f withl i ngui sti c
i nspi rati ons but set hi msel f, wi th the hel p of mathemati ci ans, to
f ormul ate in al gebrai c terms struetures that he di scovered in ki nshi p
rel ati ons, etc. Two basi c aspects of his theory are to be noted i n thi s
connexi on, the fi rst of whi ch seems to compel assent, whereas the second
poses probl ems. Fi rstl y, Lvi -Strauss concl udes that hi s structures are
di sti nct f roml aws or general rel ati onshi ps that can be establ i shed by
si mpl e observati on, for such l aws do not by t hemsel ves provi de the raison
dtre that woul d expl ai n themin thei r detai l . J ust as the physi ci st, to
i nterpret hi s phenomena, is obl i ged to go beyond the observabl e and
to construct model s embraci ng the transformati ons attri buted to the
obj ects as such, so the structures i nvoked by the ethnol ogi st are both the
product of deducti ons whi ch go beyond the l i mi t s of the phenomenon
al one and at the same timethe expressi on of a reality concei ved of as
obj ecti ve even though not i mmedi atel y di scerni bl e inthe facts. I n other
words, the structures exi st or are deemed to exi st, but remai n unknown to
the member s of the soci al group; and thi s is preci sel y the wri ter s own
posi ti on on the terrai n of the cogni ti ve structures, whi ch express what the
subj ect i s capabl e of (operati ons, etc.) at a gi ven stage of devel opment, but
Xhe concept of structure 53
not what he can say or what he is consci ous, i nhi s f ormul ated thi nki ng,
of bei ng abl e to do.
But the second tenet of the doctri ne goes farther than the fi rst; as a
resul t of hi s over- emphasi s on synchroni c consi derati ons and hi s rel ati ve
depreci ati on of hi story and genesi s, Lvi -Strauss is l ed to the assumpti on
that hi s unobservabl e structures express a sort of permanent synchrony,
or i n other words, that they bear wi tness to the exi stence of a human
mi nd conti nui ngl y i denti cal wi th i tsel f, qui te overl ooki ng the necessi ty of
psychogeneti c or bi ol ogi cal consi derati ons. He thus seems to i gnore* the
probl emof what has happened between chi l dhood and adul thood or
between the hi gher pri mate and Man.
The probl emthat then i nevi tabl y ari ses i s that of the mode of exi stence
of these structures: inwhat does thei r bei ng consi st if they are i mmut abl e?
Do they hark back toPl atoni c or phenomenol ogi cal essences - a noti on
that Lvi -Strauss di scl ai ms? Do they ori gi nate inan i nnateness i n the
sense of Chomsky s fi xed kernel ? I f so, the i nterpretati on of the great
ethol ogi st K. Lorenz, who is at once a Kant i an and a mutati oni st,
reveal s the dangerous l i mi tati ons of such a hypothesi s, and we shal l
come back to thi s. Do they not, then, resul t f r omthe compel l i ng and in
fact very general character of equi l i brati on and sel f-regul ati on
mechani sms? Such appears to be the onl y remai ni ng and, i ndeed, prima
facie reasonabl e sol uti on, but inthis case then the common and stabl e
kernel of such f ormati ve mechani sms resul ts sol el y f r omthei r functi oni ng,
a concl usi on that in no way excl udes but, on the contrary, requi res the
compl ement ary hypothesi s of the constructi on of endl essl y new structures
that resul t f r omthe i nteracti on between this functi oni ng and the i m-
medi ate demands of the moment .
6. CONCLUSION: STRUCTURALISM AND CONSTRUCTIONISM
Structural i smis a met hod and not a phi l osophy - a met hod consi sti ng in
i denti fyi ng the chara-cteri sti cs of a whol e regarded as somethi ng mor e
than a si mpl e aggregati on of antecedent el ements, and arri vi ng at these
characteri sti cs by processes whi ch may be f ormal or partl y experi mental
as the case requi res, but whi ch are al ways veri fi abl e. As such, structural i sm
* This is actually not the case, as his more recent work reveals (cf. LHomme nu).
54 Jeun Piaget
i s necessari l y i nterdi sci pl i nary; i nthe domai ns of l ogi c and mathemati cs,
whi ch are whol l y sel f-suffi ci ent di sci pl i nes, the fact remai ns that, to
establ i sh the ful l epi stemol ogy of the structures, and not stop short at
thei r techni cal aspects, they must be compared withthe structures whose
psychogeneti c constructi on can be f ol l owed. However , the study of thi s
constructi on sooner or l ater calls i nto pl ay our knowl edge of physi cal ,
bi ol ogi cal and soci al structures, and reci procal l y, comprehensi on of these
three sorts of structure requi res recourse as much to psychogeneti c
anal ysi s as to the f ormal met hods of constructi on. I n a wor d, any
structure is al ways l ocated at the i ntersecti on of a mul ti pl i ci ty of di sparate
di sci pl i nes, so that no general theory of structures can possi bl y escape
the requi rement that i t be not si mpl y mul ti di sci pl i nary but authenti cal l y
i nterdi sci pl i nary.
The fi rst i nsi ghts afforded by such a general theory, even though i t i s
sti l l an embr yo and not a ful l -fl edged di sci pl i ne, demonstrate that in all
fi el ds structural i smgoes hand i nhand withsome sort of constructi oni sm.
Whet her the l atter remai ns abstract and non- t emporal as i nmathemati cs,
where i t i s neverthel ess i mposed by the limits of formal i zati on, or
whether i t i s an i ntegral part of a real genesi s, as inall other di sci pl i nes,
the fact remai ns that the esse of the structures deri ves f romthei r
structurati on, that is, f romthe sel f-regul ati ons and operati ons of whi ch
they are the products.
Thi s bei ng granted, if we di scount, i n hypothesi s, the Pl atoni smwhi ch
never adds anythi ng to our effecti ve knowl edge (for the onl y met hod of
arri vi ng at the eternal I deas is to reconstruct t hem, and thi s constructi on
is sel f-suffi ci ent) there appear to be onl y t wo possi bl e sources, as regards
the putati ve ori gi ns of the structures, once they are concei ved of as
natural : ei ther they are i nnate or el se they are the resul t of sel f-regul a-
ti ons. Now the advantage of the hypothesi s of i nnateness i s that, i nthe
abstract, i t can be transl ated i nto a pri ori sm, ergo i nto t er ms of necessary
antecedent condi ti ons. I n the concrete, thi s means that the probl emi s
si tuated in the fieldof bi ogenesi s, whi ch superfi ci al l y may appear sounder
than that of psychogenesi s. An attempt has been made by the ethol ogi st
K. Lorenz to effect thi s synthesi s between Kant and neo- Darwi ni sm.
Accordi ng to hi m, the maj or categori es of knowl edge (and one can i ncl ude
the structures here) are the product not of experi ence or of the envi ron-
ment but of antecedent bi ol ogi cal condi ti ons comparabl e to a here-
di tary pref ormati on; j ust as the horse s hooves and the fi sh s fins devel op
The concept of structure 55
duri ng ontogenesi s well before these creatures use t hem, si mpl y by
heredi tary programmi ng, so, says Lorenz, Kant was perfectl y correct in
regardi ng our cogni ti ve categori es as predetermi ned structures acti vated
i mmedi atel y on contact wi th experi ence, wi thout thi s meani ng that they
resul t f romit. But the troubl e is that heredi ty vari es f r omspeci es to
speci es, and hence, there i s nothi ng general and above all nothi ng
necessary about i ts content. So, sel f-consi stentl y but droppi ng the
necessi ty whi ch is the most f undamental cri teri on of Kant i an a prioris,
Lorenz ends by regardi ng the latter as si mpl y i nnate worki ng hypotheses ,
and as such, therefore, they are i mposed on us but they l ack i ntri nsi c
intelligibility.
These anal yses by Lorenz seemtothi s wri ter to be of great i mportance
because they demonstrate unmi stakabl y the l i mi tati ve character of an
i nterpretati on i nt er ms of i nnateness. The consti tuent sel f-regul ati ons of
organi c life are i n fact much mor e general (si nce they are al ready i npl ay
even at the stage of the transmi ssi on of heredi tary characteri sti cs, the
genome as such havi ng i ts own sel f-regul ati on) and mani fest themsel ves
i n a far mor e easi l y veri fi abl e manner inthe evol uti on of behavi our and
duri ng the enti re psychogenesi s of cogni ti ve functi ons, i ncl udi ng
structures. To expl ai n the f ormati on and i ndeed the very nature of
structures, therefore, in the present state of our knowl edge, we fi nd
oursel ves turni ng to sel f-regul ati ons and to the constructi ons that these
entai l .
Fr omthe epi stemol ogi cal poi nt of vi ew, the process i nherent in pro-
gressi ve equi l i brati on, wi th i ts al ternati ons between di sequi l i bri umor
confl i cts of coherence and re-equi l i brati ons on new l evel s, natural l y calls
to mi nd di al ecti cal concepti ons; and on thi s poi nt agai n we need to
underscore the necessary uni on of structural i smand di al ecti cal con-
structi ons, notwi thstandi ng that some deni grators of the f ormer, as
superfi ci al as any sl aves of fashi on, have posi ted a f undament al confl i ct
between these t wo met hods of thi nki ng. I n the domai n of f ormal struc-
tures, i t is well known (and Bachel ard stressed the fact inhi s Philosophie
du non) that, as soon as a structure or even a theory has been constructed,
one tri es denyi ng an essenti al character of i t, to get beyond thesi s and
anti thesi s and reach a new synthesi s; the non- Eucl i di an or non- Ar chi me-
dean geometri es, the non- commut at i ve al gebras, the l ogi cs wi thout
excl uded mi ddl e (si nce Brouwer) and even wi thout negati on (Gri ss), etc.,
are i nstances of those di al ecti cal moment s pecul i ar to the hi story of
56 Jean Piaget
l ogi co- mathemati cal structures. I n all natural and soci al sci ences, the
reconci l i ng of the temporal , hi stori cal or geneti c perspecti ve wi th
structural i st tendenci es i mposes, a forti ori , a conti nuous di al ecti c whi ch
i s sai d to be i mmanent in the real , as well as necessary to the subj ect
seeki ng to understand the real .
I n fact, thi s is the very poi nt on whi ch we choose to concl ude thi s paper
on structural i sm, resol utel y rej ecti ng the i dea, propounded by some, that
thi s f ecund met hod necessari l y rul es out any consi derati ons of hi story,
genesi s, functi oni ng, and even a subj ects acti vi ti es.
REFERENCES
BOUDON, R., (1968) A quoi sert Zu notion de structure, Pari s.
BOURBAKI , N., (1948) L archi tecture des mathmati ques , in: F. Le Li onnai s (ed.), Les
grands courants de la pense matlimatique, Pari s.
CHOMSKY, N., (1957) Syntactic Structures, The Hague.
Di al ecti que marxi ste et pense structural e , (1968) Cahiers du Centre dEtudes Socia-
Gense et structure, (1965) Pari s.
GRANGER, G. G. , Pense formelle et sciences de lhomme.
J OI NER, J. H., (1968) Essentials of the Theory of Structures, New York.
KUHN, T. , (1968) Structure of Scientijic Revolutions, Chi cago (Ill.).
Lm- STRAUSS, C. , (1958) Anthropologie structurale, Pari s.
LvI - STRAUSS, C., (1962) La pense sazrvage, Pari s.
LEWI N, K., (1951) Fielcl Theory in Social Science, New Yor k.
PARSONS, T. , (1960) Structure and Process in Modern Societies, Gl encoe (Ill.).
PI AGET, J., (1967) Biologie et connaissance, Pari s, Gal l i mard (Seri es L aveni r de la
PI AGET, J., (1968) Le structuralisme, Pari s, Presses Uni versi tai res de France.
PI AGET, J. et al., (1967) Logi que et connai ssance sci enti fi que, in: R. Queneau (ed.),
SI NCLAI R DE ZWAART, (1967) Acquisition du langoge et dveloppement de la pense,
Le structural i sme , (1966) Aletheia, 4.
Structural i smes, i dol ogi e et mt hode , (1967) Esprit (May).
Temps modernes, (1966) 246 (Nov.).
VI ET, J., (1965) Les mthodes structuralistes dans les sciences sociales, Pari s.
listes, 76-81 (Feb.-May).
science).
Encyclopdie de la Plade, Paris, Gal l i mard.
Pari s.
3 Systems concepts
MIHAJLO D. MESAROVIC
1. CONCEPT OF SYSTEM: ORIGIN AND PRESENT ROLE
Duri ng the l ast t hree decades there has been an ever-i ncreasi ng i nterest in
the phenomena whi ch can broadl y be cal l ed i nf ormati on-processi ng
and deci si on- maki ng. These are of great i mportance insci ence engi neeri ng
and management . Observati on data are transf ormed for speci fi c purposes.
A need was felt for a concept to rel ate resul ts obtai ned f r omvari ous
subfi el ds, devel oped i ndetai l but i ni sol ati on, i.e. for a systemas under-
stood inthe newand rather broad sense of systems theory or systems
sci ence, and i ni ts appl i ed f orms (such as systems engi neeri ng or systems
management). I n general , a systemdenotes the existence of reIationships
between data or variables or, mor e speci fi cal l y, a transf ormati on of a set
of data i nto another set. As di fferent physi cal theori es deal wi th di fferent
ki nds of physi cal phenomena, vari ous aspects of i nf ormati on-processi ng
and deci si on- maki ng theori es can deal withvari ous types of systems, thei r
behavi our, transformati ons, control , and so on.
Does a f ormal , mathemati cal , defi ni ti on of a systemi ntroduce a too
severe restri cti on, especi al l y indeal i ng withcompl ex bi ol ogi cal or soci al
probl ems? It can readi l y be shown that the l i mi tati on, if any, is onl y
nomi nal . For exampl e, if a descri pti on i s expressed onl y as a set of state-
ments, the f ormal aspects of the observati ons can sti l l be expressed as a
system. To every set of statements a mathemati cal model can be assi gned,
i.e. a systemto represent the f ormal aspects of the theory. If the system
has very littlestructure and inthis sense does not reveal much about the
functi oni ng of the real -l i fe phenomena, thi s i s due to the l i mi tati on of our
knowl edge about the behavi our of the real -l i fe systemrather than due to
the appl i cati on of f ormal , mathemati cal , methods.
Consi der now some met hods for mor e detai l ed descri pti on of a
system. First step inthat di recti on i s to recogni ze cause-effect rel ati on-
58 Mihujlo D. Mesurovic
shi ps i n the system; this is done by parti ti oni ng the f ami l y of system
obj ects i nto t wo sets: the sets of inputs (or sti mul i ), X, and the set of
outputs (or responses), Y. The systemis then represented as a rel ati on on
t wo sets.
scxx Y (1)
where X = VI X ... x Vk, Y = Vk+1 x ... x Vn. The systemingeneral
is then consi dered to transformi nputs i nto outputs.
There are t wo basi c probl emareas i n the systems fi el d: 1) How to
descri be the systemmor e speci fi cal l y, i .e., how the i nputs are transf ormed
i nto outputs? 2) What are the basi c properti es of a systemof a gi ven ki nd
(i.e. witha gi ven type of descri pti on)?
Starti ng f r omthe concept of a general systemas a rel ati on on abstract
sets there are t wo approaches by whi ch the systemcan be descri bed in
mor e detai l . I n the fi rst approach the systemi s consi dered to process
certai n i nf ormati on fl owand a mor e detai l ed speci fi cati on i s gi ven in
t er ms of a mechani sm by whi ch this i s done. Thi s i s referred to as
termi nal approach and the systemi s consi dered as a termi nal system(or
causal system). In the second approach the systemis consi dered to be
pursui ng a certai n goal (or purpose) and the output of the systemis
descri bed i nterms of a goal -seeki ng process, as a response to a gi ven
i nput (sti mul us). These systems are referred toi ngeneral as goal -seeki ng
systems or mor e speci fi cal l y as control systems, deci si on- maki ng systems
and the like.
2. TERMINAL SYSTEM
I f S i s a functi on, i .e. withevery i nput, x e X, there i s associ ated a si ngl e
output, i t i s referred to as a functi onal system
S : X + Y (2)
A basi s then exi sts to descri be the systemby an i nternal mechani sm
whi ch i s essenti al l y a procedure to speci fy the ri ght output for any gi ven
el ement of the i nput set. Thi s is usual l y done by means of a set of equati ons
or a transf ormati on tabl e incase of systems withl ower cardi nal i ty, For
exampl e, l et X be a sequence of real numbers restri cted to the i nterval
TIo =(0,1, ..., l O), i . e. anyxEXi s a sequencex=(XO, XI, ..., XZO) wherext
Systems concepts 59
is a real number ; f urthermore l et Y be al so a sequence on T10 but such
that the fi rst el ement i s al ways a gi ven number , say y0 = v.. A functi onal
systemS : X + Y can be then defi ned by the di fference equati on
i.e. for any 2 E X, the correspondi ng output p E Y i s a sequence such that
i ts fi rst el ement i s M whi l e all other el ements are obtai ned (generated) by
(3). Equati on (1) represents a constructive specijcation of the system. The
type of the constructi ve speci fi cati on whi ch can be used to descri be a
systemdepends of course, on the type of sets X and Y. E. g. if X and Y are
sets of conti nuous functi ons constructi ve speci fi cati on can be gi ven by
means of di fferenti al or parti al di fferenti al equati ons.
If S i s a proper rel ati on i t fi rst has to be transf ormed i nto a functi on
before a constructi ve speci fi cati on can be gi ven. Two approaches have
been devel oped: state space approach and probabi l i sti c approach.
I n the state space approach a new obj ect, Z, i s i ntroduced fi rst so that
the system, S c X x Y, is made a f uncti on
S : Z x X - + Y (4)
If Xand Y are ti me functi ons Z i s al so a set of timef uncti ons, e. g. z is a
sequence of el ements f roma set 2, z = ( ~ 0 ~ ~ 1 , ..., zn); the fi rst el ement of
z is termed initial state, an arbi trary el ement i s termed the state of the
systemand the set 2 is the state space. The constructi ve speci fi cati on is
now gi ven int er ms of t wo sets of equati ons: fi rst gi vi ng the el ements of z
f romx and the initial state, e.g.
(5)
Zf = 428-1 + 23t2-1
and the second gi vi ng the el ements of the outputs for the gi ven el ements of
z, e.g.
Yi = 32% (6)
I f at for any gi ven ti me, i, the state of the system, zi, and the remai ni ng
i nput (xi, xi+1, ... xn) are gi ven, the remai ni ng output sequence of the
systemi s compl etel y speci fi ed. I n thi s sense the state of the systemcom-
pl etel y speci fi es the i nternal condi ti ons of the syst emat any gi ven ti me.
I nstead of i ntroduci ng new obj ects (likestate space), one can regai n
functi onal i ty in descri bi ng a systemby consi deri ng groups of si mi l ar
i nputs (sti mul i ) and then descri bi ng how such groups of i nputs map i nto
60 Mihajlo D. Mesarovic
groups of outputs. Thi s l eads to a probabi l i sti c approach. One fi rst des-
cri bes the i nput and output sets i nprobabi l i sti c t er ms; i.e. Xis parti ti oned
i nto subsets and wi th each el ement of a gi ven subset there i s associ ated a
number smal l er than uni ty. If some addi ti onal techni cal condi ti ons are
sati sfi ed the i nterpretati on of such a constructi on is that the number
i ndi cates the probabi l i ty wi th whi ch a gi ven el ement occurs. The system
itself i s then descri bed inreference tothese probabi l i ti es or some deri ved
stati sti cal characteri sti cs. Apparentl y one regai ns a functi onal descri pti on
of the systemby movi ng toa di fferent l evel of descri pti on, i.e. rel ati ng
subsets rather than el ements of the system s obj ects.
Syst ems whi ch are functi onal or for whi ch there i s gi ven a state space
constructi ve speci fkati on are termed deterministic systems. Systems
whi ch are descri bed vi a probabi l i sti c approach are termed probabilistic or
statistical systems. Apparentl y probabi l i sti c systems are but a speci al cl ass
of rel ati onal , non-determi ni sti c, systems si nce the appl i cati on of
probabi l i sti c approach requi res the sati sfacti on of a number of fai rl y
restri cti ve techni cal condi ti ons. Of parti cul ar i nterest is the probabi l i sti c
descri pti on of a systemas an information channel inwhi ch the si gnal s are
descri bed int er ms of a l ogari thmi c-type probabi l i sti c measure.
Syst ems can be cl assi fi edaccordi ng to the type of thei r obj ects. E.g. if X
and Y are sets of t i me functi ons, one tal ks about time systems. If X and Y
are descri bed al gebrai cal l y, one tal ks about algebraic systems. Si mi l arl y,
if the state space i s a fi ni te set, one tal ks about finite state systems;
otherwi se, infinite state systems. Also one tal ks about numeri cal or
symbol i c type systems inreference to whether X and Y are descri bed in
t er ms of number s or symbol s. I f both X and Y have mor e than one
component sets the systemi s multivariable.
Anot her cl assi fi cati on i s made inreference to the type of constructi ve
speci fi cati ons. Thus, one tal ks about di fferenti al (equati ons) or di fference
(equati ons) system, about static and dynamic systems or systems wi th
memor y or del ays i n reference to the properti es of the constructi ve
speci fi cati on whi ch descri bes the evol uti on of the systems response in
time.
There are a number of properti es whi ch are defi ned speci fi cal l y inthe
f r amewor k of a termi nal descri pti on of systeni s, e.g. causal i ty, stabi l i ty,
control l abi l i ty, reproduci bi l i ty, etc. Systems are al so cl assi fi edi nreference
to these properti es, e.g. as stabl e, control l abl e in a gi ven sense, etc. It
shoul d be menti oned that in conj uncti on with the termi nal descri pti on
Systems concepts 61
one shoul d not use systems properti es whi ch are perti nent to the goal -
seeki ng approach such as l earni ng, adaptati on, sel f-organi zati on, etc.
3. GOAL-SEEKING APPROACH
In the goal -seeki ng approach the response of a systemto any sti mul i is
descri bed inreference to the pursui ng of a goal or purpose. A si mpl est
way to formal i ze the noti on of a goal -seeki ng is to represent i t as a
deci si on- maki ng systemin a general . Some of the aspects of the goal -
seeki ng however will remai n then i nformal . To descri be a system
S c X x Y as deci si on- maki ng there shoul d be assumed t wo i tems:
- A f ami l y of deci si on probl ems, b = {A (x) : x E XI; apparentl y the
fami l y of deci si on probl ems, , is parametri zed by the i nput set X, i.e.
for each x E X there is associ ated a gi ven deci si on probl emA (x). Let M
denote a deci si on set for all A(x), i.e. a sol uti on for A(x) is an el ement of
My denoted by fi@).
- A mappi ng Q : M 3 Y whi ch generates an output correspondi ng to
any sol uti on of the systems deci si on probl em.
Systems behavi or i s then descri bed in reference to and M: for any
sti mul us, x E X, the system s response i s determi ned by (obtai ned by
transf ormati on f rom) a sol uti on fi@) of the deci si on probl emA(x), i.e.
Y = Q(fi(x)).
For i l l ustrati on consi der a si mpl e exampl e of a deci si on- maki ng
system. Let both X and Y be t i me functi ons defi ned on the i nterval [O,l].
Let there be gi ven another set of timefuncti ons M defi ned al so on [0,1].
For each x let there be gi ven an opti mi zati on probl em, denoted by A(x)
and defi ned by the statement:
Fi nd &(x) E M so that the defi ni te i ntegral
1
G(m,x,y) = J [O-X)~ + c m2] dt
O
i s mi ni mal under the constrai nt gi ven by di fferenti al equati on
- -
dy - ay + bm,
dt
where a, b and c are constrai nts.
62 Mihajlo D. Mesarovic
I t can easi l y be shown that for each x there is a uni que sol uti on d(x) and
therefore a uni que y(x) as gi ven by the constrai nt equati on. The deci si on
probl ems = (A(x) : x E X> and the constrai nt equati on defi ne therefore
a functi onal systemS : X 3 Y such that for any x E X, the pai r (XJ) is in
the systemS if and onl y if y is obtai ned by the constrai nt equati on f rom
a sol uti on of A(x).
Cl assi fi cati on of the deci si on- maki ng systems can be made accordi ng
to the type of deci si on- maki ng probl emi nvol ved; e.g. opti mi zati on,
stati sti cal deci si on systems etc.
I f X and Y are t i me functi ons defi ned on T and for any t E T the val ue of
the deci si on vari abl e fi(t), i .e. sol uti on at timet, is obtai ned by a mappi ng
on i nputs and outputs one has a feedback control system descri bed by a
pai r of mappi ngs
whose composi ti on i s the overal l systemS c X x Y. Constructi ve
speci fi cati ons of FI and F2 descri be actual evol uti on of the f eedback
systemi n ti me.
Other types of goal -seeki ng systems are recogni zed inreference to the
probl em. E.g. adaptati on and l earni ng i s defi ned in reference to the
uncertai nty i nvol ved i nthe deci si on process. Adaptation i s acti vi ty of a
goal -seeki ng systemai med at reduci ng the uncertai nty. I n mor e compl ex
si tuati ons el ements of A(x) i tsel f can change; e.g. the structure of the
functi ons and rel ati ons i nvol ved can change i n order to i mprove the
perf ormance of the system. This acti vi ty i s referred to as self-organization.
Many other systems properti es perti nent to goal -seeki ng (such as e.g.
l earni ng) can be defi ned inthe same f ramework.
4. LARGE-SCALE COMPLEX SYSTEMS
It is to an extent a rel ati ve questi on what one woul d consi der a l arge-scal e
system. As a rul e i t is not i nthe systemi tsel f but i nthe approach taken to
represent the systemwhether one shoul d consi der a systemas l arge-scal e.
For exampl e what is a l arge-scal e systemto a psychol ogi st is but a
component for a soci ol ogi st. It i s essenti al for the l arge-scal e system
approach that the compl exi ty or l argeness of the systembe properl y
Systems concepts 63
refl ected in the model , i.e. descri pti on of the systemas such. The fi rst
prerequi si te for thi s i s that the systemconsi st of subsystems and that this
parti ti oni ng be recogni zed expl i ci tl y. Thi s l eads toa mul ti -l evel or strati fi ed
descri pti on of a system. On any parti cul ar l evel one consi ders a fami l y of
subsystems as compl etel y i sol ated but on the next hi gher l evel these are
i ncorporated i nto l arger subsystems. The basi c questi on i s how the sub-
systems, as defi ned on the precedi ng l ower l evel , i nteract. Convi nci ng
arguments can be presented that any l arge-scal e systemrequi res a
hi erarchi cal , strati fi ed, descri pti on.
Another i mportant type of l arge-scal e systems requi res both the
concepts of hi erarchy and goal -seeki ng. Systems descri bed inthe precedi ng
secti on have a si ngl e goal even i f the systemwas mul ti -vari abl e. I n general ,
of course, a systemmay contai n subsystems whi ch have di fferent goal s
that can be parti al l y or compl etel y inconfl i ct. I n this respect the fol l owi ng
cl assi fi cati on of goal -seeki ng systems is of i nterest:
4.1Single-level single-goal systems. These are the systems descri bed in
the precedi ng secti on. Thei r behavi our can be very compl i cated, of
course, i nvol vi ng predi cti on, non- numeri cal deci si on- maki ng, l earni ng
etc., but there is at l east one si mpl i fyi ng feature: there i s no confl i ct
wi thi n the boundari es of the system. Control and deci si on- maki ng
theory as devel oped over the last t wo decades to a consi derabl e l evel of
sophi sti cati on i s concerned wi th this type of systems.
4.2 Single-level multi-goal. These are the systems whi ch contai n a fami l y
of i nteracti ng goal -seeki ng subsystems each with a goal of i ts own and
wi thout any supremacy rol e assi gned to a parti cul ar uni t. Fr omthe
conceptual standpoi nt the compl exi ty of behavi our of this type of system
i s consi derabl y greater than that of the systems inthe precedi ng category.
For this cl ass of systems t wo theori es have been devel oped: (1) game
theory todeal wi th the si tuati ons where the subsystems are incompeti ti on;
(2) theory of t eams to deal withthe si tuati on where the subsystems are
cooperati ng toward a common obj ecti ve. Al though these approaches have
created a conceptual f ramework i n whi ch to deal wi th these ki nds of
systems much as yet has to be l earned about how toresol ve a confl i ct ina
si ngl e-l evel mul ti -goal system.
4.3 Multi-level multi-goal systems. A system, in thi s cl ass, contai ns as
64 Mihajlo D. Mesarovic
before a fami l y of i nteracti ng subsystems each one of t hemhavi ng a
di sti nct goal possi bl y inconfl i ct withothers, but al so there is a hi erar-
chi cal , supremal i ty type rel ati on between the subsystems so that some of
the subsystems can i nfl uence - condi ti on, constrai nt or even control -
the goal -seeki ng acti vi ti es of others. Thi s of course i s an organi zati onal
type of systemas f ound inbi ol ogi cal , soci al si tuati ons as well as in the
compl ex technol ogi cal si tuati ons (automati on).
Al t hough the concept of hi erarchy was recogni zed as i mportant for
qui te some time, onl y recentl y a mathemati cal theory of mul ti -l evel
systems and wi thi n this f r amewor k a comprehensi ve theory of coordi na-
ti on, i.e. control rel ati onshi p between uni ts on di fferent l evel s, have been
devel oped (Mesarovi c, 1970). This type of systems is of' i mmense
i mportance i n many sci enti fi c areas as well as inengi neeri ng and manage-
ment. It has been stated that the future of bi ol ogy depends upon the
understandi ng of how to cross bi ol ogi cal l evel s, i.e. how the properti es
on one l evel are refl ected on adj acent l evel s. Si mi l arl y for a theory of
organi zati on, whi ch will deal expl i ci tl y withthe i nternal structure of an
organi zati on, a mul ti -l evel systems theory i s i ndi spensabl e. It is hard to
overemphasi ze the i mportance of mul ti -l evel , mul ti -goal systems. I ndeed,
i t has been argued qui te convi nci ngl y that the concept of hi erarchy i s
i nti matel y rel ated wi ththe very noti on of a l arge-scal e systemand that
any l arge-scal e systemhas hi erarchi cal features.
5. APPLICATIONS
Systems concepts pervade many a field. Li ke appl i ed mathemati cs,
systems concepts are used inal most any area where the devel opment of a
f ormal theory deal i ng wi th structural questi ons is attempted. Broadl y
speaki ng appl i cati ons of systems i deas can be grouped i nto three areas
accordi ng to the rol e they have in the met hodol ogy i n the respecti ve
fi el ds, namel y, sci ences, engi neeri ng and management .
5.1 Application in sciences. In non- physi cal sci ences, such as soci al and
pol i ti cal sci ences, psychol ogy, economi cs, etc. systems theory is provi di ng
a f r amewor k for any type of f ormal studi es whether purel y mathemati cal
or based on computer si mul ati on. Systems theory is i ncreasi ngl y bei ng
i denti fi ed withthe appl i cati on of quanti tati ve methods in general . Further-
Systems concepts 65
more, i t i s provi di ng a new set of metaphors and paradi gms for model i ng
and conceptual i zati on, In the past, concepts f r omthe physi cal sci ences
where predomi nant; one tal ked about pressures, forces, energy, etc. inthe
context of soci al , pol i ti cal and economi c si tuati ons. Newmetaphors f r om
the systems fi el ds i nvol ve concepts such as f eedback, i nf ormati on f l ow,
game-theoreti c rel ati onshi p, hi erarchi es, etc. These are openi ng com-
pl etel y new avenues whi ch can l ead to a dramati c i mprovement of our
understandi ng of the soci al and economi c systems especi al l y when
coupl ed wi th computer-assi sted anal ysi s.
I n the physi cal and bi ol ogi cal sci ences the systems concepts have al so
i ntroduced a new di mensi on. Thi s i s best i l l ustrated inbi ol ogy. Cl assi cal l y,
sci enti c expl anati on inbi ol ogy is gi ven ei ther int er ms of physi cal l aws
or chemi cal pri nci pl es. Syst ems concepts provi de a newf r amewor k for
sci enti fi c expl anati on of bi ol ogi cal probl ems; namel y, the bi ol ogi cal
phenornena are descri bed ini nf ormati on processi ng and deci si on- maki ng
t er ms e.g. as i nf ormati on channel s, f eedback control systems, mul ti -l evel
systems, etc. Resul ts have al ready been grati fyi ng i n parti cul ar in
physi ol ogy. However , i t is onl y fair tostate that many methodol ogi cal and
conceptual questi ons have yet to be resol ved. For exampl e, a systems
bi ol ogi cal expl anati on is still questi oned on occasi on as not bei ng qui te
as f undamental as a bi ophysi cal and bi ochemi cal expl anati on, as bei ng
somewhat temporary, merel y an expedi ent inthe absence of a fi nal under-
standi ng whi ch, by assumpti on, shoul d be in terms of physi cal l aws or
chemi cal pri nci pl es. This posi ti on is not attai nabl e. Sol i d arguments can
be adduced to showthat a systems bi ol ogi cal expl anati on i s on a f unda-
mental bi ol ogi cal l evel and is not reduci bl e to l ower l evel s, j ust as
chemi stry is not reduci bl e to physi cs i nany practi cal sense.
5.2 Engineering. Systems i deas are appl i ed in engi neeri ng in many
di fferent ways. First, in the synthesi s and desi gn of man- made systems
whi ch are to perf orm i nf ormati on-processi ng or deci si on- maki ng
functi ons inparti cul ar inthe desi gn and appl i cati on of computers. Second,
in the anal ysi s of l arge technol ogi cal compl exes whi ch i nvol ve sub-
systems of di verse nature, el ectri cal , mechani cal , etc. Thi rd, inthe eval ua-
ti on of the economi cs and soci al i mpact of l arge technol ogi cal systems.
Last, but certai nl y not l east i n soci otechnol ogi cal areas such as trans-
portati on, urban pl anni ng, pol l uti on control etc. These are man- made
systems of enormous soci al i mpl i cati on al though technol ogy- based.
66 Mihujlo D. Mesarovic
The functi oni ng of these systems is not descri bed on the l evel of physi cal
component s and the study has tobe conducted inthe systems f ramework.
I t is thi s area whi ch is at the hearth of what is termed systems engineering.
5.3 Management. Systems i deas are used i nmanagement i na number of
ways. First of all the appl i cati on of the so-cal l ed moder n management
techni ques, such as programmi ng, operati on research, etc. is done inthe
systems f ramework. Next , automati on of management functi ons such as
e.g. management (computer) i nf ormati on systems i s devel oped and used
wi thi n a systems vi ewof the management functi ons and the company as a
whol e. Fi nal l y, the systems model s and anal ysi s i n busi ness domai n,
e.g. of organi zati ons, marketi ng, economy, etc. provi de a basi s for a
si gni fi cant i mpr ovement of the management deci si on process.
A wor d on the so-cal l ed systems approach as used inengi neeri ng and
management i s inorder. I n the past many of the compl ex engi neeri ng and
management probl ems where not subj ect to deducti ve or quanti tati ve
anal ysi s because of the l ack of f r amewor k and tool s for anal ysi s. Systems
theory and comput er si mul ati on have provi ded these. I n the past one was
concerned onl y wi th some of the subsystems in sol ati on (those whi ch can
be studi ed by cl assi cal mathemati cal methods) the enti re systembei ng
hopel essl y compl i cated for a quanti tati ve anal ysi s. I n the new systems
approach one starts wi th the enti re systemand descri bes i ts operati on inas
realistic t er ms as possi bl e avoi di ng the pitfalls of usi ng too speci fi c
mathemati cal model (as e.g. di fferenti al equati on systems). Systems
concepts and comput ers al l owthe anal ysi s of such poorl y structured and
weakl y descri bed systems. The emphasi s is on a real i sti c representati on
of the systemini ts enti rety. The useful ness of such an approach hardl y
needs to be emphasi zed.
5.4 I mpl i cati ons of systems concepts in other areas is al so of i nterest.
Phi l osophi cal l y systems approach i s inthe tradi ti on of l ogi cal posi ti vi sm.
However , i t al so represents a maj or departure of the tradi ti onal approach
inthe sense that i t is much mor e of a relativistic character. There is no
commi t ment to a reducti oni st i dea but rather onl y tothe methodol ogy of
usi ng mathemati cal constructs to study some structural rel ati onshi p of
real -l i fe phenomena. Al so, i t provi des newapproach to many probl ems
(such as l earni ng, i nformati on, etc.) of central i nterest to vari ous
branches of phi l osophy, i n parti cul ar epi stemol ogy. Fi nal l y, i nthe era of
Systems concepts 61
i ncreased speci al i zati on, i nf ormati on expl osi on and exponenti al i ncrease
i n detai l ed factual knowl edge, the systems concepts offer one of the f ew
ways of putti ng the j i g-sawpuzzl e together, of comprehendi ng the
worl d around us as i t appears , recogni zi ng the i nterrel ati onshi ps between
what are tradi ti onal l y consi dered as i sol ated detai l s. Systems concepts
promi se to be of the utmost i mportance i n organi zi ng knowl edge.
Structural aspects of whatever phenomena one i s concerned withcan be
descri bed in t er ms of the ki nd of systems used in model l i ng. Systems
concepts therefore have a potenti al to provi de a f r amewor k for organi za-
ti on and cl assi fi cati on of facts i nmost di verse fi el ds. Thi s i s of parti cul ar
i mportance ineducati on.
6. THEORETICAL BACKGROUND
It i s not easy to gi ve a hi stori cal account of the concepts of systemand
the systems approach in general because the field has devel oped very
broadl y and l acks spectacul ar advances - breakthroughs - that can be
associ ated wi th si ngl e person or group. I n engi neeri ng i t was a natural
tendency to be concerned wi th l arger and l arger systems consi sti ng of
subsystems of di verse nature and of i ncreasi ng economi c and soci al
i mportance. Take for exampl e el ectri cal engi neeri ng. The fi rst step
toward greater general i ty was made by studyi ng a general rotati ng
el ectri c machi ne whi ch can represent ei ther generators or motors of
di fferent vari eti es. Next was to i ncl ude transformers and to deal witha
general el ectri cal or el ectromechani cal energy transf ormer. Wi t h the
advent of el ectroni c ci rcui try the concept of el ectri cal net work devel oped.
This was f ol l owed by the concept of a general i zed net work based onl y on
energy exchange regardl ess of the f or m(el ectri cal , mechani cal , chemi cal )
i n whi ch the exchange i s taki ng pl ace. Fi nal l y, the noti on of a system
whi ch deal s onl y wi th the exchange wi th the envi ronment whi ch i s defi ned
on an abstract, i nf ormati on type, basi s rather than int er ms of energy or
materi al exchange. Si mi l ar trends can be traced inother fi el ds. I ncreased
compl exi ty requi red greater general i ty i ndescri pti on and anal ysi s. On the
other hand i t was the advent of automati on and coni puters whi ch made
some of the systems i dea of normati ve as well as descri pti ve val ue si nce
one had the possi bi l i ty to materi al i ze di fferent i nf ormati on processi ng
and deci si on- maki ng processes. Thi s was all accel erated by soci otech-
68 Mhajlo D. Mesurovic
nol ogi cal devel opments of unprecedented scal e and compl exi ty (e.g. such
as space expl orati on, i ncreased urbani zati on, popul ati on expl osi on, etc.).
A sl i ghtl y better def i ned task is that of traci ng the attempts to provi de
a general systems theory. In thi s context the fol l owi ng maj or devel -
opment s shoul d be menti oned:
A) Proposal to l ook for some general l aws whi ch are val i di nsoci al and
bi ol ogi cal fi el ds (von Bertal anfy, 1968; Boul di ng, 1956) the devel opment
bei ng non- mathemati cal , i ntended as a scientific phi l osophy.
B) Proposal for the devel opment of a theory of cyberneti cs or steeri ng ,
control (Wi ener, 1948).
C) Proposal to devel op a mathemati cal theory of general systems
whi ch on the one hand will be essenti al l y rel ati vi sti c, i.e. i t will not make
any commi t ment as tothe exi stence of some general l aws (cf. A), whileon
the other hand i t will combi ne the control (deci si on-maki ng) and i nf orma-
ti on, si gnal -processi ng phenomena (Mesarovi c, 1964; 1968).
It i s hardl y necessary to say whi ch posi ti on is f avoured here. The fi el d,
however, i s sti l l not fully devel oped and onl y the future will showwhi ch
of the approaches is most useful . The i mmense i mportance of systems
concepts, however, has al ready been establ i shed.
REFERENCES
BERTALANPY, L. VON, (1950) An outline of general systems theory, The British Journal
BERTALANFY, L. VON, (1968) General System Theory, G. Braziller.
BOULDING, K. E., (1956) General systems theory-skeleton of science, General Systems
MSAROVIC, M. D., (1964) Foundations f or a general systems theory, in Views on
MESAROVIC, M. D., (1968) Auxiliary functions and constructive specification of
MESAROVIC, M. D., MACKO, D.;TAKAHARA, Y., (1970) Theory of Multilevel Systems,
WIENER, N. , (1948) Cybernetics, N e w York, John Wiley.
of the Philosophy of Science.
Yearbook, Ann Arbor (Mich.), University of Michigan.
General Systems Theory, N e w York, John Wiley.
general systems, Journal of Mathematical Systems Theory, Springer-Verlag.
N e w York, Academic Press.
4 Symmetry concepts and the fundamental theory of
matter
ABDUS SALAM
Symmetry, us narrw or us wide as you may de-
jne itsmeuning, is one ideu by whchmun through
the ages tried to comprhend and create order,
beauty and perfection ( Her mann Weyl ) 1
Fr omthe dawn of all ci vi l i zati on man has wondered and asked questi ons
- questi ons about the col our of the sunset, about the l umi nosi ty of the
stars, about rai nfal l and cl oud-burst, about the traj ectory of a bul l et and a
space satel l i te, and eventual l y, about lifeitself. But inall thi s questi oni ng
there has occurred one recurri ng theme. Man has al ways bel i eved that the
answers to these questi ons, when they come, must fol l owf r omj ust a very
fewgeneral pri nci pl es. Man has al ways hel d to an unreasoni ng fai th in an
eventual symmet ry, an eventual si mpl i ci ty, inany basi c l aws whi ch may
govern the uni verse. The hi story of sci ence is the hi story of a search for
such all embraci ng, such uni fyi ng concepts. And al most i nvari abl y the
successful concepts have been the ones di sti ngui shed by thei r aestheti c
appeal .
To i l l ustrate most rapi dl y what I mean and the manner in whi ch I
wi sh to devel op my t heme, consi der a basi c symmet r y whi ch we bel i eve
the uni verse in whi ch we l i ve in possesses - the transl ati on symmet r y
of space. Transl ati on symmet r y is one of the most fami l i ar types of
ornamental symmetri es - fami l i ar inart and archi tecture. When a pattern
repeats i tsel f after bei ng transl ated a finitedi stance - l i ke the recurrence
of the del i cate arches of the Pal ace of the Doges i nVeni ce - we speak of
transl ati onal symmet ry possessed by the structure. Assume now that
space i nwhi ch we l i ve inpossesses the symmet ry of such exact recurrence
- mor e preci sel y that the resul ts of an experi ment perf ormed here on
earth are i denti cal to the resul ts of the same experi ment perf ormed on
Mars; that the transl ati on of an experi mental Set-up f r omearth to
Mar s does not affect the resul ts of the experi ment; that the l aws of
physi cs are transl ati on symmetri c. Cl earl y if thi s was not the case, there
woul d be no reproduci bi l i ty of natural phenomena. The l aws of physi cs,
70 Abdus Sulum
if they di ffered f r oml ocati on to l ocati on - f r omearth to Mar s -
woul d be i nfi ni tel y mor e compl i cated to i nfer and sci ence, as weknow i t,
woul d not exi st. Now the astoundi ng thi ng about thi s postul ate of
transl ati on symmet r y of space is that i t can be tested in our own
l aboratori es on earth wi thout goi ng to any other part of the uni verse.
One can show- and I shal l di scuss thi s ingreater detai l l ater - that the
fami l i ar l awof conservati on of moment umi s a di rect consequence of this
postul ate. Thus to the preci si on that moment umis conserved in any
col l i si on, one may assert that space is transl ati on-symmetri c and vi ce
versa. If all space is l i kened to the Doges Pal ace and resul ts of a l ocal
experi ment l i kened to one of the arches, the symmet ry i mpl i ed by
recurrence of the arches makes not onl y for the beauty of the structure:
i t does somethi ng sti l l mor e prof ound. It gi ves us an expl anati on of a
f undament al phenomenon ( moment umconservati on) whi ch one has
observed so often happeni ng underneath each arch. I n physi cal sci ence,
therefore, symmet ry concepts possess not j ust an aestheti c appeal ; they
provi de prof ound correl ati ons of postul ate and experi ment.
Havi ng set the stage for our study of symmet ry pri nci pl es l et me
bri efl y summari ze the parti cul ar aspects I shal l concentrate on. I shal l
di scuss speci fi cal l y the fol l owi ng concepts:
1. Space- ti me symmet r y; the postul ate of transl ati onal and rotati onal
symmetri es of space and time.
2. Symmetri es of space and timerefl ecti on.
3. Symmetri es associ ated wi th the intrinsic properti es of matter -
properti es likeel ectri c charge, hypercharge and uni tary charge.
4. The rol e of approxi mate symmetri es whi ch combi ne space- ti me and
i ntri nsi c symmetri es menti oned above.
I shal l not di scuss the symmetri es one meet s i n the study of macro-
scopi c matter - symmetri es of crystal s for exampl e, and the ornamental
symmetri es inarchi tecture, scul pture and art - parti cul arl y ori ental art.
These symmetri es are aestheti cal l y i mportant but they are not rel ated to
f undament al physi cal l aws in the deep manner the other symmetri es I
menti oned are.
2. ROLE OF GROUP THEORY
Bef ore we di scuss space- ti me and i nternal symmetri es of matter, i t may
Symmetry concepts and the fundamental theory of matter 71
be menti oned that the di rect mathemati cal construct for descri bi ng a
symmet ry i s the group concept. A set of obj ects (the arches inthe Doges
Pal ace for exampl e) possess a group property if the symmet r y operati ons
(likethe transl ati on operati on of one arch on tothe next) when perf ormed
on a gi ven member of the set i nterchanges i t i nto another member of the
set. Groups are of t wo vari eti es, dependi ng on the type of group
operati on . They are ei ther discrete when the number of group operati ons
i s finite(e.g. a crystal l ographi c group), or continuous - l i ke the group of
rotati ons whi ch, for exampl e, transf ormone poi nt on the ci rcumf erence
of a ci rcl e on to a nei ghbouri ng poi nt.
Gr oup theory i s one of the best devel oped di sci pl i nes i nmathemati cs.
It came to fl ower towards the end of the last century - parti cul arl y the
theory of conti nuous groups (inthe wor k of Sophus Lie). I ts systemati c
use i nphysi cs i s, however, of recent ori gi n, and dates back no further
than the begi nni ngs of quant umtheory. To see why thi s has been so,
consi der the postul ated symmet ry of space we di scussed before. I sai d
earlier that the postul ate of transl ati on symmet ry of space l eads to the
l awof conservati on of moment um. The argument for thi s i s classical;
i t basi cal l y depends on how we defi ne moment um. Consi der now the
rol e of quant umtheory i nfurther sharpeni ng thi s resul t. Quant umtheory
brought group theory (and, withi t, the powerf ul mathemati cs i nvented
by group theori sts) to bear on the descri pti on of physi cal phenomena.
Why di d thi s happen? The reason has been spel t out by Yang and Wi gner;
i t l i es inone basi c ci rcumstance; i t l i es in the basi c postul ate of quant um
theory that the quant umstates of a physi cal systemf or ma linear
manifold.
To i l l ustrate, consi der space rotati ons i nthe three- di mensi onal space
represented mathemati cal l y by the rotati on group O(3). Throughout the
hi story of physi cs we have started wi th the assumpti on that l aws of
physi cs remai n unchanged for space rotati ons. I n the fi nal anal ysi s i t is
an empi ri cal postul ate, to be tested by i ts consequences. We bui l d thi s
postul ate i nto physi cs by demandi ng that all basi c equati ons of mot i on
whi ch represent f undamental physi cal l aws shoul d not change i n f or m
when wri tten i n t er ms of co-ordi nates rotated in rel ati on to those we
started f rom. I n physi cal t er ms thi s means we are asserti ng that if an
experi mental set-up and the detecti ng devi ces are rotated through a
certai n angl e, the resul ts of the experi ment are unal tered. One can show
by a cl assi cal argument that thi s postul ate l eads to the conservati on l aw
72 Abdus Salam
of angul ar moment um. An appl i cati on of thi s rotati on symmet r y to
classical traj ectori es - pl anetary orbi ts for exampl e - al so tel l s us that,
gi ven a certai n orbi t, we may i nfer by rotati on the exi stence of other
physi cal l y possi bl e ones.
Now this resul t incl assi cal physi cs is i mportant but i t i s by no means
very deep. It does not l ead to newi nsi ghts. Contrast this withthe case of
quant ummechani cs. The same statement can be made about quant um
orbi ts. I n quant umtheory there i s, however, the further postul ate that
all possi bl e orbi ts f orma l i near mani f ol d and that one can sel ect f r om
thi s mani f ol d a l i nearl y i ndependent complete set int er ms of whi ch all the
orbi ts can be expressed l i nearl y, i.e.
Now denote the rotati on operator correspondi ng to a rotati on g by
U(g). Fr omrotati on i nvari ance, if [ Y> is an orbi t I Y' >= U(g) I Y>
i s another possi bl e one. Speci al i ze to I Y > = I Yi>. Fr omcompl eteness,
we may i nfer that
Uk) I Y+ = 2 a~ I Y5>
5
Cl earl y the gi ve us at once a representation - inthe techni cal sense of
matri ces representi ng rotati ons - of the group g. Wi t h the quant um
postul ate of the orbi ts f ormi ng a l i near mani f ol d, we i mmedi atel y stri ke
a l evel or ri chness with the mathemati cal representati on theory of
groups, unsuspected, unconcei ved of at the l evel of cl assi cal dynami cs.
Let us pursue this further. I n quant umtheory we are concerned onl y
with uni tary representati ons. Thi s is connected with the quant um
mechani cal theory of measurement . I shal l not go i nto the measurement
theory of physi cs inany detai l ; we shal l here merel y accept that we shal l
al ways deal wi thuni tary representati ons. Wr i t e an i nfi ni tesi mal uni tary
rotati on operator inthe form
U(g) 1 + i 5 J5 = 1, 2, 3)
The standard commut at i on rel ati ons for the real operators J8 whi ch
represent the three component s of the angul ar moment umvector read:
Her e
[ J d j ] = i E$3k Jk
E$/ = l(--l);
Symmetry concepts und the fundamental theory of matter 73
_P.
(Photo Alexis N. Vorontzoff)
The delicate arches of the Doses Palace in Venice exhibit translation symmetry.
74 Abdus Salam
if i, j, k = 1, 2, 3 or any cycl i c (anti -cycl i c) permutati on of these number s
and zero otherwi se. These commut at i on rel ati ons express the wel l - known
statement that a rotati on by an angl e 0 around an axi s A fol l owed by a
rotati on through an angl e cp around an axi s B does not gi ve t he same
fi nal confi gurati on of the systemas these t wo rotati ons perf ormed i nthe
reverse order. Now i t is well known that the operator for total angul ar
moment umJz = J12 + Jz2 + J3z commut es wi th all the three Ji.
[Jz,&] = O. Fr omthe mathemati cal theory of the group 0(3), i t is also
well known that the group O(3) possesses matri x representati ons,
l abel l ed by t wo di screte nuni bers j and j3, symbol i cal l y
Jz lj,j3> =Ki + 1) lj,j3>
J3 I j,j3> = j3 I jJ3>
where j can take i ntegral or hal f-i ntegral val ues; O, 3, 1, 3/ 2 . . . and j3 ranges
bet ween +j and -j. The symmet ry postul ate for space rotati ons com-
bi ned withthe compl eteness of the basi c states of a quat ummechani cal
systemal l ows us therefore to state that all systems i nnature must be a
super-posi ti on of di screte states, with wel l -defi ned val ues of angul ar
moment um. The quant umnumber j representi ng as i t does the total
angul ar moment um(or the i ntri nsi c spi n) of a basi c state as well asj 3 the
component of the spi n al ong the z-axi s, are both quantized; both are
integers or half-integers (in uni ts of the Pl anck s constant). We all know
that quanti zati on, the di screteness of physi cal quanti ti es is the essence of
quant ummechani cs. It is the expl oi tati on of the representati on theory of
groups of the symmet ry group of three di mensi onal rotati ons whi ch has
automati cal l y guaranteed for us that angul ar moment umshal l al ways be
quanti zed. There coul d concei vabl y be no mor e beauti ful synthesi s of
symmetri es, group theory, quant umtheory and experi ment.
Summari zi ng, rotati on symmet ry was the basi c postul ate; conservati on
of angul ar moment um, i ts classical consequence; quanti zati on and
di screteness of angul ar moment um, i ts quantum-mechanical consequence.
The whol e is one mi racul ous bl end.
3. ROTATION SYMMETRI ES OF TIME A N D SPACE
We have so far consi dered transl ati on and rotati on symmetri es of the
three-di mensi onal space we i nhabi t. The great revol uti on of 1905 for
Symmetry concepts and the fundamental theory of matter 75
whi ch Ei nstei n was responsi bl e, consi sted of the speci al rel ati vi ty theory
postul ate whi ch stated that there is i nnature a compl ete symmet ry between
space and ti me so far as transl ati ons and rotati ons are concerned. Mor e
speci fi cal l y, the Ei nstei n postul ate was that the resul ts of an experi ment
are not onl y space-transl ati on and space-rotati on i nvari ant: they are
ti me-transl ati on and space- ti me (Lorentz) rotati on i nvari ant. That l aws
of physi cs do not change f r omyesterday to today, nor do they change
when studi ed i n a ref erence-f rame space-ti me-rotated (more si mpl y,
movi ng with a uni f ormvel oci ty) rel ati ve to a f rame whi ch i s fi xed. The
di rect cl assi cal consequences of space- ti me rotati on symmet r y are l egi on -
these, as is wel l known, i ncl ude equi val ence of mass and energy (i.e. the
rel ati on E = mc2) as well as the time di l ati on of uni f orml y movi ng
obj ects (an astronaut l i ves l onger, the faster he moves, provi ded his
l i feti me reckoni ng i s done, usi ng a stati onary clock). We shal l not be
concerned here with this (classical) aspect of these symmetri es; rather
our concern will be with the quant ummechani cal aspects. These wer e
fi rst di scovered by Di rac and then el aborated group-theoreti cal l y by
Wi gner i n a cl assi c paper publ i shed in 1939. What Di rac and Wi gner
showed was the fol l owi ng.
3. 1 As a consequence of the space- ti me rotati on symmet r y, all mol ecul es,
all atoms, all nucl ear and subnucl ear parti cl es carry an i ntri nsi c angul ar
moment um. The parti cl es are not j ust chunks of matter; they resembl e
spi nni ng tops, spi nni ng cl ockwi se or anti cl ockwi se rel ati ve to thei r
di recti on of moti on.
3.2 The magni t ude of this spi n is measured i ni nteger or hal f-i nteger uni ts
of the Pl anck s constant. Each massi ve parti cl e wi th spi n J can exi st in
(2J + 1) states correspondi ng to the (2J + 1) di fferent pol ari zati ons of
thi s spi n. For exampl e, a beamof el ectrons (or protons, or neutrons),
each one of whi ch i s endowed witha spi n of J = -2, is composed of hal f
the number of parti cl es spi nni ng anti -cl ockwi se (mathemati cal l y, wi th
J3, the component of spi n al ong the di recti on of mot i on equal to +&) and
hal f-spi nni ng cl ockwi se withJ3 = -3. For a parti cl e of non- zero mass
and spi n J = 1 there are three pol ari zati on di recti ons J3 = +1, J3 = O,
J3 = -1 and so on.
3. 3 For parti cl es with zero rest-mass (and whi ch on account of this
16 Abdus Salam
property al ways travel with t he vel oci ty of light) there are j ust t wo
pol ari zati ons whatever be the total spi n val ue. For a spi n J and zero rest-
mass parti cl e, the magni tude of l eft-spi n is J3 = fJ, and of ri ght spi n
J3 = -J, withno other pol ari zati ons avai l abl e. For massi ve parti cl es the
(2J + 1) di sti nct spi n-pol ari zati on states menti oned above are equi val ent
to each other inthe sense that they can be transf ormed i nto one another by
space rotati ons. Thi s is not possi bl e for the case of parti cl es of zero mass.
Her e the t wo states of spi n-pol ari sati on +J and -J are compl etel y
i nequi val ent.
3.4 I n addi ti on to transl ati on and rotati on symmetri es, one may postul ate
other symmetri es. The ones I shal l consi der most speci fi cal l y are the
di screte symmetri es of space and time rejlection.
Space-refl ecti on i s somethi ng very fami l i ar inour everyday experi ence.
If we l ook i nto a mi rror, we see the space-refl ected worl d whi ch, as we
well know, is not i denti cal with our worl d. I n a mi rror, a ri ght hand
refl ects i nto a left hand; cl ockwi se moti ons go i nto anti -cl ockwi se moti ons.
To assert that l aws of physi cs are symmetri c for space refl ecti on is a
t remendous asserti on. It is the asserti on that if ri ght hands exi st, some-
where i n the Uni verse there must al so be l eft hands. For el ementary
parti cl es thi s is the asserti on that if ri ght-pol ari zed, ri ght-spi nni ng
el ectrons exi st, so must l eft-spi nni ng, l eft-pol ari zed el ectrons.
Ti me-refl ecti on i s somethi ng we are l ess fami l i ar wi th. We refl ect time
i f we run a filmbackwards. I n a back- run film a di ver can be seen ri si ng
f romthe water and settl i ngon to the di vi ng board - a phenomenon one
does not frequentl y come across inreal life. Ti me refl ecti on symmet ry
woul d assert that di vi ng and counter-di vi ng are bot h possi bl e physi cal
phenomena.
Now in 1928, Di rac postul ated (not exactl y in the manner I am des-
cri bi ng) that l aws of physi cs, even t hough this may not be the most
obvi ous of alI thi ngs, are i nfact symmetri c if we refl ect space and time
si mul taneousl y - that i s, we l ook i nto a mi rror and at the same time
i nterchange the past and the future. Thi s postul ated space- ti me refl ecti on
symmet ry, together with the space- ti me rotati on symmet r y I spoke of
earl i er, had an amazi ng consequence. Di rac was abl e to showthat all
parti cl es innature - al l atoms, all mol ecul es, all nucl ear parti cl es, and all
obj ects made f r omthem- exist inpai rs, To every parti cl e there corres-
ponds an anti -parti cl e. An anti -parti cl e is an obj ect wi th the same spi n,
Symmetry concepts and the fundamental theory of matter 17
same mass but opposi te electric or nucl ear charge. I f Di rac was ri ght the
exi stence of the negati vel y charged el ectron woul d i mpl y the possi bl e
exi stence of one posi ti vel y charged (the so-cal l ed positron). I f the proton
exi sts, so must an anti -proton. If the hydrogen at omexi sts, thi s must
i mpl y the possi bl e exi stence of an anti - atomof anti - hydrogen wi th the
same energy l evel s. Thi s wor k of Di rac i s some of the most moment ous,
the most far-reachi ng inthe hi story of physi cs. It was, of course, brilliantly
conf i rmed by the experi mental di scoveri es of anti -el ectrons and anti -
protons. At one stroke Di rac had doubl ed the number of possi bl e struc-
tures in the uni verse - and all f roma pure mathemati cal deducti on
fol l owi ng upon a set of postul ated symmetri es.
3. 5 Let us now separatel y consi der the space-refl ecti on symmet ry. The
human f rame is space-refl ecti on symmet ri c so far as our external
appearances are concerned; we possess both ri ght and l eft hands. Not
so, when we consi der i nternal organs. Most of us have hearts on the
left onl y. The weaker f ormof mi rror refl ecti on symmet ry pri nci pl e
however merel y asserts that there shoul d be the possi bi l i ty of the
exi stence of some ri ght-hearted i ndi vi dual ; the fact that they are rare may
si mpl ybe a consequence of the conti ngent geneti c factors - a resul t of asym-
metri c boundary condi ti ons whi ch happened to have perpetuated. Thi s, as
we know, is i ndeed the case. There do exi st i ndi vi dual s wi th ri ght hearts,
t hough they are rare. Refl ecti on symmet ry of space seems to hol d innature.
Thi s at l east was the opi ni on hel d unti l 1957, f romthe ti me of Lei bni z
who was the fi rst to refl ect on this symmet ry. We bel i eved that so far as
f undamental physi cal l awwas concerned, there was no i nner di fference
between the ri ght and the left. Mathemati cal l y, one may say that
equati ons of mot i on of physi cs whi ch descri be the uni verse remai n
unchanged when the co-ordi nate x is repl aced by --x.
In the summer of 1956, T. D. Lee and C.N. Yang, t wo Chi nese physi -
cists worki ng in the Uni ted States, by a careful anal ysi s of exi sti ng
experi mental data, came to the concl usi on that whereas the pri nci pl e had
been checked inthe i nteracti ons of protons, neutrons, el ectrons, i t had
never been checked where neutri nos were concerned. They suggested
that di rect checks be made to fi ndout if the pri nci pl e was all that uni versal .
I heard Professor C. N. Yang in Sept ember 1956 questi on the space-
refl ecti on symmet ry postul ate in a l ecture in Seattl e. It seemed sacri l e-
gious to me - as i ndeed i t appeared tomost other physi ci sts - that nature
78 Abdus Salam
shoul d gi ve up a symmet r y pri nci pl e l i ghtl y. That ni ght a Mi l i tary Air
Transport pl ane was very generousl y pl aced at the di sposal of the
conferees to the Seattl e Conf erence by the Uni ted States Air Force to fly
back to London. The pl ane was too noi sy, too uncomf ortabl e to sl eep.
Over the dark Atl anti c I kept bei ng tormented by what Professor Yang
had sai d. I felt that if nature must sacri fi ce the space-refl ecti on
symmet r y pri nci pl e i t woul d do so onl y if thi s pri nci pl e confl i cted wi th
some symmet r y pri nci pl e aestheti cal l y even mor e appeal i ng. One knew at
that timethat the neutri no s rest-mass was very smal l ; i t struck me that if
one assumed that thi s mass was exactl y zero - i.e. that the neutri no
travel l ed al ways wi th exactly the l i ght-vel oci ty - one coul d show i mme-
di atel y and wi thout any el aborate cal cul ati ons - that this si tuati on woul d
confl i ct wi th mi rror- symmetry for neutri nos. Ri ght-spi nni ng neutri nos
woul d exi st but l eft-spi nni ng neutri nos woul d not. Li ke Hof f man in
Of f enbach s opera, when a ri ght-spi nni ng neutri no l ooked i nto a mi rror, i t
woul d see nothi ng. Coul d nature be sacri fi ci ng the mi rror symmet r y
pri nci pl e for another symmet r y ( nowcal l ed the y5- symmetry) whi ch must
govern the behavi our of all spi n .?i parti cl es whenever they possess zero
rest- mass and travel l i ght vel oci ty.
I n earl y 1957, the experi ments of Wu and Leder man in the Uni ted
States compl etel y conf i rmed the i deas stated above. Nature does not
possess mi rror- symmetry where neutri nos are concerned; ri ght-spi nni ng
neutri nos exi st but l eft-neutri nos do not. I n 1957 thi s symmet ry appeared
a weak substi tute for the l oss of mi rror symmet ry. I n 1969 when 75-
symmet r y has been seen to pervade all parts of physi cs, nucl ear and
el ectromagneti c as well, we know that the l oss has been f avourabl y
wei ghted inthe bal ance of gai ns. Thi s i s a devel opment, however, whi ch I
cannot write about in detai l , for this new ys- symmet ry is di ffi cul t to
expl ai n except inmathemati cal t erms.
I have wri tten of this epi sode of physi cs because i t i l l ustrates the
ide$xe whi ch most of us share and wi thwhi ch I started this essay. We
have al ways f ound that whenever a postul ated symmet ry pri nci pl e was
appeari ng to fail in natural phenomena, thi s must be due to some still
deeper symmet ry, wi thwhi ch i t must be inconfl i ct. We may, at a gi ven
ti me, fail to comprehend the aestheti cs of nature. When, however, the
full and fi nal pi cture emerges one has i nvari abl y f ound that the symme-
tri es thi s exhi bi ts are prof ounder still.
Symmetry concepts and the fundamental theory of matter 79
4. SYMMETRIES OF THE INTERNAL STRUCTURE OF PARTICLES
So far I have di scussed the space- ti me symmetri es, the quanti zati on of
spi n usi ng representati ons of the space group O(3) and the parti cl e anti -
symmet ry whi ch ari ses when one thi nks of space- ti me refl ecti on symmet r y
as a f undamental postul ate. I n this secti on we turn symmetri es of i nternal
structure of parti cl es and thei r cl assi fi cati onschemes on thi s basi s. The
gui di ng pri nci pl e once agai n will be the use of group theory. To i l l ustrate
this I shal l go bri efl y over the experi mental si tuati on i n parti cl e physi cs as
i t has devel oped f r om1926 onwards.
Ar ound 1926, t wo so-cal l ed f undamental parti cl es wer e known to
physi cs; the proton and the el ectron. These are ti ny chunks of matter -
the proton wi th a mass of around 10- 24 grams, the el ectron some 2000
t i mes l i ghter. Bot h are el ectri cal l y charged, the proton posi ti vel y and the
el ectron negati vel y. They were el ementary and f undament al i nthe sense
that all matter - all the 92 at oms - wer e then (erroneousl y) bel i eved to be
made f r omj ust t hese t wo obj ects.
As I sai d before, protons and el ectrons are not j ust si mpl e chunks of
matter. Bot h these parti cl es carry i ntri nsi c spi n, group theoreti cal l y the
parti cl es corresponded to the spi nor representati on of the group 0(3),
wi thj = $,j3 = 4, -4. I n the vocabul ary I have used earl i er the proton
(or the el ectron) f ormed a twof ol d (2j + 1 = 2) mul ti pl et.
The astoni shi ng thi ng about these t wo parti cl es was - and sti l l is -
the numeri cal equal i ty of the el ectri c charge they carri ed. The masses of
the el ectron and proton are so di fferent - the rati o is 1 : 2000. Thei r
el ectri c charge, however, is equal . Further, likeangul ar - moment um, all
el ectri c charges innature appear to be quanti zed - quanti zed ini ntegral
mul ti pl es of the charge on the el ectron. These facts we must i ncorporate
inour descri pti on of nature, ei ther by the statement that every systemin
nature i s made of protons and el ectrons or by a descri pti on mor e
embraci ng whi ch may conti nue to hol d, even if protons and el ectrons are
no l onger bel i eved to be el ementary.
Wi th the success of the symmet r y and group theory i deas inunder-
standi ng quanti zati on of spi n, the pattern for this mor e embraci ng
descri pti on seemed cl ear i n t er ms of a group-theoreti c i dea. The group
representati ons of a rotati on group in two- di mensi ons are l abel l ed by
posi ti ve or negati ve i ntegers. Assume that there exi sts a two- di mensi onal
i nternal ) space representi ng the i nternal degrees of f reedomof structure
80 Abdus Sulum
of matter. Cal l i t the charge space , and assume that equati ons of mot i on
of physi cs are i nvari ant for rotati ons in thi s space. Thi s rotati on sym-
metry will i mpl y (through the quant ummechani cal procedure sketched
earl i er) both charge quanti zati on and charge- conservati on. The pattern is
the same as that for the three- di mensi onal rotati on group; the l ogi cal
argument is the same. The di fference, however, is that, incontrast to the
three- di mensi onal physi cal space, the new twi n di mensi ons of charge
space apparentl y cannot be di rectl y apprehended. They are apprehended
onl y through thei r mani festati on - the quanti zed el ectri c charge.
Unti l 1930, the onl y known i nternal characteri sti c of an el ementary
parti cl e was j ust this one quanti zed enti ty - the el ectri c charge. In 1930,
wi th Chadwi ck s di scovery of the neutron, there came a break. The
neutron was the thi rd el ementary parti cl e; i t was al most as massi ve as
the proton but was el ectri cal l y neutral . The proton and the el ectron
attracted each other when cl ose together through the i ntermedi acy of the
cl assi cal el ectrostati c force but thi s force was cl earl y i rrel evant for
neutrons si nce they were neutral . Two neutrons, or a proton and neutron,
however di d exhi bi t a strong attracti on when cl ose to each other. Thi s was
a new force of nature. At comparabl e di stances one f ound empi ri cal l y
that i t was at l east 100 ti mes stronger than the el ectri cal force. Thus to an
excel l ent approxi mati on - to the approxi mati on that one coul d negl ect
el ectri cal forces rel ati ve to the nucl ear - protons and neutrons were t wo
states of j ust one si ngl e enti ty, the so-cal l ed nucl eon.
4.1 Isotopic symmetry
This si tuati on of a si ngl e enti ty (nucl eon) exi sti ng i n t wo di sti nct states
(proton and neutron) was a si tuati on one had encountered before. One
had seen that a spi n Q parti cl e possesses t wo pol ari zati on states I j = 4,
j3 = + Q> and I j = $, j3 = -&>. Coul d one once agai n postul ate the
exi stence of a new i nternal three- di mensi onal space - wi th three i nfi ni tesi -
mal rotati on generators Il, 12 13, sati sfyi ng the commut at i on rel ati ons
[k If] = i EZI~ Ik
The I = 3 reprsentati on of thi s new group coul d then be i denti fi edwith
the nucl eon with13 = 3 representi ng the proton state and 13 = - 3 the
neutron. Thi s suggesti on ori gi nated withKemmer , Hei senberg and Breit
around 1934-1938. The new i nternal space was named the i sotopi c
Symmetry concepts und the fundamental theory of mutter 81
space ; the nucl ei , whi ch are composi tes of nucl eons, f ormed mul ti pl ets,
correspondi ng to the i rreduci bl e representati ons of this i sotopi c-rotati on
group. All nucl ei carri ed i sotopi c spi n, i n addi ti on, of course, to the
ordi nary spi n whi ch henceforth I shal l cal l Poincar spi n.
The next devel opment in parti cl e physi cs came in 1935 wi th some
specul ati ons of Yukawa. Yukawa recal l ed that all accel erati ng el ectri c
charges emi t el ectromagneti c radi ati on i n accordance wi th Maxwel l s
l aws. The quant umaspects of the el ectromagneti c force are the photons.
Yukawa rai sed the questi on: what is the anal ogue of a phot on for the
nucl ear force? What type of radi ati on do nucl eons emi t when they are
accel erated? He conj ectured that there exi st in nature photon- l i ke
obj ects, the so-cal l ed mesons, parti cl es wi th masses i ntermedi ate between
el ectrons and nucl eons, whi ch are emi tted by accel erati ng nucl eons.
Fr omthe group-theoreti c poi nt of vi ew, these parti cl es, if they di d exi st,
woul d once agai n correspond to i rreduci bl e representati ons of the i so-
topi c group. Further, if likephotons these are emi tted- shed by nucl eons -
si ngl y at a time, conservati on of i sotopi c spi n woul d demand that thei r
I -spi n be an i nteger and not hal f-i nteger.
Yukawa s i deas wer e put f orward in 1935. These were persuasi ve i deas.
The search for these parti cl es was i nterrupted by the war, but j ust after
(in 1947) Professor E. C. Powel l announced the di scovery i ncosmi c rays
of the Yukawa parti cl es - the so-cal l ed pi ons . These wer e three pi ons,
correspondi ng to an i sotopi c spi n I = 1.
n++I I =1, 13=1>
7t O+I I =1, 13=0>
n- +p=1, 13=1>
The Poi ncar spi n J of the parti cl es turned out to equal zero (J = O).
Summari zi ng, the cl assi fi cati on scheme of parti cl es concerned wi th the
nucl ear force proceeds i nt er ms of three types of symmetries:
(1) The external space- ti me symmet ry, gi vi ng ri se to Poi ncard spi n J.
(2) Two i nternal rotati on symmetri es gi vi ng ri se to
(a) i sotopi c spi n,
(b) el ectri c charge Q.
4. 2 The SU(3) symmetry
Af ter 1947 came further experi mental di scoveri es. A whol e host of new
82 Abdars Sulain
obj ects wer e di scovered; parti cl es wi th di fferent masses, di fferent charges,
di fferent Poi ncar and di fferent i sotopi c spi ns. By no stretch of the
i magi nati on coul d one call these el ementary parti cl es any more. But
whether these were or wer e not composi tes of any si mpl er enti ti es one had
to fi nd a quant umdescri pti on for these. I n J anuary 1964 the si tuati on
coul d be summari zed as fol l ows.
As a resul t of pati ent and pai nstaki ng experi mentati on, both wi th
cosmi c rays and the gi ant accel erators at CERN, Brookhaven, Dubna,
Berkel ey and el sewhere, one coul d cl assi fy the newl y- di scovered parti cl es
i nto the fol l owi ng mul ti pl ets:
(1) 8 parti cl es of Poi ncar spi n 3 (the nucl eon fami l y).
(2) 9 parti cl es of Poi ncar spi n 4 (exci ted nucl eons).
(3) 8 mesons of Poi ncar spi n zero.
(4) 9 exci ted mesons of Poi ncar spi n one.
It is i rrel evant for my purposes to go i nto detai l s concerni ng these
mul ti pl ets; in parti cul ar the i denti fyi ng nomencl ature, etc., wi thi n a
mul ti pl et i s compl etel y i rrel evant. However , there i s one common poi nt
I need to i l l ustrate about these mul ti pl ets, and for this I shal l ref er to the
8-fol d nucl eon mul ti pl et whi ch consi sted of the fol l owi ng component s :
Table 1
I I3 Q Y = 2 (Q - 13)
3 1 1
- 4 O 1
A0 O O O O
1 1 O
O O O
-1 -1 O
: : I
] 1
z-
1 %
3 O -1
E- I a -1 -1
Thi s &fol d of parti cl es - all of nearl y equal mass, all of same Poi ncar
spi n - consi sts of four di sti nct i sotopi c mul ti pl ets (p,n), (A), (Z+XOZ-) and
(Eo, E-) wi th thei r I and I3 val ues gi ven inthe accompanyi ng tabl e. For
each i sotopi c mul ti pl et the quant umnumber Y (=2 h Q-13) (the so-cal l ed
hypercharge) happens to possess the same (i ntegral ) ei genval ue. (See the
Symmetry concepts und the fundamental theory of matter
83
last col umn of Tabl e 1. The quant umnumber Y takes the val ues 1, O, O,
-1.) Fi xi ng on 13 and Y, cl earl y there must exi st a hi gher symmet r y
group, perhaps a group of rank 2 si nce we are deal i ng wi th at l east t wo
si rnul taneousl y di agonal i zabl e operators 13 and Y, accordi ng to the
representati ons of whi ch one mi ght cl assi fy these parti cl es. That there
was some hi gher symmet ry at wor k i n the physi cs of the nucl ear i nter-
acti on was cl ear qui te earl y, around 1956-1957. That the way toprogress
l ay al ong a systemati c search for a Liegroup of rank 2 was onl y very
i mperfectl y understood unti l 1961. The maj or uncertai nty inthe approach
was, of course, al ways experi mental . I have bl andl y stated that the
nucl eon mul ti pl et was f ound to consi st of ei ght member s, all of Poi ncar
spi n &, and I have speci fi ed the I -spi ns and hypercharge Y for each parti cl e
as if every experi mental l y di scovered obj ect carri ed a chai n around i t
wi tha l abel on whi ch one coul d read off i ts characteri sti cs. I n real life
thi ngs are never like thi s. Unti l this day, for exampl e, one does not
experi mental l y know wi th mor e than 95 per cent statistical conf i dence
that the Poi ncar6 spi n of the E30 and E- parti cl es is i ndeed 3. My col l eague,
Professor P. T. Mat t hews, has gi ven an i l l umi nati ng i l l ustrati on of the
difficulties whi ch beset the wor k of an experi mental physi ci st. There
i s onl y one experi ment a parti cl e physi ci st can perf orm; he can
scatter one set of parti cl es off another and, by counti ng the number s
whi ch fly off i n a col l i si on in a gi ven di recti on, try to fi nd the spi ns
and i sotopi c spi ns, etc. of the di fferent end- products. I t s like pl ayi ng
a hose of water on a statue i n a dark r oomand bei ng al l owed to
col l ect the water that spl ashes off f romthe statues face. One can appre-
ci ate the hardshi ps of the experi mental physi ci st if one wer e posed the
probl emof del i neati ng the statues features by measuri ng onl y the
quanti ti es of water that spl ashed off per square i nch of i ts surface.
Returni ng to the cl assi fi cati on probl em, Sophus Lie had al ready
tabul ated all groups of rank t wo. There are just f our of these wi th the
techni cal nomencl ature: A2, B2, C2 and G2. One had to choose one of
these as representi ng the symmet ry. The probl emof fi ndi ng the correct
group and therefore the correct symmet ry coul d be f ormul ated thus.
Associ ati ng I -spi n wi th the group structure U2 and associ ati ng hyper-
charge Y wi th the group structure U1, fi nd a group of l ank t wo whi ch
possesses both U2 and U1 as subgroups, and has an 8- di mensi onal
representati on wi th component s as i ndi cated i nTabl e 1.
Once agai n l et me stress, no worki ng physi ci st can ever state a physi cal
84 Abdus Salam
probl eminthi s f or m, except after the event. There are so many reserva-
ti ons, so many hesi tati ons, so much one must take on trust and, most
difficult of all, so much one must di scount.
Stated i nthe f or mI have used above, the probl emhad one sol uti on -
the hi gher symmet r y group must be AZ, whi ch i s al so known under the
name SUS. Thi s sol uti on was fi rst proposed by the J apanese physi ci sts
Ohnuki , Ogawa and Sawada in 1959. The SU3 symmet r y was named
' uni tary symmetry' . Unf ortunatel y the J apanese authors made a wrong
physi cal i denti fi cati on of parti cl es; the representati ons of the group they
happened tochoose di d not decompose inthe manner of Tabl e 1. A l ater
versi on suggested i n 1961 by Mur r ay Gel l ' Mann a\ nd Yuval Neeman
fittedfacts better.
Thi s was encouragi ng, but not encouragi ng enough. I n addi ti on to the
8-fol d mul ti pl et of nucl eons, there was al so the other mul ti pl et consi sti ng
of ni ne exci ted nucl eons wi th the fol l owi ng assi gnments:
Table 2
Z Y
; : i I 1 O
Y*-
=*O
' =*- } 3 -1
Y
Now there i s no 9- component i rreduci bl e representati ons of the group
U(3); the nearest whose ( Uz x UI) decomposi ti on woul d yi el dthe ei gen-
val ues shown in Tabl e 2 for the generators 13 and Y must contai n ten
obj ects. One cruci al parti cl e was mi ssi ng; a parti cl e whi ch f romthe
group-theoreti c structure of the representati on concerned must possess
i sotopi c and hypercharges I = O, Y = -2. The empi ri cal f ormul a
Q = 13 + Y/2 woul d gi ve i ts el ectri c charge as -1. The parti cl e was
hopeful l y gi ven the name fi- before i t was experi mental l y di scovered. The
fate of the SU(3) symmet ry hung on i ts di scovery.
Thi s was the si tuati on i n J anuary 1964. I n February 1964, among a
Symmetry concepts und the fundumental theory of matter 85
scan of mi l l i ons of photographi c bubbl e- chamber pi ctures t wo Q-s were
di scovered at Brookhaven. The producti on and decay of these parti cl es is
spectacul ar; the fact that there are even now a total of a dozen known
speci mens of thi s parti cl e makes i t hi ghl y preci ous. The hi gher symmet r y
scheme SU3 was vi ndi cated; the i sotopi c-spi n group Q(3) w SU(2) had
been general i zed to the uni tary-spi n group SU3 as the sti l l hi gher, the
still mor e embraci ng symmet ry of the f ami l y of the nucl ear parti cl es.
The fol l owi ng questi on now ari ses. The group SU(3) has a represen-
tati on consi sti ng of three obj ects. Thi s is the so-cal l ed f undamental
representati on. All other representati ons can be f or med f r omit. The
physi ci sts have begun to call i t the Quar k representati on. I ts three
component s, the t hree i ndi vi dual quarks can be shown to possess the
fol l owi ng val ues of I -spi n and hypercharge:
I3 Y
41 z 5
92 - 3 3
q3 O -3
1
1
The el ectri c charges of the three quarks woul d, f r omthe rel ati on
Q = (13 + yY2) equal p, -3, -3- (inuni ts of the el ectri c charge). Any
one of the nucl eons i nTabl e 1 woul d be coi nposed of t hr ee quarks; any
one of the mesons of a quark and anti - quark. Quarks, if they exi st,
would then be the basi c matter, the el ementary parti cl es. So far the
experi mental search for t hemhas proved frui tl ess but one must r emember
we have not gone to very hi gh energi es ei ther withour present generati on
of accel erators. The i mportant poi nt to note real l y is that i nexorabl y the
symmet r y concepts, charge, i sotopi c charge, uni tary charge, have l ed to
a seri es of steps whi ch have gui ded the quest for the most f undamental
consti tuents of matter. No mean tri umph for the symmet r y pri nci pl es.
4.3 The SU(6) symmetry
The next devel opments inappl i cati ons of symmet ry pri nci pl es to parti cl e
physi cs came soon after, inSeptember 1964. And these were sti l l mor e
spectacul ar. At thi s stage, one had a total of four compl ete mul ti pl ets:
86 Abdus Salam
Table 3
N
N*
(nucleon) J = 3. Number of particles 8
(excited nucleon) J = 3. Number of particles 10
~ ~~
M (meson) J = O. Number of particles 8
M* (excited mesons) J = 1. Number of particles 9
One had t wo vari eti es of quant umnumbers:
(1) External ; Poi ncar- spi n; group structure O3 w SU2;
(2) I nternal ; I -spi n and hypercharge E group structure SU3.
Now when counti ng the numbers of parti cl es i n any i nternal SU(3)
mul ti pl et, no account had been taken of spi n pol ari zati ons, of spi n
mul ti pl i ci ty. Supposi ng we do treat each Poi ncar spi n pol ari zati on as
di sti nct; the count woul d then be as fol l ows:
Table 4
N (nucleons) J = 3
N*
8 X (U + 1) = 16 distinct particles
(excitednucleons)J= 4, 10 x (2J + 1) = 40 distinct particles
M (mesons) J = O, 8 x (2J + 1) = 8 distinct particles
M*
9 x (2J -t 1) = 27 distinct particles (excited mesons) J = 1,
In Sept ember 1964, the questi on was rai sed: i s i t concei vabl e that one di d
have innature a symmet ry hi gher than SU(3), a symmet ry compri si ng
both external Poi ncar spi n and the i nternal uni tary spi n for whi ch
nucl eons - both those whi ch we have cal l ed exci ted and those others
presenti ng the ground state of the structure-appear as j ust one mul ti pl et.
Coul d we possi bl y obl i terate the di sti ncti on of external and i nternal
symmetri es? Coul d we thi nk up a symmet r y group wi th both SU(3) and
SU(2) as subgroups? One obvi ous candi date was SU(6). This was tri edi n
Sept ember 1964 by F. Gursey, L. Radi cati and B. Saki ta. Among i ts
i rreduci bl e representati ons SU(6) does possess t wo representati ons wi th
di mensi onal i ti es 35 = 8 + 27 and 56 = 16+ 40. These preci sel y are the
number s of di sti nct parti cl es of the mesoni c (M+ M*) and nucl eoni c
(N + N*) vari ety. SU(6) is perhaps a sti l l hi gher symmet ry of nucl ear
i nteracti ons,
So far, so good. One was maki ng progress but how was one to under-
stand thi s mergi ng of the external Poi ncar and the i nternal uni tary
spi n? Several answers have been gi ven but none uni versal l y accepted. The
myst ery of thi s mergi ng remai ns.
Symmetry concepts and the fundamental theory of matter 87
5
Thi s then is a bri ef account of the rol e symmetri es have pl ayed i n the
evol uti on of the f undamental theory of matter. I shal l end by rei terati ng
once agai n the f undamental fai th of the physi ci st inthe el oquent wor ds
of the Hol y Book of one of the great rel i gi ons of thi s worl d.
Thou seest not in the creati on of the All Merci ful any i mperfecti on.
Return thy gaze, seest t hou any fi ssure, then return thy gaze agai n and
agai n. And thy gaze comes back tothee dazzl ed, aweary (A. J. Arberry,
The Koran Interprted).
REFERENCES
GELL-MANN, M. and NEEMAN, Y., (1964) The Eightfold Way, N e w York, Benjamin,
LANDAU, L. D., (1957) Journ. Eksp. Theo. Fiz., vol. 32, p. 405 (in Russian).
LEE, T. D. and YANG, C. N. , (1956) Phys. Rev., vol. 104, page 254, Journal of the
SALAM, A., (1957) Il Nuovo Cimento, vol. 5, page 299, Journal of the Italian Physical
WEYL, H., (1952) Symmetry, Princeton (N.J.), Princeton Univ. Press,
WIGNER, E. P., (1967) Symmetries and Refections, Bloomingdale und.), Indiana
American Physical Society.
Society.
Univ. Press.
5 Sign
LADISLAV TONDL
1. SEMI OTI CS AS THE THEORY OF SIGNS
The tradi ti onal defi ni ti ons of man as a tool - maki ng ani mal , rati onal
ani mal or the l i ke are someti mes further qual i fi ed, to make man i nto a
si gn-usi ng ani mal , animal symbolicum (E. Cassi rer, 1923-1931). I n
consi deri ng t wo types of soci al acti vi ti es - cogni ti ve processes and
communi cat i on processes - much attenti on i s pai d to the rol e of si gns,
and systems of si gns, in man s acti vi ti es. Cogni ti ve processes and com-
muni cati on processes are l i nked, reci procal and i nterl ocki ng, and
semi oti cs, the general theory of si gns and systems of si gns, has been
devel oped pri mari l y to deepen our knowl edge of t hem. The i nterest of
si gns i s obvi ousl y not confi ned to t hemal one: sign processes are al so of
great i mportance el sewhere to cul ture and behavi our - inart, rel i gi on
or pol i ti cs for i nstance. Her e, however, thei r characteri sti cs and rel ati ons
as a rul e deri ve f r omthose exhi bi ted by si gn-processes incogni ti on and
communi cati on.
The t er m semi oti cs was ori gi nal l y used by J ohn Locke to desi gnate the
general theory of si gns, fol l owi ng i deas about the f uncti on of si gns
and si gn-si tuati ons that date back to the Stoi cs. The fi rst mor e systemati c
attempt to devel op semi oti cs i nto a theory of si gns was made by Charl es
S. Pei rce (1931-1935; 1940), who i ntroduced the concept of si gn process
or semi osi s, whi ch he concei ved as a tri adi c rel ati on between si gns,
obj ects and i nterpretants. This means that any obj ect, event or property
of an obj ect, etc. may become the si gn of an arbi trary obj ect, assumi ng
that i t can be i nterpreted by someone, i.e. if an i nterpretant of the si gn
exi sts. It is i mmateri al whether such an obj ect is represented by materi al
thi ngs, events or their properti es, behavi our, states of the mi nd, obj ects
that are products of our thought, i magi nati on, and so f orth; Pei rce al so
worked out several si gn cl assi fi cati on schemes and traced connecti ng
l i nks between semi oti cs and l ogi c.
90 Ladisluv Tond1
The concepts of Pei rce wer e fol l owed up i n parti cul ar by Charl es
S. Morri s (1938, 1946, 1964), who inaddi ti on, connected inhi s works the
i nfl uences of Pei rce s pragmati smand the i nfl uences of l ogi cal posi ti vi sm.
An i nterest in the probl ems of si gns was al so characteri sti c of other
phi l osophi cal trends. Fr omneo- Kanti an posi ti ons anal ogous probl ems
wer e tackl ed by Ernst Cassi rer (1923-193 l), who especi al l y underl i ned the
i nnate human qual i ty of symbol i zati on and creati on of the real i ty by
means of thi s symbol i c creati vi ty. Attempts at devel opi ng theori es and a
typol ogy of si gns can be f ound inthe wri ti ngs of the representati ve of
phenomenol ogy E. Husserl (1913-1921), who stressed, inparti cul ar, the
speci al posi ti on of wor d si gns (Ausdriicke), incontradi sti ncti on to si gns
of other kinds (Anzeichen).
The sources of i nterest in si gns, thei r rol e inthe cogni ti ve and com-
muni cati on processes and in semi oti cs i n general can obvi ousl y on no
account be reduced to phi l osophi cal school s or wor ks i nspi red by the
i nvesti gati on of phi l osophi cal probl ems al one. For the devel opment of
an anal ysi s of si gns and si gn systems, as well as the el ementary probl ems
of semi oti cs, other fields of sci ence or scientific topi cs have al so been of
great i mportance, especi al l y those rel ati ng to l i ngui sti cs, l ogi c, psychol ogy
and aestheti cs, to menti on but some of t hem.
Semi oti cs has l ong been l i nked with l i ngui sti cs. Starti ng f r omthe
cl assi cal wor k of Mi chel Bral (1904), i ts probl ems have been regarded as
a l egi ti mate part of the sci ence of l anguage. I n tradi ti onal , l i ngui sti cal l y-
ori ented semi oti cs, i t was the wor d (word si gn) and i ts meani ng or the
changes occurri ng inthi s meani ng that werethe mai n poi nt of attenti on,
but i ts scope has now been substanti al l y extended (parti cul arl y inconnec-
ti on wi th machi ne transl ati on) through the appl i cati on of new l ogi cal
and mathemati cal met hods to the study of natural l anguages, the
anal ysi s of the rel ati ons between natural and formal i zed l anguages,
and so on.
Logi c, inanci ent and medi aeval ti mes, was much concerned wi th si gns
and meani ngs, the rel ati onshi p between wor d and concept, the conno-
tati on and denotati on of concepts, and the l i ke. Thi s i nterest has
qui ckened si nce the end of the ni neteenth century, si nce no sati sfactory
sol uti on of f undamental questi ons i nl ogi c, the el uci dati on of probl ems
connected withcertai n paradoxes, withthe concepts of i denti ty, l ogi cal
consequence, model , and so on, is possi bl e wi thout due regard bei ng
pai d to rel ati onshi ps with whi ch semi oti cs i s also concerned. The
Sign 91
pi oneers here i ncl ude Frege, Russel l , Kotarbi hski , Carnap, Chur ch,
Qui ne and Marti n.
General semanti cs will, i t i s hoped, be an aut onomous di sci pl i ne for the
study of the functi ons and meani ng of si gns. Phi l osophi cal l y, i ts pr opo-
nents hol d an ecl ecti c mi xture of opi ni ons dr awn f romvari ous school s
and di sci pl i nes. On the whol e, the most frui tful sti mul i have come f r om
general psychol ogy, soci al psychol ogy, psychi atry, neurol ogy, psycho-
l i ngui sti cs and aestheti cs. The meaning of meaning by Ogden and
Ri chards (1923) for exampl e, anal yzes the emoti ve and aestheti c effect of
words, di sti ngui shes between the emoti ve and the i ntel l ectual uses of
l anguage, and f ormul ates a basi c scheme of rel ati onshi ps between si gn,
obj ect and thought (the Ri chards tri angl e of reference). For Korzybski ,
Hayakawa, Chase and some other authors, general semanti cs provi des a
means of psychi cal and soci al therapy , of mental or soci al hygi ene ,
maki ng for the conci l i ati on of soci al confl i cts; i t i s concerned wi th the
magi c effect of words, fi ghts the so-cal l ed Ari stotel i an l anguage system
(i denti fi ed pri mari l y wi th two- val ued l ogi cal systems). Some of these
wri ti ngs i ncorporate el ements of Freud s psychoanal ysi s, Pavl ov s
theory of condi ti oned refl exes and certai n di scoveri es of l ogi c and
l i ngui sti cs.
A somewhat di fferent l i ght is t hrown on the functi on of si gns and si gn
systems, especi al l y i n communi cat i on processes, by research incul tural
anthropol ogy, comparati ve and hi stori cal ethnography, hi stori co-
comparati ve l i ngui sti cs and psychol ogy - especi al l y by research on
pri mi ti ve mental i ty and the rel ati ons between thought and l anguage
among pri mi ti ve peopl es. The hypothesi s of l i ngui sti c rel ati vi sm(Sapi r
and Whorf ) sees l anguage as a whol e and i ndi vi dual wor d si gns as not
onl y a certai n i mage of the soci o-pragmati cal l y determi ned real i ty but at
the same timei ts gui de. Psychol ogi cal research concerned wi thsi gns and
meani ngs, and withthe measurement of meani ngs, inart and ingeneral ,
has been extensi ve and vari ed.
Semi oti cs (semasi ol ogy, semanti cs), accordi ngl y, as the general theory
of si gns, does not provi de a homogeneous or wel l -ordered pattern of
sci enti fi c resul ts; di fferences inconcepti ons, termi nol ogy, and approach
are so enor mous as to precl ude a neat summar y of resul ts that woul d
embrace t hemall.
92 Ladislav Tondl
2. THE NATURE OP SIGNS
In everyday l anguage, the t er m si gn (and others wi th al l i ed meani ng,
e.g. symbol , si gnal , token ) i s used i n a very broad sense. Smoke is a
si gn of a fi re, a col our or ani mal i s a symbol of a pol i ti cal party, a col oured
rocket i s a parti cul ar si gnal , a nod is a si gn of agreement, and so on. Thi s
heterogenei ty of si gns and of the contexts inwhi ch we encounter t hem,
and accordi ngl y the di fferences al so, suggest l hat the approach in con-
si deri ng t hemmust be equal l y broad. Bertrand Russel l for exampl e
poi nted out that for an organi smO any el ement of the cl ass of sti mul i Ais
the si gn of some el ement of cl ass By if the real i zati on of thi s sti mul us cal l s
forth, i norgani smO, a response correspondi ng wi th the el ement of cl ass
B. Thi s si gn scheme worki ng wi th the concepts of sti mul us and response is
actual l y cl ose to the concept of the condi ti oned refl ex i n the context of
Pavl ov s physi ol ogy. The mechani smof the condi ti oned refl ex apparentl y
al so suppl i es the physi ol ogi cal basi s of what Pei rce and Morri s call the
si gn process or semi osi s . Morri s di scerns fi ve component s of the si gn
process: 1) the si gn, 2) the obj ect to whi ch the si gn refers (designatum),
3) the i nterpreter, 4) the i nterpretants, i .e. the response on the part of the
i nterpreter or the di sposi ti on of the i nterpreter to respond i n a defi ni te
manner to a gi ven si gn, and possi bl y 5) the contexts in whi ch the si gns
occur. I n viewof thei r deci si ve i mportance the fi rst three component s and
thei r mutual rel ati ons are di scussed bel ow.
The above concepti on of si gn is extremel y broad. It coul d be appl i ed to
any response of an organi sm(or of a cyberneti c equi pment model l i ng the
behavi our of the organi sm) whi ch has the character of a condi ti oned
refl ex. Hence, a di sti ncti on must be made between si gn processes inthe
broadest sense, and si gn processes rel ated to soci al communi cati on. I n the
Pavl ov reacti on of the dog to the bel l , all the component s of semi osi s can
be di sti ngui shed. It is not tiedto soci al communi cat i on and i ts component s
are not soci al l y condi ti oned. Accordi ngl y, we do not encounter si gns in
the proper sense, i.e. soci al artefacts that are only ex post i nterpreted as
si gns of somethi ng. No mor e exact di sti ncti on can be dr awn between the
t wo types of si gn processes, or between natural si gns and si gns of
proper (thi s di sti ncti on roughl y correspondi ng to the tradi ti onal
di fferenti ati on between the signum naturale and the signum ad placitum).
Thi s expl ai ns why, i n semi oti c l i terature, we meet withsuch a vari ety of
si gn typol ogi es and cl assi fi cati ons. Her e we shal l be mai nl y concerned
Sign 93
wi thsi gnsproper, butwecannotwhol l y excl udethei ntrusi on ofcertai nchar-
acteri sti c traits of natural si gns and si gn processes based on natural si gns.
We may fi nd, acti ng as a si gn, any materi al obj ect, i ts property, an
event, the property of an event, or the l i ke, whi ch inthe communi cat i on
process can serve the purpose of communi cati ng i nf ormati on, i.e. any
materi al obj ect, etc. that can be i nterpreted. By the i nterpretati on of si gns
we understand a process whi ch enabl es the i nterpreter, on the basi s of the
si gn, tomake a deci si on as to the obj ects to whi ch the si gn i s assi gned. We
usual l y say that for the i nterpreter the si gn has a certai n meani ng,
i rrespecti ve of the fact that the concept of meani ng is i nterpreted in
di fferent ways. Thi s means that there can be no si gns whi ch woul d be
devoi d of meani ng, and conversel y whatever can be subsumed under the
concept of meani ng can onl y be i sol ated by means of a si gn l i nked
withthe gi ven meani ng. Fr omthe above characteri sti c i t is cl ear that a
si gn must, necessari l y, have the fol l owi ng properti es: (A) Communi cabi l -
i ty, i.e. the abi l i ty to communi cat e i nf ormati on i n space and ti me; (B)
I nterpretabi l i ty.
I n cogni ti ve and communi cati on processes si gns of most di verse types
are brought i nto pl ay. The di fferences i n the outward appearance, uses
and functi ons of si gns i ncogni ti on and communi cat i on are so great as to
make i t difficult to wor k out a si ngl e typol ogy of si gns. Normal l y wor d
si gns or t er ms are di sti ngui shed f romsi gns of other types. I n the cogni ti ve
and communi cat i on processes wor d si gns enj oy a speci fi c posi ti on: by
means of wor d si gns we i nterpret si gns of other ki nds, and by means of the
properti es and rel ati ons pecul i ar to wor d si gns we reconstruct properti es
and rel ati ons bel ongi ng to other si gns.
Among si gns that do not possess the character of wor d si gns we i ncl ude,
for i nstance, di verse si gnal s and agreed si gns. For a worker, the sound of
a hooter means i t i s timeto fi ni shhis wor k, a l i ght si gnal at a crossroads
pl ays the part of a di recti on addressed to those operati ng the means of
transport. Also certai n ki nds of behavi our can have the functi on of a
si gn e.g. noddi ng the head is a si gn of agreement, wavi ng the hand is a
si gn of farewel l .
A speci al group of si gns i s f ormed by those when the rel ati onshi p
between the si gn and the denoted obj ect i s obvi ous, because of sensory
si mi l ari ty, as i nthe case of pi ctures, sketches, drawi ngs or photographs.
I n such i nstances we speak of i conographi c si gns, whi ch represent
materi al , sensori l y percepti bl e obj ects. I n the process of soci al com
94 Ladislav Tondl
muni cati on through the medi umof si gns we al so represent obj ects that
are not sensori l y percepti bl e, such as human vi rtues, feel i ngs, rel i gi ous
i deas, pol i ti cal views. Usual l y we say that the cross symbol i zes Chri sten-
dom, the crescent I sl am, the owl symbol i zes wi sdom, the l i on bravery,
and so forth. Many of these si gns are l i nked with thousand- year ol d
tradi ti ons of i ndi vi dual cul tures, and the mechani smof thei r i mpacts need
not be i ntel l ectual onl y, but al so emoti onal .
The above outl i ned di sti ncti on of some el ementary types of si gns
natural l y represents but one of the pl ural i ty of possi bl e vari ants of the
typol ogy of si gns, with the di fferent l i nks and transi ti ons (between
i ndi vi dual types of signs). Besi des the typol ogi es of the sai d or si mi l ar
ki nds, whi ch poi nt pri mari l y to exteri or di fferences and l i nks bet ween
i ndi vi dual ki nds of si gns, account shoul d al so be taken of other di sti nc-
ti ons, di fferenti ati ng the si gns i na mor e abstract manner , on the basi s of
their f uncti on in the communi cat i on process. The si gn is percei ved,
regi stered and di sti ngui shed al ways as a whol e, as a shape, as an el ement
of a certai n cl ass of si gns. A fai thful Chri sti an i nterprets the cross, i.e. the
symbol of his fai th, to a certai n effect - regardl ess of whether he i s f aced
wi th a mi ni ature obj ect he carri es around his neck, or a gi ant structure
erected at the peak of a church steepl e or of a mountai n. Thi s means that
inthe course of a si gn-process i t is al ways a certai n si gn-event that i s
percei ved, but whi ch i s i nterpreted int er ms of the membershi p of this
si gn-event ina certai n si gn cl ass. A wri ttenwor d of a defi ni te l anguage is
understood inthe same way, i rrespecti ve of the type of l etteri ngi t has been
writteni n. For these reasons i t i s useful to di scern the si gn as a si gn-event
and the si gn as a si gn desi gn. (The termi nol ogi cal di sti ncti on of the t wo
types of si gns differs withi ndi vi dual authors, but the f undamental s of the
di fferenti ati on essenti al l y agree. I n the same sense S. C. Pei rce (1931-
1935; 1940) made a di sti ncti on between token and type , R. Car nap
between si gn-event and si gn-desi gn , H. Rei chenbach (1947) bet ween
token and symbol , K. Dr r (1954) between si gns in concreto and
si gn in abstracto, A. A. Mar kov (1954) between concrete l etters and
abstract l etters. I nasmuch as we di scern and i nterpret si gns in their
enti rety, in the anal ysi s of semi oti c probl ems in l ogi c, l i ngui sti cs,
i nf ormati on theory and other fields i t i s mor e customary to consi der the
tsi gn-desi gns as the poi nt of departure for the i nvesti gati on of probl ems
of thi s ki nd. However , there are al so seri ous attempts of an extremel y
nomi nal i sti c vi ewpoi nt, whi ch regard the si gn-events as the starti ng
poi nt of semi oti c anal ysi s.
Sign 95
3. LEVELS OF SEMIOTIC ANALYSIS
Semi oti c anal ysi s i s pri mari l y based on three component s of the si gn
process: 1) si gns, 2) obj ects to whi ch the si gn refers, 3) i nterpreters. (For
the i ndi vi dual component s the l i terature empl oys di fferent t er ms, for
exampl e for 3): communi cant , user of the si gn, user of the l anguage,
speaker, etc.) I n thi s connexi on the i nterpreter is assumed to be abl e to
understand the gi ven si gns: this means that a response by the i nterpreter,
or the di sposi ti on of the i nterpreter to respond to the gi ven si gn in an
adequate manner, are both possi bl e. Taki ng i nto consi derati on the three
basi c component s of the si gn process and assumi ng that we are faced
wi th a si gn-systemhavi ng a l i ngui sti c structure, we can di sti ngui sh three
l evel s of semi oti c anal ysi s:
- I n so far as we take i nto account onl y the si gns and their mut ual
rel ati ons, abstracti ng f r omthe rel ati onshi p betweeen the si gns and the
remai ni ng component s of the si gn-process, we operate at the syntactic
level of semiotic analysis.
- If, i n addi ti on, we take account of the rel ati ons between the si gns
(or the texts consi sti ng of these si gns) and the obj ects to whi ch they refer,
abstracti ng f r omthe rel ati onshi p to the i nterpreters of the si gns, we
operate at the level of semantic analysis.
- If we were al so to i ncl ude the rel ati onshi p to the i nterpreters, and
possi bl y their moti ves, goal s, reacti ons, di sposi ti ons and so f orth, we
woul d be operati ng at the level of pragmatic analysis.
It is cl ear that this di fferenti ati on of the three l evel s of semi oti c anal ysi s
represents a certai n abstracti on. Every actual si gn-process natmal l y
presupposes that all of the rel ati onshi ps ment i oned are brought i nto pl ay.
Fr omthis angl e the syntacti c l evel of semi oti c anal ysi s i s actual l y the
most abstract, and the pragmati c l evel the most concrete of the three.
The rel ati ons among the t hr ee component s may assume the f or mof
rul es, whi ch accordi ng to the nature of these rel ati ons are characteri zed
as syntacti c, semanti c and pragmati c rul es.
The si gn-systems compri si ng syntacti c and semanti c rul es are al so
characteri zed as sign systems with a linguistic structure or l anguages
(inthe general sense of the word). Fr omthi s standpoi nt we may make a
di sti ncti on between si gns that are el ement s of a si gn-systemwi th a
l i ngui sti c structure (for i nstance written or uttered words, symbol i cal
means of mathemati cs or l ogi c, etc.), and si gns that do not f orml i ngui sti c
96 Ladislav Tond1
structures (for exampl e, the hammer and si ckl e, fl ags, coats- of - arms,
etc.). The actual basi s of soci al communi cat i on as well as the el ementary
means of expressi on and preservati on of the resul ts of cogni ti ve acti vi ty
are si gn-systems wi th l i ngui sti c structure.
Among si gn-systems withl i ngui sti c structure rank, inthe fi rst pl ace,
natural l anguages e.g. Engl i sh, Chi nese, Swedi sh. Fr om natural
l anguages we somet i mes di sti ngui sh the so-cal l ed artificial l anguages,
such as Esperanto, the structure of whi ch does not however substanti al l y
di ffer f romthose of the natural l anguages. A speci al group is f ormed by
the so-cal l ed f ormal i zed si gn-systems or formal i zed l anguages whi ch
wor k wi th constructi ve symbol i cal means and whose structure is fi xed
by exactl y and uni quel y f ormul ated rul es.
Al so the f ormal i zed si gn-systems use means servi ng to express certai n
obj ects, events or operati ons. These means, whi ch have been constructed
in mathemati cs and l ogi c, can have ei ther a constant or a vari abl e
meani ng, the scope of the vari abi l i ty bei ng al ways determi ned. Fr om
this poi nt of vi ewthe si gns of a f ormal i zed si gn-systemcan be di vi ded
i nto constants and vari abl es. When devel opi ng a formal i zed si gn-system
certai n mandat ory steps must be taken: firstly a compl ete l i st of si gns
must be dr awn up, and the vari ous ki nds of si gns must be di sti ngui shed.
The next step consi sts indetermi ni ng the mode inwhi ch these si gns are to
be used, i.e. how they can be made up i nto expressi ons qual i fi ed ina
certai n manner, for exampl e wel l - f ormed f ormul ae, sentences. I n thi s
respect i t is usual to speak of the rul es of f ormati on of a formal i zed
si gn-system. The rul es of f ormati on determi ne exactl y whi ch connecti ons
and whi ch sequences of si gns have a certai n qual i fi ed character, i.e.
whi ch are sentences, wel l - f ormed f ormul ae, etc., and whi ch sequences of
si gns are l acki ng such a character. The next step inthe devel opment of
a formal i zed si gn-systemis the determi nati on of i nference rul es whi ch
woul d enabl e us to deri ve one set of expressi ons f r omother expressi ons.
These steps, whi ch consti tute the syntacti c component of the devel opment
of a formal i zed si gn- system, enabl e us to undertake certai n operati ons
i nvol vi ng si gns as well as expressi ons f or med theref rom, wi thout knowi ng
the meani ng of the si gns and expressi ons. If however, we are todetermi ne
the meani ng, what we need are the semanti c rul es, the sumof whi ch
consti tutes the semanti c component of the constructi on of a formal i zed
si gn-system. I f any of the t erms (correctl y f ormed accordi ng to the rul es of
f ormati on) are characteri zed as axi oms (i.e. thei r val i di ty is postul ated,
Sign 97
other expressi ons admi tted as val i d onl y i nso far as they can be deduced
f r omthe axi oms), the gi ven f ormal i zed si gn-systemis an axi omati zed
system.
The devel opment of semi oti c anal ysi s at all the menti oned l evel s
presupposes that the si gns and thei r mutual rel ati ons as well as thei r
rel ati ons toother obj ects have become obj ects of i nvesti gati on, the resul ts
of whi ch we natural l y al ways express through the i ntermedi ary of other
si gns. It is therefore expedi ent to make a di sti ncti on bet ween obj ect-si gns
and anal ogousl y obj ect-si gn systems etc. , obj ect-l anguages and the l i ke,
inrespect of whi ch we f ormul ate statements - and metasi gns, metasys-
tems, metal anguages, by means of whi ch we f ormul ate our statements.
Such a di sti ncti on has been carri ed out, inparti cul ar, inl ogi c, because
mi xi ng obj ect-si gns and metasi gns may resul t i nparadoxes.
By anal ogy withthi s di sti ncti on we must di scern exactl y when the sign
has been used to denote an obj ect, and when a si gn is on the contrary
menti oned as somethi ng that denotes itself. Thi s shoul d be cl ear f r om
the fol l owi ng exampl e:
London is the capi tal of the Uni ted Ki ngdom.
London has 6 l etters.
Al though the si gns London and capi tal of the Uni ted Ki ngdom
denote the same obj ect, namel y the city si tuated on the Thames, we
cannot mutual l y i nterchange London and London , because in the
f ormer i nstance the si gn i s used todenote another obj ect, whereas inthe
the latter case i t i s meant toexpress the si gn itself.
4. SIGN AND MEANING
If i n the communi cat i on process we come across si gns, we are in a
posi ti on to compr ehend t hemprovi ded we can answer the fol l owi ng
or si mi l ar questi ons: what does ... mean , what is the meani ng of ...,
what does ... refer to, and so on. I n all these and si mi l ar questi ons i t i s
presumed that we have assi gned the gi ven si gn to other extral i ngui sti c
obj ects. (It i s onl y natural that the di sti ncti on of l i ngui sti c and extra-
l i ngui sti c obj ects i s rel ati ve, i.e. is onl y feasi bl e withregard to the gi ven ob-
j ect-l anguage. ) Such an assumpti on is val i d, however, onl y condi ti onal l y:
i n si gn-systems withl i ngui sti c structure no such obj ects can be assi gned to
certai n si gns; the functi on of thi s speci al ki nd of si gns i s l i nked withthe
98 Ladislav Tondl
syntacti c structure of the si gn system. For these reasons i t i s essenti al to
ampl i fy the i ntroductory note al so to cover the questi on as to what
assumpti ons must be made for us to be capabl e of comprehendi ng the
si gns: we are abl e to comprehend t hemif we are fami l i ar wi th thei r
extral i ngui sti c meani ng - or if we know the functi on of these si gns inthe
syntacti c structure of the si gn system.
I n natural l anguages, accordi ng to an ol d tradi ti on we keep apart the
so-cal l ed syncategorematic words whi ch have no extral i ngui sti c meani ng
by themsel ves, but onl y i nconj uncti on withwords havi ng an i ndependent
extral i ngui sti c meani ng of thei r own. I n other words: syncategoremati c
si gns are by t hemsel ves not abl e to make any statements on the gi ven
uni verse of di scourse - they can onl y make such statements inconj uncti on
withcategoremati c words.
I n f ormal i zed l anguages we di sti ngui sh logical signs from extralogical
signs. I f the functi on of the si gns remai ns i nvari ant wi th respect to any
uni verse of di scourse, we are conf ronted wi th l ogi cal si gns. Logi cal si gns
as such do not ref er to any extral i ngui sti c obj ects: i ncogni ti ve processes
and communi cat i on processes al i ke, they can be used onl y inconj uncti on
with extral ogi cal si gns. An exact and unequi vocal di sti ncti on between
l ogi cal and extral ogi cal si gns i n si gn-systems wi th l i ngui sti c structure
is not si mpl e and as a resul t there i s no uni f ormconcepti on of such a
di sti ncti on. Si gn- systems wi th l i ngui sti c structure are hi erarchi cal l y
arranged. For exampl e, words can be made up i nto a cl ause, cl auses can be
made up i nto compl ex sentences: all these expressi ons we are abl e to
i nterpret, namel y make deci si ons regardi ng the obj ects or states of the
gi ven uni verse of di scourse to whi ch we assi gn the expressi ons. By
anal ogy thi s i s al so val i d for formal i zed si gn systems. Of all these si gns
and expressi ons inrespect of whi ch such an assi gnment is possi bl e we say
that they are capabl e of bei ng semanti cal l y anal yzed or semanti zed.
The semanti zati on of si gns and of semanti zabl e expressi ons i ngeneral
does not proceed inaccordance witha general scheme. What is essenti al
i nthi s connexi on i s the character of the gi ven uni verse of di scourse, the
syntacti c structure of the sel ected si gn-systemand the l evel of the seman-
ti zabl e expressi ons. (It i s for exampl e evi dent that the process of semanti -
zati on will differ i nthe case of words representi ng names of obj ects, inthe
case of cl auses, compl ex sentences, sententi al contexts, etc.)
Whi l e i t i s not possi bl e to poi nt out a uni f ormscheme of semanti zati on,
i t i s expedi ent to sketch some el ementary trends inwhi ch semanti zati on
Sign 99
can be ori ented, especi al l y inconsi derati on of the fact that si gn-systems
come i nto pl ay i ncogni ti ve processes and coi nmuni cati on processes. The
l i nkages between cogni ti ve processes and communi cat i on processes, and
hence al so the l i nks between the basi c trends of semanti zati on can be
outl i ned i n the fol l owi ng di agram:
Bl ock A whi ch we may characteri ze as a source of i nf ormati on, i s the
obj ect area under i nvesti gati on: bl ock B can be a human observer, or an
observer fittedwi th equi pment extendi ng hi s sensory ability to di scern, or
fi nal l y any measuri ng or experi mental equi pment abl e to regi ster sti mul i
comi ng f r ombl ock A. Bl ock C i s a systemof expressi ons. The rel ati ons
between A, B and C schemati cal l y represent the cogni ti ve process,
rel ati ons between C, C' and C" etc. the communi cat i ve process. Knowl -
edge concerni ng A, enabl i ng us to achi eve the semanti zati on of the
el ement s C, i s then acqui red by t wo ways: on the basi s of what B.
Russel l characteri zed as knowl edge by acquai ntance, and on the basi s of
the communi cat i on process (Russel l spoke inthis connexi on of knowl edge
by descri pti on. In the f ormer case extral i ngui sti c means can al so be used
for the semanti zati on of si gns or expressi ons made up of si gns - in the
l atter case other si gns or other expressi ons are al ways empl oyed for
semanti zati on. The procedures of the semanti zati on of si gns whi ch are
based on extral i ngui sti c means, such as poi nti ng to the obj ect to be
denoted, or poi nti ng out the event to be denoted, or carryi ng out the
operati on i nquesti on and the l i ke, are characteri zed as ostensi ve proce-
100 Ludisluv Tondl
dures. I n process of masteri ng the meani ng of the si gns, ostensi ve
procedures represent but the most el ementary l evel s of semanti zati on.
Hi gher l evel s of semanti zati on are based on the appl i cati on of other si gns,
for exampl e defi ni ti ons, expl i cati ons, i nterpretati ons etc. The same
appl i es to the ways in whi ch chi l dren l earn l anguages, the ways and
met hods of studyi ng other l anguages, l earni ng of the meani ng of un-
fami l i ar si gns, and so forth.
Leavi ng asi de the l ogi cal si gns, the semanti zati on of si gns and ex-
pressi ons f ormi ng the el ements of bl ock C can be understood as assi gn-
ment. The concept assi gnment is encountered i na si mpl e, and somewhat
nai ve questi on: what corresponds to . . .. Thi s questi on postul ates the
assi gnment to bl ock C of el ements of bl ock A, and possi bl y of bl ock B
(inthis event we speak of the expressi ve functi on of the si gn-system, i n
contradi sti ncti on to other functi ons), or ul ti matel y el ements of bl ock
C, C, etc. This questi on actual l y presupposes structural si mi l ari ty of
bl ocks A, B and C, whi ch i s a most probl emati c supposi ti on. Onl y i nthe
si mpl est case has the assi gnment the character of a one- to- one rel ati on,
for exampl e withproper nouns of i ndi vi dual obj ects. In most cases the
assi gnment has a much mor e compl ex f or m, for exampl e wi th general
expressi ons, si gns denoti ng theoreti cal concepts, and so on.
Semanti zati on of si gn-systems presents a seri ous probl em: the probl em
of l evel s for the assi gnment. Assumi ng that bl ocks A, B and C are f ormed
by certai n el ements, we must not l ose si ght of the fact that these el ements
may be categori zed incertai n ways, that they f ormcertai n whol es and
establ i sh certai n rel ati ons, that they are hi erarchi cal l y arranged incertai n
l evel s, etc. If in bl ock C we use natural l anguages, such as Engl i sh,
Ger man, Russi an and the l i ke, we shal l fi nd that the expressi ons appeari ng
inCar e made up of words, cl auses, compl ex sentences etc. If, however, we
make use of a f ormal i zed l anguage the l ogi cal basi s of whi ch i s the pred-
i cate cal cul us, we must di sti ngui sh i ndi vi dual s, predi cates, sentences
and f ormul ae. In thi s manner we can reach a questi on that coul d be
f ormul ated in the fol l owi ng way: wi th regard to whi ch category of
di scerni bl e el ement s and at whi ch l evel are we enti tl ed to thi nk about
assi gnment? Anal ogousl y si mi l ar questi ons can al so be f ormul ated wi th
respect to the structure of the other bl ocks.
If i t is possi bl e - invi ewof the gi ven uni verse of di scourse and i nview
of the gi ven si gn-system- to f ormul ate a sati sfactory answer to the gi ven
questi on, i.e. sel ect sui tabl e levels for assi gnment, then i t i s feasi bl e to
Sign 101
i sol ate, inthe si gn-system, those si gns and expressi ons f ormed theref rom
whi ch can be regarded as names of the obj ects denoted. The anal ysi s of
the meani ng of si gns based on the name- rel ati on is, however, not the
onl y i magi nabl e approach to the probl ems of meani ng. Besi des vari ous
vari ants of the theory of the name- rel ati on a number of other concepti ons
have been devel oped: i n contemporary semi oti cs there does not exi st a
si ngl e, uni versal l y accepted concepti on of meani ng. Al so the scope of
signs that can be regarded as names is, as a rul e, not uni f orml y determi ned.
I f we regard certai n si gns as names of the obj ects to be denoted, we
shal l not be abl e to avoi d certai n difficulties. Di ffi cul ti es ari se, in
parti cul ar, i f we are conf ronted wi th the task of determi ni ng whi ch si gns
are i denti cal , what are the condi ti ons of the i nterchangeabi l i ty of si gns,
what means that t wo si gns have the same meani ng, and the like.
The fi rst attempt at a mor e systemati c concepti on of the name- rel ati on
is represented by the concepti on of Frege who di sti ngui shes between the
denotati on of a name and i ts sense. (The authors who devel oped and
modi fi ed Frege s concepti on have someti mes empl oyed a di fferent
termi nol ogy. ) Frege starts wi th the fol l owi ng questi on: are the names
eveni ng star and morni ng star i denti cal ? And if so, then in what
respect? His answer i s si mpl e: i t is conveni ent to assi gn to the name not
onl y the denoted obj ect, but al so i ts sense. Semanti c anal ysi s can therefore
be carri ed out in t wo ways: names have denotati on and sense. The
establ i shment that the names eveni ng star and morni ng star have
di fferent senses, but the same denotati on i s not enti rel y tri vi al , for such
an establ i shment assumes a certai n amount of knowl edge. The same can
be demonstrated on the of t- quoted exampl e by Russel l : Wal ter Scott
and the author of Waverl ey denote the same person - this fact not bei ng
however al ways known - whereas the sense of the t wo si gns di ffers.
In l ogi c as well as i n l ogi cal semanti cs, the approaches so far most
thoroughl y devel oped are those whi ch are based on the concepts of
denotati on (or extensi on).
Approaches of thi s ki nd have been worked out, inparti cul ar, by A.
Tarski , R. Car nap and A. Church. Much mor e difficult is the devel opni ent
of the semanti c approaches based on the concept of sense (i ntensi on).
The cri ti cs of detonati onal - or extensi onal - semanti cs woul d ri ghtl y
poi nt out that the reason why we understand a si gn i s not that we can
poi nt out i ts denotati on, but because we know i ts sense. We al so under-
stand si gns and expressi ons wi th respect to whi ch we are absol utel y
102 Lrrdduv Tond1
unabl e to make reference to thei r denotati on, such as the ' round square'
or the perpetuum mobile and the l i ke. The ways of understandi ng the
sense or i i l tensi on of si gns differ greatl y: the sense can be understood as
a postul ated abstract enti ty, as a concept, as a set of all l ogi cal l y equi val ent
expressi ons or all synonymous expressi ons, as what is i nvari ant i n any
correct transl ati on of the si gn, etc. Whereas the cri teri a for the establ i sh-
ment of an i denti cal denotati on are consi derabl y si mpl er, such cri teri a for
the establ i shment of an i denti cal sense are much mor e compl i cated, and as
yet subj ect to extensi ve di scussi ons.
The semanti c anal ysi s of si gns, whi ch is based on the concepts of
denotati on and sense, actual l y abstracts f romthe fact that there al ways i s
an i nterpreter of a si gn. Wi th the si gn the i nterpreter al ways associ ates
certai n psychi cal states or bel i efs of hi s own, connects certai n external or
i nternal reacti ons or di sposi ti ons to such reacti ons, certai n deci si ons of
his own. Hence we do not debar - especi al l y i nthe semanti c anal ysi s of the
si gns i n the communi cat i on process - such assi gnments as are based upon
the atti tude of the i nterpreter as the subj ect of thought, acti on, deci si on,
and so on.
I f we now revert to our scheme, the previ ousl y sketched ways of the
semanti zati on of si gns representi ng el ements of bl ock C can be under-
stood as vari ous manners of assi gnment, and in thi s respect we can
operate wi th both the el ement s of bl ock A and el ement s of bl ock B, or
el ements of bl ocks C, C', C", etc. Wi th the moder n cyberneti c approach to
the mut ual connexi ons of these bl ocks, f eedbacks between the di fferent
bl ocks must al ways be presumed. Thi s i mpl i es for exampl e that because
of the di screte character of el ements C we often tend to regard as di screte
al so the el ements A, or el ement s B. It shoul d be stressed here that a
si gn-systemusual l y onl y affords the possi bi l i ty of a di screte representati on
of obj ects, events, experi ences, and so forth, whi ch inthemsel ves may be
ei ther di screte or conti nuous.
5. SIGNS A N D CYBERNETIC MACHINES
I f we characteri ze a si gn as any materi al obj ect, i ts property, an event, a
property of an event etc., that inthe communi cat i on process can serve to
communi cat e i nf ormati on or that can be i nterpreted, i t shoul d be under-
l i ned that the i nterpreter does not have tobe, of necessi ty, man. By means
Sign 103
of cyberneti c equi pment present-day man can i mi tate the abi l i ty to
i nterpret si gns, and hence al so the abi l i ty to process i nf ormati on.
Theref ore, contemporary computers can al so be characteri zed as si gn-
usi ng machi nes, Man as the i nterpreter of si gns i s capabl e of respondi ng
to si gns in a certai n way, whether thi s response will mani f est i tsel f i n
maki ng certai n deci si ons, i n the f ormof outward behavi our or i n the
f or mof mental processes. The cyberneti c machi ne al so reacts to si gns of
a certai n ki nd by maki ng deci si ons inf avour of certai n operati ons. By
means of these operati ons the machi ne sol ves the tasks entrusted to it.
To sol ve a task means to convert the gi ven task i nto a sequence of
el ementary operati ons that the machi ne i s abl e to carry out. The i nstruc-
ti on to carry out a sequence of operati ons with a viewof tackl i ng a
certai n cl ass of tasks on the basi s of initial data - i s termed an algorithm.
An al gori thmmust fulfil the fol l owi ng requi rements:
- i t must be determi ned, i .e. at no stage of the operati ons must i t
permi t doubts as to how to proceed further;
- i t must be abl e to depart f romvari abl e initial data, i.e. i t must
provi de an i nstructi on for the sol uti on of a certai n cl ass of rel ated tasks;
- i t must be capabl e of l eadi ng tothe sol uti on requi red.
A task can be consi dered al gori thmi cal l y sol vabl e if i t i s feasi bl e to
construct an al gori thmfor the i nvesti gati on of the task. The si gn-system
through the medi umof whi ch an al gori thmi c sol uti on and hence al so
i ndi vi dual el ementary operati ons as well as thei r sequence can be ex-
pressed i s characteri zed as an algorithmic language. An al gori thmi c
l anguage must be capabl e of recordi ng the set of el ementary operati ons,
thei r mutual rel ati ons, thei r sequences, and so forth. Thi s means, for
exampl e, that i t must be l ai d down what rel ati ons and sequences are
permi ssi bl e and what are not. If the sumof the el ementary operati ons
is represented by an operati onal code of a cyberneti c machi ne, then the
al gori thmi c l anguage for thi s set of operati ons can be referred to as the
l anguage of the machi ne .
Al gori thmi cal l anguages, in whi ch data and i nstructi ons can be put
i nto a cyberneti c machi ne wi th the ai mof sol vi ng a certai n task, can be
regarded as a speci al ki nd of formal i zed si gn-systems.
The initial data needed for the sol uti on of a gi ven task as wel l as the
respecti ve i nstructi ons must therefore be put i nto the machi ne in the
l anguage of the machi ne . Thi s however necessari l y i mpl i es that such
data and i nstructi ons must be transl ated, i.e. for the data and i nstructi ons
104 Ladislav Tond1
expressed i n the ori gi nal si gn-systemwe must substi tute data and
i nstructi ons expressed inthe l anguage of the machi ne . Thc rul e govern-
i ng this transl ati on is usual l y referred to as a code. I n the general sense
we cal l codi ng any representati on of a defi ni te obj ect or event by means
of other obj ects or events, i.e. a substi tuti on of one set of si gn for
another set of si gns, whi l e mai ntai ni ng the semanti c characteri sti cs of the
si gns. (The questi on as to what semanti c characteri sti cs are actual l y
preserved i s a subj ect of di scussi ons, and may be answered in vari ous
ways. )
6. METHODOLOGICAL SIGNIFICANCE OF SEMIOTICS
The probl ems wi th whi ch semi oti cs i s concerned do not f or ma cl osed
and homogeneous domai n, but tonch upon and transcend a seri es of
sci enti fi c di sci pl i nes: wherever we handl e data that can al ways be ex-
pressed by means of si gns and expressi ons f or med theref rom, i t i s neces-
sary to take i nto account semi oti c probl ems. Thi s is inthe fi rst pl ace true
of those di sci pl i nes inwhi ch accuracy of expressi on has been brought to
the hi ghest degree of perfecti on: mathemati cs and l ogi c. In the j usti fi -
cati on of the f oundati ons of mathemati cs we shal l come up agai nst
probl ems that are cl osel y ti edto semi oti cs. Also i nl ogi c we shal l fi nd i t
i mpossi bl e to tackl e certai n key probl ems wi thout taki ng i nto consi d-
erati on the resul ts of semi oti c anal ysi s, especi al l y the resul ts of the
so-cal l ed l ogi cal semanti cs deal i ng wi th the semanti c anal ysi s of for-
mal i zed l anguages. The j usti fi cati on of certai n el ementary concepts i n
l ogi c, such as the concepts of l ogi cal consequence and the concept of
model , i s based on l ogi cal semanti cs.
Also inthe l anguage of the empi ri cal sci ences probl ems l i nked with
semi oti c occupy an i mportant posi ti on. I n sci ence we al ways wor k with
vari ous si gns and expressi ons whi ch must be i nterpreted. I n this con-
nexi on i t is essenti al to make a di sti ncti on between those procedures by
means of whi ch we make deci si ons on the obj ects of the provi nce under
i nvesti gati on (in thi s connexi on we somet i mes speak of onti c deci si on-
maki ng), and those procedures by means of whi ch we take deci si ons about
si gns and expressi ons f or med therefrom(inthis connexi on we commonl y
speak of semanti c deci si on-maki ng). I n the f ormer case the posi ti on i s
that the obj ects of the gi ven uni verse of di scourse can be, for exampl e,
Sign 105
descri bed, measured, expl ai ned, forecast, and the l i ke. In the l atter case
the posi ti on i s that the si gns and expressi ons made theref romcan be
expl i cated, i nterpreted, some expressi ons can be val i dated, conf i rmed, etc.
Cogni ti ve sci enti fi c acti vi ty, of course, i nvol ves procedures of both types,
mutual l y i nterl ocki ng and compl ement i ng each other.
Cont emporary semi oti cs has subj ected to a t horough study al so the
probl ems of the onti c commi t ment s of i ndi vi dual si gns and expressi ons
made up of t hem. The probl emwe are conf ronted wi th i s whether the use
of si gns or expressi ons of a certai n category whi ch represent the names of
assumed extral i ngui sti c obj ects does not commi t us to a recogni ti on of
the exi stence of these obj ects, or possi bl y inwhat context and under
what condi ti ons such onti c commi t ment s can be accepted. Wi t hout
taki ng i nto consi derati on the rel ati onshi ps subj ected to study i nsemi oti cs
we cannot achi eve the di sti ncti on between the onti c character of i ndi vi dual
extral i ngui sti c obj ects, and the semanti c character of the si gns and
expressi ons f ormed f romt hem. The probl ems of onti c commi t ment s in
formal i zed l anguages are essenti al to maki ng the deci si ons as to what
enti ti es are permi ssi bl e as val ues of free vari abl es.
For the l anguages of the empi ri cal sci ences great i mportance must be
attached to the study of the cri teri a for the meani ng of si gns and ex-
pressi ons, as well as a study of the met hods whi ch enabl e us to di sti ngui sh
meani ngl ess and meani ngf ul expressi ons. Whi l e we have not succeeded
inworki ng out a uni f orm, absol utel y val i d cri teri on of meani ng, bi ndi ng
for al l si gns and all expressi ons of the l anguage of the empi ri cal sci ences
(attempts to el aborate a uni f ormcri teri on for val i dati on, operati ons, or
other empi ri cal criteria havi ng fai l ed) - wi th respect to a gi ven si gn-
system, a gi ven uni verse of di scourse and a gi ven cl ass of tasks i t is
al ways mandat ory to defi ne rel ati vel y dependabl e cri teri a of the sense of
si gns and expressi ons.
REFERENCES
B ~ A L , M., (1904) Essai de smantique, science des significations, Paris.
CARNAP, R., (1942) Introduction to semantics, Studies in Semantics, Vol. I, Cambridge
CASSIRER, E., (1923-1931) Philosophie der symbolischen Formen, Berlin.
CHASE, S., (1938) The Tyranny of Words, New York, Harcourt, Brace and Company.
CHASE, S., (1954) The Power of Words, New York, Harcourt, Brace and Company,
(Mass.).
1 O6 Ladisluv Tondl
CHURCH, A. , (1956) Introduction to Mathematical Logic, Pri nceton (N.J.), Pri nceton
DRR, K. , (1954) Lehrbuch der Logistik, Basel -Stuttgart.
FODOR, J. A. and KATZ, J . J., (eds.) (1964) The Structure of Language, New Yor k,
FREGE, G. (1892) ber Si nn und Bedeutung , Zeitschrgt fr Philosophie und philo-
HAYAKAWA, S. I. (1941) Language in Action, New Yor k, Harcourt, Brace and Company.
HUSSERL, E. , (1913-1921) Logische Untersuchungen, Vol . I and II, Hal l e, M. Ni emeyer.
KLEENE, S. C. , (1950) Introduction to Meta-mathematics, New Yor k, Van Nost rand
KORZYBSK~, A., (1933) Science and Sanity, An Introduction to non-Aristotelian Systems
KOTARBI ~SKI , T., (1961) Elementy teorii poznania, logiki formalnej i metodologii nauk,
LOCKE, J., Essay concerning Human Understanding, Book I V, Chap. XXI.
MARKOV, A. A. , (1954) Teorija algoritmov, Moscow, T. M. I .
MARTI N, R. M. , (1958) Truth and Denotation. A Study of Semantical Theory, Chi cago
#
MORRIS, C. W. , (1938) Foundati ons of the theory of si gns, in International Ency-
MORRI S, C. W. , (1946) Signs, Language and Behaviour, NewYor k, Prenti ce-Hal l .
Mo ms , C. W. (1964) Signification and Significance, M. I . T. Press.
NARSKI J , I. S., (1969) Problema znaka i znatchneya Izd, Moscow, Moskovskogo
OGDEN, C. K. and RI CHARDS, I. A. , (1923) Tfze Meaning ofMeaning, K. Paul , Trench,
PEI RCE, C. S., (1931-1935) Collected Papers, 6 Vol . , C. Hart shorne and P. Wei ss
PEI RCE, C. S. (1940) The Theory of Signs, inJ. Buchl er (ed.), The Philosophy of C. S.
QUINE, W. V. O., (1953) From a Logical Point of View, Harvard Uni v. Press.
QUI NE, W. V. O. , (1960) Word and Object, Har var d Uni v. Press.
REI CHENBACH, H. , (1947) Elements of Symbolic Logic, New Yor k, Macmi l l an.
RUSSELL, B., (1905) On denoti ng , Mind, 14,479-493.
RUSSELL, B., (1910-1912-1913) Principia Mathematica, Vol . 1-111 (together wi th
RUSSELL, B., (1912) The Problems of Philosophy, New Yor k.
RUSSELL, B., (1940) An Inquiry into Meaning and Truth, New Yor k- London, Al l en and
RUSSELL, B., (1947) Human Knowledge, its Scope and Limits, London- New Yor k, Al l en
TONDL, L. , (1966) Problmy smantiky, Prague.
TMCHTENEI ROT, B. A. , (1960) Algorithmi i maschinnoe reschenye zadatcfi, Moscow.
Uni v. Press.
Prenti ceHal l .
sophische Kritik, 100,25-50.
Company.
and General Semantics, Lancaster, The Sci ence Press.
Wr cl awand War saw, Ossol i neum. (Fi rst edi ti on in1929.)
(Ill.), Uni v. of Chi cago Press.
clopaedia of Unified Science, I, 2, Chi cago (Ill.), Uni v. of Chi cago Press.
Uni versi teta.
Trubner Ltd.
(eds.), Har var d Uni v. Press.
Peirce, New York, Harcourt , Brace and Company.
Whi t ehead, A. N.), Cambr i dge (Mass.), Cambr i dge Univ. Press.
Unwi n.
and Unwi n.
6 Language
YEHOSHUA BAR-HILLEL
Duri ng the last fi fteenyears, si nce the appearance of NoamChomsky s
fi rst book (1957), concepti ons of l anguage have undergone a radi cal
change, so that some timecan reasonabl y be spent on bri ngi ng themi nto
f ocus. I nteresti ngl y enough, these revi sed concepti ons inpart del i beratel y
hark back to vi ews that wer e preval ent duri ng the rati onal i st Cartesi an
peri od of the seventeenth and ei ghteenth centuri es and mi ght be cal l ed
neo- Cartesi an Li ngui sti cs, paraphrasi ng the ti tl e of another mor e recent
book by Chomsky (1966), inwhi ch these connexi ons are cl earl y del i n-
eated.
I regard this Chomskyan revol uti on to be of such deci si ve i mportance
that I i ntend to devote most of the present survey to i ts descri pti on, ful l y
aware of the i nj usti ce that may thereby be caused to some, perhaps not
much l ess ori gi nal and frui tful , contri buti ons by others. But the space
al l otted to me is l i mi ted, and I therefore feel that my deci si on is
j usti fi abl e.
Let us then see what are the maj or tenets and consequences of thi s
revol uti on. I shal l not di scuss theminanythi ng likechronol ogi cal order,
nor rank t hemaccordi ng to thei r obj ecti ve i mportance (si nce I woul d not
know how to make such a ranki ng), but accordi ng to a scheme whi ch
mi ght l ook rather arbi trary but behi nd whi ch, I bel i eve, there is some
rati onal j usti fi cati on.
1. NATURAL LANGUAGES ARE SPECIES-SPECIFIC
The ways i n whi ch humans, ani mal s, and certai n artefacts such as
computers communi cat e wi th other member s of thei r speci es and, on
occasi on, even wi th member s of other speci es, are many and vari ed. It i s
of no parti cul ar i mportance for our present purposes to go at any depth
108 Yehoshua Bar-Hi l l el
i nto the termi nol ogi cal questi on as to whether the t erm communi cat i on
is or is not appropri ate for the many thi ngs that happen, say, bet ween the
pr ogr ammer and the comput er, though I will say a fewwor ds about this
questi on l ater.
However , what has become cl ear duri ng the last years i s that natural
l anguages, as we know them, are speci es-speci fi c and that to the best of
our empi ri cal knowl edge - and I stress empi ri cal - nothi ng of the ki nd
exi sts inbeast or machi ne. There i s no conti nui ty in any seri ous sense
bet ween the communi cat i on syst ems i nuse amongst themand the parti c-
ul ar communi cat i on systemthat rel i es essenti al l y on natural l anguages
i nuse among humans (see Chomsky, 1966, 1968; Lenneberg, 1964).
Thi s i mmedi atel y rai ses the age-ol d questi on of the ori gi n of human
l anguages, whi ch has probabl y gi ven ri se to no f ewer myths or l ess
specul ati on than any other of the questi ons wi th whi ch manki nd has
been grappl i ng si nce the dawn of hi story.
In vi ewof this anti qui ty, i t seems strange that so many seri ous thi nkers
have come to the concl usi on that i t mi ght sti l l be premature to take i t
up. They bel i eve that there are a l arge number of other questi ons to be
sol ved fi rst; and, i f these do not fi ndat l east a tentati ve sol uti on, there i s
l i ttl e chance of maki ng much progress i ntraci ng the ori gi n of the human
facult de lunguge, to use the wel l - known Saussuri an phrase.
Whi l e this probl emof the ori gi ns has l ost much of i ts appeal and i s
nowadays deal t withonl y i n popul ar or, at best, semi - popul ar publ i ca-
ti ons but hardl y at all inthe scientific l i terature, a rel ated probl emhas
come to the fore, vi z. that of the acqui si ti on of l anguage. I t, too, is ol d,
inei ther of i ts t wo mai n f orms: the acqui si ti on of a mother t ongue in
earl y chi l dhood, and the acqui si ti on of a second l anguage, at any age.
The second vari ati on is usual l y regarded as bei ng pri mari l y pedagogi cal
and, as such, has attracted countl ess studi es by l i ngui sts and educators,
The acqui si ti on of one s mot her tongue seems to be of deeper i mportance,
perhaps the deepest probl emof psychol i ngui sti cs al together, yet sti l l
si mpl e enough for al most anybody who has ever had chi l dren of hi s own
to rai se to be abl e to make useful and rel evant observati ons. I ndeed, the
l i terature abounds instori es tol dby psychol i ngui sts about thei r personal
experi ences with thei r chi l dren. (Noti ce that most of thi s materi al is
observati onal and anecdotal , si nce, for obvi ous reasons, one woul d
hardl y perf ormcontrol l ed experi ments inl anguage acqui si ti on wi th young
chi l dren.)
Language 109
Though there is no dearth of observati onal materi al on fi rst l anguage
acqui si ti on and of both observati onal and experi mental materi al on
second l anguage acqui si ti on, the theory, i n bot h cases, is still in its
i nfancy - infact, i t i s doubtf ul whether the t er m theory shoul d be used at
all for whatever i s avai l abl e. Thi s i s not meant to belittlethe wor k of such
pi oneers as Wi l hel mStern (1907), J ean Pi aget (1926), or Lev Semonovi ch
Vygotsky (1962), or the mor e recent wor k reported upon (wi th bi bl i og-
raphy) in Part One of R. C. Ol dfi el d and J . C. Marshal l (1968); rather
i t refl ects the obj ecti ve state of affai rs.
The maj or drawback seems agai n to l i e i n the fact that theoreti cal
l i ngui sti cs as such is sti l l at a very rudi mentary stage. Noti ce that the
fi rst textbook to carry the phrase theoreti cal l i ngui sti cs in i ts ti tl e
(J ohn Lyons, 1968) was publ i shed onl y four years ago; rel ated phrases
appear int wo earl i er books by NoamChomsky (1965a,b). I mysel f used
thi s phrase (Bar-Hi l l el , 1962).
The Chomskyan revol uti on accordi ngl y postul ates that a seri ous
treatment of the probl emof l anguage acqui si ti on and, in parti cul ar, fi rst
l anguage acqui si ti on, must be based on theoreti cal constructi ons f r oma
general theory of l anguage.
2. PRIORITY OF LINGUISTIC COMPETENCE OVER LINGUISTIC
PERFORMANCE
Acqui ri ng a l anguage is equated with acqui ri ng the knowl edge of a
l anguage, a certai n l i ngui sti c competence, i.e. masteri ng a parti cul ar
systemof rul es that rel ate sound and meani ng (to use the standard,
somewhat ol d-fashi oned termi nol ogy). Havi ng acqui red this competence,
an i deal speaker-hearer (it is perhaps not superfl uous to remark that inthe
current l i ngo, one means by the t er m speaker any producer of some
l i ngui sti c uni ts, whether by speech, wri ti ng, typi ng or by any other
appropri ate mode of communi cat i on; the same hol ds, of course, for
hearer) can express any meani ng he wi shes to communi cat e by an
appropri ate sequence of sounds (perhaps in mor e than one way) and
assi gn to sound sequences he has heard uttered wi thi n certai n contexts
the meani ng (or meani ngs) that is (are) usual l y expressed by such sequen-
ces (in these contexts), despi te the fact - and here l i es the crux of the
matter - that he mi ght never before have heard thi s parti cul ar sequence
110 Yehoshua Bur- Hi l l el
i n any context, nor have been in a posi ti on to want to express thi s
parti cul ar meani ng.
Real speakers-hearers (as opposed to i deal ones) have l i mi tati ons that
affect thei r l i ngui sti c perf ormance, both in communi cat i ng and in
understandi ng. Probabl y the most i mportant are the l i mi tati ons of the
short-termmemor y, whi ch, among other thi ngs, will affect the degree
of syntacti c compl exi ty of the sentences whi ch speakers can utter or
hearers can understand wi thout havi ng to guess. The producti on or
percepti on of sentences of great syntacti c compl exi ty requi res memor y
ai ds that are not normal l y avai l abl e. However , these sentences sti l l bel ong
to the l anguage known by any speaker of thi s l anguage; he knows, in
pri nci pl e, how to deal witht hem, and can do so successful l y inpracti ce as
soon as he manages to extend his l i mi ted short-termmemor y by some
artificial devi ce.
Other factors that i mpai r success incommuni cati ng or understandi ng
are, for i nstance, defects i n the arti cul atory or audi tory mechani sms,
or i n the central nervous system, but al so such si mpl e and frequent
factors as gaps inthe background knowl edge of speaker and hearer or
even l ack of good- wi l l on the part of the hearer. A speaker, when ad-
dressi ng a certai n hearer, will often expect hi mto use his background
knowl edge and his comprehensi on of the context as well as of the
l i ngui sti c Co-text to understand the uttered sentence in a uni que way,
to di sambi guate i t, if necessary, on thi s basi s, and this inspi te of the fact
that he mi ght be ful l y aware that the sentence as such is syntacti cal l y and/
or semanti cal l y ambi guous. The expected di sambi guati on mi ght fail,
however, for the si mpl e reason that the presupposed background
knowl edge is j ust not there or that the hearer refuses, for whatever reason,
to use the rel evant knowl edge inhi s possessi on for thi s purpose. Thi s is
onl y one i l l ustrati on of the countl ess si tuati ons inwhi ch communi cat i on
will fail because the actual perf ormance will fall short of what coul d have
been achi eved wi th the avai l abl e competence.
Though all this may sound rather si mpl e, even trite, i t shoul d be
stressed that nei ther rhetori cs nor the theory of argumentati on, nor thei r
moder n counterparts, psychol i ngui sti cs, soci ol i ngui sti cs and com-
muni cati on theory, the sci ences that shoul d study, among other thi ngs,
the condi ti ons under whi ch communi cat i on will succeed or fail, have been
abl e to devel op a real l y sati sfactory theory, inspi te of countl ess sporadi c
observati ons and occasi onal l aboratory experi ments. (Let me call the
Language 1 11
reader s attenti on, al most at r andom, to the fol l owi ng rel evant publ i ca-
ti ons: Perel man and Ol brechts-Tyteca (1958), Saporta (1961), Bri ght
(1966), Gumper t z and Hymes (1969) and Cherry (1957).)
Thi s i s usual l y expl ai ned, at l east inpart, by sayi ng that these obser-
vati ons and experi ments had no underl yi ng sound theory of l i ngui sti c
competence to back t hemup. Now that we seemto be well on the way
towards such a theory, i t is perhaps no l onger premature to expect
greater progress inthe foreseeabl e future.
3. SYNTAX - THE LINK BETWEEN SOUND AND MEANING
Li ngui sti cs i s often di vi ded i nto gr ammar and semanti cs, and gr ammar
i s subdi vi ded i nto phonol ogy and syntax. Among these di sci pl i nes,
phonol ogy, the study of l i ngui sti c sound, al ready has a mor e or l ess
sati sfactory theory, t hough di scussi ons between the proponents of the
vari ous competi ng phonol ogi cal theori es will doubtl ess conti nue for a
l ong timeto come ( Chomsky and Hal l e, 1968). Semanti cs, the study of
meani ng, is sti l l i ni ts very earl i est theoreti cal stages (sernanti cs bei ng a
catch-al l term, i t mi ght be useful to di sti ngui sh between l i ngui sti c
semanti cs, l ogi cal semanti cs and phi l osophi cal semanti cs: for the fi rst
di sci pl i ne, see Ul l man (1962), Kat z (1966); for the second, see Tarski
(1956), Carnap (1946), Hi nti kka (1969); for the thi rd, see Schi l pp (1963),
Wi ttgenstei n (1953), Ziff (1960), Qui ne (1960), Kat z (1966), and Bar-Hi l l el
(1970b)), despi te t wo thousand years of study and, on occasi on, consi der-
abl e sophi sti cati on. (For an excel l ent short hi story of semanti cs, see the
arti cl e Semanti cs, Hi story of in The Encyclopaedia of Philosophy, VI I ,
358-406. ) I t i s in the theoreti zati on of syntax, the systemof rul es that
provi de a bri dge between the t wo, that the Chomskyan school has made
i ts most si gni ti cant contri buti on so far, one that amount s to a real
breakthrough.
The study of syntacti c structure as such is as ol d as the study of
l anguage in general , but onl y very recentl y has i t reached a degree of
ri gour and preci si on comparabl e to that of the better-devel oped natural
sci ences. Formul ati ons such as the one whi ch charges the syntax of a
l anguage wi th the task of generati ng all and onl y the sentences of this
l anguage as well as of assi gni ng t hemthe i ntui ti vel y correct structure(s)
attracti ve as this f ormul ati on may sound, if onl y we are not fri ghtened
1 12 Yehoshua Bar- Hi l l el
away by the hyper- modern term generate (this t ermhas been taken over
f r oman admi ttedl y rather obstruse subfi el d of mathemati cal l ogi c, but i n
our context i t has neverthel ess an extremel y si mpl e meani ng) borrowed
f r ommathemati cal l ogi c, are of very recent ori gi n, but they tend to dri ve
home the fact that syntax has, wi thi n a very short peri od, attai ned a degree
of theoreti zati on whi ch other humani sti c studi es are sti l l worki ng hard to
achi eve. It has al ready reached the stage where one feel s onesel f enti tl ed
to speak of mathemati cal l i ngui sti cs. I nci dental l y, thi s t ermis not wi thout
i ts dangers and i s i ndeed often mi sunderstood. Most peopl e, i ncl udi ng
some l i ngui sts, when they hear this t er m, al most automati cal l y thi nk of
appl i cati on of stati sti ca1met hods to l i ngui sti c materi al , of f requency
counts and the l i ke. Stati sti cs, however, appl i es pri mari l y to speech and
not to l anguage, t hough the possi bi l i ty of appl yi ng advanced stati sti cal
met hods certai nl y endows t he study of speech witha good amount of
wel l -deserved respectabi l i ty. But thi s possi bi l i ty i s not suffi ci ent by i tsel f
to turn l i ngui sti cs i nto a theoreti cal sci ence. This deci si ve devel opment
happens onl y when a degree of formal i zati on has been achi eved suffi ci ent
for al gebrai c met hods to become appl i cabl e. It is therefore onl y the ri se
of al gebrai c l i ngui sti cs, the use of l ogi co-al gebrai c methods i n the study
and descri pti on of l anguage, whi ch serves testi mony to the fact that
l i ngui sti cs has come of age and i s now compet i ng withthe best of the
theoreti cal sci ences for a pl ace in the sun. (For some earl y papers in
obj ecti ve l i ngui sti cs, see Luce, Bush and Gal anter, 1965, Part I I , Vol. I I ,
and Bar-Hi l l el , 1964, Part 11.)
No w that gr ammar s are f ormul ated, to an i ncreasi ng degree, in a way
that al l ows for a ri gorous eval uati on of thei r strength and compl exi ty, i t
is possi bl e, for the fi rst ti me i n the hi story of l i ngui sti cs, to seri ousl y
compar e competi ng gr ammar s, and to provi de somethi ng likeproof s for
the i ntui ti ve i nadequacy of certai n gr ammar s, and even of whol e cl asses
of gr ammar s. Onl y now, when the so-cal l ed i mmedi ate-consti tuent
gr ammar s, regarded by many l i ngui sts of the l ast generati on as the most
adequate types of gr ammar s for the descri pti on of natural l anguages
(Lyons, 1968, p. 209f), can:be ri gorousl y f ormal i zed i nthe f or mof phrase-
structure gr ammar s (p. 215f) or, al ternati vel y, of categori al gr ammar s
(p. 227f), can one di scuss the questi on of thei r adequacy withsuffi ci ent
ri gour. Thi s ri gour forced and enabl ed the l i ngui sts todi sti ngui sh between
weak and strong adequacy, where a gr ammar is cal l ed weakl y adequate
when i t generates all and onl y the i ntui ti vel y wel l - f ormed sentences of the
Language 113
l anguage in questi on, but strongl y adequate, if, in addi ti on, i t assi gns
to each of the generated sentences i ts appropri ate siructure(s). It is onl y
now that one can at last cl ai mto be abl e to prove that certai n types of
gr ammar s, such as the so-cal l ed context-free phrase-structure gr ammar s,
are i nadequate for the generati on of all natural l anguages (and i t i s
i rrel evant, for our purposes, whether one can say that these cl ai ms have
been defi ni tel y substanti ated). Many l i ngui sts are convi nced that such
gr ammar s are i nadequate for assi gni ng to all sentences their correct
structures and i nsuffi ci ent to serve as a basi s for establ i shi ng the vari ous
semanti c rel ati onshi ps that i ntui ti vel y hol d bet ween sentences of super-
ficially qui te di fferent f orms such as
(1) J ohn l oves Mar y
and
(2) Mar y is l oved by J ohn
whose semanti c rel ati on i s cl earl y a very strong one i ndeed, vi z. nothi ng
l ess than that of expressi ng the same cogni ti ve content, while assi gni ng
the same syntacti c structure to sentences whose i ntui ti ve semanti c struc-
ture i s, insome i mportant sense, qui te di fferent, such as
(3) J ohn is eager to pl ease
and
(4) J ohn is easy to pl ease
thereby l eavi ng vitally i mportant semanti c rel ati ons unexpl ai ned in the
fi rst case, whilei nti mati ng spuri ous semanti c rel ati ons, i nthe second case.
These and si mi l ar facts made many l i ngui sts real i ze that generati ve
gr ammar s of the phrase-structure type, i n spi te of thei r superi ori ty in
ri gour and degree of formal i zati on over the ol der i mmedi ate- consti tuent
gr ammar s, or perhaps j ust because of this superi ori ty whi ch made possi bl e
a ri gorous treatment of i ssues that before coul d be di scussed onl y on an
i nf ormal , non- commi ttal l evel , are strongl y i nadequate, t hough some
l i ngui sts may regard t hemas bei ng weakl y i nadequate.
4. DEEP STRUCTURE AND SURFACE STRUCTURE
The parti cul ar way in whi ch the Chomskyan school f ormul ated the
shortcomi ngs of phrase structure grammars in expl ai ni ng the exact
rel ati onshi p between the t wo pai rs of sentences menti oned inthe prece-
di ng secti on turned out to be extremel y hel pful , and mi ght well be
1 14 Yrhoshua Bar-Hillel
Surface strarctiire of (3)
Sentence
Nomi nal / \ \ Phrase Verbal Phrase
Noun I
Verb / \ Nominal Phrase
/\ Verb-infinitive
Adjective
/I
to please
I
John is easy
Surface structure of (4)
I
N o u n
/
Verb
John is eager
regarded as the most i mportant speci fi c contri buti on so far of this school
to l i ngui sti cs al together. The f ormul ati on chosen was to say that the
sentences (1) and (2), though exhi bi ti ng qui te di fferent surface (or super-
ficial) structures - noti ce, e.g., that they have di fferent subj ects, that the
verb in the one sentence i s in the acti ve mood and in the other in the
passi ve mood - have the same deep (or depth) structure (or, perhaps,
onl y very cl osel y rel ated deep structures; i t woul d take us too far to go
here i nto the fi ne, t hough hi ghl y i nteresti ng, details), whi l e the opposi te is
the case for the other pai r of sentences: (3) and (4) have the same surface
Language 115
Deep structure of (3)
Nomi
N
JO
1 .v
Adj
e
structure (up to the l exi cal items), and thi s even for the most detai l ed
parsi ng , but very di fferent deep structures.
J ust for the sake of i l l ustrati on and for whatever i ts worth in the
absence of detai l ed expl anati ons for whi ch we have no space, l et me
present here one of the most often used met hods of depi cti ng the surface
and deep structures of the sentences (3) and (4.), the so-cal l ed tree f orm
(see above).
The transi ti on f r omthe deep structure to the surface structure i s per-
f or med by so-cal l ed rul es of transf ormati on, a t er mChomsky took over
f r omhis teacher Harri s (1957), t hough not wi thout consi derabl e changes
in i ts meani ng, and whi ch shoul d be careful l y di sti ngui shed f r omthe
homonymous one that pl ayed such an i mportant rol e i n Carnap s
1 16 Yehoshuu Bur-Hillel
Noun2
Deep strirctirre of (4)
Sentence
Nomi nai Phrasel
N
JI
Logical Syntax (1937), where i t meant what he l ater cal l ed rul es of
deducti on. This i s not the onl y uncti on of the rul es of transformati on but
the most i mportant one. The t er ms deep structure and surface structure
themsel ves are al so not ori gi nal withChomsky si nce they had been used
previ ousl y, wi th approxi matel y the same meani ng, though not withthe
same degree of ri gour, by the phi l osopher L. Wi ttgenstei n (1922) and the
l i ngui st C. Hockett (1958; the title of Chap. 29 is Surface and deep
gr ammar ) .
The concepti ons of deep structure , surface structure and rul es of
transf ormati on have l atel y undergone consi derabl e changes, somet i mes
of a very f undamental character. Si nce these changes are sti l l very much i n
process and i t is by no means cl ear at present i n whi ch di recti on they
Language 117
are goi ng, I shal l conti nue presenti ng here the cl assi cal concepti on of
thei r functi on. Accordi ng to thi s concepti on, i t i s the deep structure of
a gi ven sentence (or i ts vari ous deep structures, i n case the sentence i s
syntacti cal l y ambi guous) whi ch recei ves i ts meani ng by the rul es of
semanti c i nterpretati on, whi l e the surface structure recei ves i ts phoneti c
i nterpretati on through phonol ogi cal rul es. The functi on of the trans-
f ormati on rul es i s, as sai d before, to map deep structures onto surface
structures. Accordi ng to the generati ve-transf ormati onal approach, a
person who kiiows a speci fi c l anguage has control of a gr ammar that
generates (or characteri zes) the infiniteset of potenti al deep structures,
maps t hem(by appropri ate rul es of transformati on) onto associ ated
surface structures and determi nes the semanti c and phoneti c i nterpreta-
ti ons of these abstract structures by the appropri ate semanti c and pho-
neti c rul es, respecti vel y.
Let i t be sai d, once agai n, that the t er m generate carri es no psychol og-
i cal connotati ons. Generati ve grammars do not tel l us how speakers
produce the sentences they utter on some gi ven occasi on nor why they
produce these utterances and not others. Generate i s not a psycho-
l i ngui sti c or soci ol i ngui sti c t erm, but a purel y theoreti cal l i ngui sti c one;
i t has nothi ng to do withperf ormance but bel ongs whol l y with compe-
tence.
5. LINGUISTICS - A BRANCH OF COGNITIVE PSYCHOLOGY?
Probabl y the most dari ng and therefore al so most controversi al thesi s
of the Chomskyan school i s the cl ai mthat bot h the theory of l i ngui sti c
perf ormance and the theory of l i ngui sti c compet ence, that i s, l i ngui sti c
theory i ngeneral , consti tute a branch of psychol ogy, and mor e speci fi cal l y,
of cogni ti ve psychol ogy ( Chomsky, 1968, p. 1 ) .
Such bl unt f ormul ati ons, needl ess to say, tread on many peopl e s toes,
thus i nfri nge on many vested i nterests and therefore al i enate many
l i ngui sts for whomthe aut onomy and i ndependence of l i ngui sti cs is a
matter of deep convi cti on. It i s i mportant, however, to real i ze that the
Chomskyan cl ai ms have nothi ng incommon withbut, on the contrary,
stand in the strongest possi bl e opposi ti on to, other cl ai ms whi ch mi ght
sound superfi ci al l y si mi l ar, namel y cl ai ms to the effect that l i ngui sti cs is
reduci bl e to behavi ori sti c S-R (Sti mul us- Response) psychol ogy, and
1 18 Yehoshuu Bur- Hi l l el
whi ch i denti fy l i ngui sti cs wi th the study of verbal behavi our. Such
cl ai ms were made most forceful l y in B.F. Ski nner (1957) and equal l y
forceful l y cri ti ci zed in a revi ewby Noam Chomsky (1959). Chomsky
hi msel f is strongl y opposed to any concepti on that sees l anguage
as a habi t structure , a network of associ ati ve connecti ons, a skill
expressi bl e as a systemof di sposi ti ons to respond ( Chomsky, 1968,
p. 22). He prefers to expl ore the possi bi l i ti es for devel opi ng expl anatory
theori es, and thei r i mpl i cati ons with regard to the compl exi ty and
abstractness of the underl yi ng mechani sms. For hi m, the search for
expl anatory theori es must begi n withan attempt todetermi ne the systems
of rul es that consti tute knowl edge of a l anguage and f or ma correct
gr ammar of that l anguage. As soon as one consi ders the probl emof the
conf i rmati on of proposed gran-i mars on the basi s of empi ri cal evi dence,
the actual perf ormance of the speakers-hearers pl ays, of course, a deci si ve
rol e. However , Chomsky regards i t as a vital methodol ogi cal pri nci pl e
that si nce actual speech behavi our i s the resul t of the i nteracti on of several
factors, i t i s necessary, for an understandi ng of the preci se character of
thi s i nteracti on, that each of these separate factors be fi rst fairly well
understood by itself.
Chomsky is ready to go one step further and l ook for expl anatory
theori es of a deeper sort. When a nati ve speaker acqui res inhi s youth
the gr ammar of hi s mot her tongue, he has mastered a certai n expl anatory
theory, on the basi s of very restri cted and degenerate data, i.e. f rom
whatever he mi ght have heard f rommember s of hi s fami l y and pl aymates,
f r omcorrecti ons of his own babbl i ngs by adul ts and the rei nforcements
of hi s f3rst successes at verbal communi cat i on, etc. But the gr ammar he
constructs for hi msel f goes far beyond this restri ctedevi dence. There must
be i nnate inhi mcertai n pri nci pl es that successful l y gui de hi mindetermi -
ni ng the f or mof the gr ammar he arri ves at i nsuch a short ti me. Chomsky
proposes to cal l the set of these pri nci pl es uni versal gr ammar (1968,
p. 57, n. 30), f ol l owi ng what he bel i eves to be tradi ti onal usage. ( Though
he is probabl y wr ong on thi s hi stori cal count, thi s is a mi nor i ssue i nour
context. ) Now the questi on at stake is whether the study of thi s uni versal
gr ammar can i ndeed be i nterpreted as part of the study of the structure
of human i ntel l ectual capaci ti es. For if we grant this - and we are sti l l a
l ong way f r omi t, the arguments presented by Chomsky inf avour of this
vi ewbei ng so far not suffi ci ent toconvi nce all doubters - t hen the study of
the structure of gr ammar s of natural l anguages as well as the expl orati ons
Language 119
of the vari ous systems of meta- rul es that expl ai n how the chi l d arri ves
at a correct gr ammar of hi s nati ve l anguage shoul d l ead us to a better
understandi ng of cogni ti on in general . Chomsky i s ful l y aware of the
fact that, as of today, any suggesti ons one can make al ong these l i nes are
tentati ve and restri cted inthei r coverage. J udgi ng f r omthe short hi story
of these attempts, i t seems fair to predi ct that no presentl y exi sti ng
theory will survi ve for even a fewyears. But Chomsky and hi s fol l owers
are ready ta sti ck thei r necks out and are convi nced that the sci ence of
l i ngui sti cs, as any other sci ence, will advance onl y by conj ectures and
refutati ons, to i nvoke Popper s memorabl e phrase (1962).
Si nce the revi val of the t er m uni versal gr ammar has created a lot of
mi sunderstandi ngs, l et me stress here that inthis new, somewhat i di osyn-
crati c usage, uni versal gr ammar is pri mari l y a set of condi ti ons that any
human l anguage must meet , al most a transcendental gr ammar in a
quasi - Kanti an sense, and onl y secondari l y a skel etal substructure of rul es
that any such l anguage must contai n. I tend to bel i eve that Chomsky is
here underesti mati ng the degree of val i di ty of the second mor e tradi ti onal
vi ew; on one occasi on (Bar-Hi l l el , 1969), I mysel f proposed a vi ew
accordi ng towhi ch l ogi c shoul d be i denti fi ed wi th uni versal semanti cs and
therefore wi th part of uni versal l i ngui sti cs, in the secondary sense,
though there i s no difficulty in rei nterpreti ng this i denti fi cati on in the
pri mary sense, perhaps even mor e hel pful l y so.
6. THE CHOMSKYAN APPROACH VERSUS THE SEMIOTIC APPROACH
The Chomskyan di chotomy, theory of competence vs. theory of per-
f ormance, i s rel ated to the wel l - known Saussuri an di sti ncti on between
langue and parole (de Saussure, 1955) and to di sti ncti ons dr awn by others
between l anguage and speech, t hough i t woul d take us too far to go i nto
the communal i ti es and di versi ti es of the vari ous concepti ons.
However, l et me poi nt out that thi s viewdi verts, on purpose, f r om
another cl assi fi cati on whi ch harks back to Pei rce (1931-1935) and had
been revi ved inour century by Morri s (1938) and Car nap (1946). Pei rce
di vi ded semi oti c, the general theory of si gns, i nto three l ayers of i ncreasi ng
abstracti on. The l owest l evel , inwhi ch not onl y the rel ati ons between the
si gns among themsel ves and between the si gns and the thi ngs denoted by
these si gns inthe outsi de worl d are consi dered, but al so the uses of these
120 Yehoshuu Bar-Hi l l el
si gns inthe vari ous communi cati ve acts are expl i ci tl y taken i nto account,
he cal l ed pragmati cs. At the next l evel of abstracti on inwhi ch the uses
of the si gns are l eft out of consi derati on, one moves up i nto semanti cs.
Fi nal l y, when even the rel ati ons between the si gns and the worl d are put
asi de, one arri ves i nsyntax (or syntacti cs, as i t was someti mes cal l ed in
thi s connecti on). (It shoul d be noti ced that the t erms pragmati cs , seman-
tics and syntax are someti mes used i n an excl usi ve sense, so that prag-
mati cs i s understood to deal onl y withthe rel ati ons of the si gns tothe si gn
users, semanti cs onl y wi th the rel ati ons of the si gns to the thi ngs si gni fi ed,
and syntax onl y withthe rel ati ons of the si gns among themsel ves. The
term semi oti c i s then used in the sense whi ch pragmati cs has i n i ts
i ncl usi ve usage.)
Li ngui sti cs was consi dered by Pei rce and hi s fol l owers to be the
theory of a certai n speci fi c subcl ass of si gns, the l i ngui sti c si gns, or
symbol s, and the tri parti te di vi si on i nto pragmati cs, semanti cs, and
syntax, was carri ed over to it.
Accordi ng to thi s scheme, then, semanti cs and syntax are arri ved at
f r ompragmati cs by successi ve abstracti ons. Car nap, however, drew
the extremel y i mportant di sti ncti on between descri pti ve semanti cs
and syntax, on the one hand, and pure semanti cs and syntax, on the other,
stressi ng that whi l e descri pti ve semanti cs and syntax are i ndeed obtai ned
by abstracti on f r om(descri pti ve) pragmati cs, so that inthis sense prag-
mati cs i s the basi s for all l i ngui sti cs, this i s not the case for pure semanti cs
and syntax whi ch are i ndependent of pragmati cs (Carnap, 1946, s 5).
It i s i nteresti ng to note that, f r omthe very begi nni ng, the Chomskyan
school , t hough usi ng freel y, as everybody el se di d, the semanti cs-syntax
termi nol ogy (though wonderi ng, on occasi on, where to dr awthe border-
l i ne between the t wo and someti mes even doubti ng whether such a
borderl i ne can be useful l y dr awn at all; see, e.g., NoamChomsky (1965b,
Chap. 4, Secti on 1; the i ssue i s still very much i n acti ve di scussi on),
shi ed away f r omusi ng the t er m pragmati cs and someti mes cl ai med in
effect that the study of the i mpact of the soci o-physi cal setti ng (the t er m
they used on occasi on - Kat z and Fodor, 1963 - as an approxi mate
counterpart of pragmati c aspects) shoul d be left out of l i ngui sti cs
al together, the mai n reason for thi s deci si on bei ng that a compl ete theory
of the i mpact of the soci o-physi cal setti ng is out of reach (remai ni ng
somewhat vague about the questi on whether this i s so at present or
whether this will remai n so as a matter of principle). They were qui te
Language 121
happy to l et anybody i nterested pi ck up these l eft-overs f r oma compl ete
l i ngui sti c descri pti on, whi ch i deal l y contai ns exactl y three component s,
phonol ogi cal , syntacti c and semanti c, perhaps hopi ng that psychol i ngui sts,
soci ol i ngui sts, ethnol i ngui sts, communi cat i on theoreti ci ans, and thei r
l i kes woul d decl are i nterest inthis mess.
I do not thi nk that we can possi bl y l eave i t at that. The rel ati on
between l i ngui sti c theory and speech behavi or must surel y be a much
cl oser one, and an efficient di vi si on of l abor (and there is nothi ng wr ong
wi th di vi si on of l abor as such) shoul d run on di fferent l i nes. As a matter
of fact, by combi ni ng some of the vi ews of Carnap and Chomsky,
al ready descri bed, wi th some new i deas of Hi nti kka (1968), i t seems that
a much mor e sati sfactory pi cture can be arri ved at. First, l et us recal l that
Carnap s use of the t er m pure preceded hi s deci si ve re-eval uati on of the
status of theoreti cal concepts and of theori es ingeneral . Let us, then, i n
accordance wi th thi s re-eval uati on, tal k about theoreti cal semanti cs
and syntax, or better, about semanti c and syntacti c theory, inpreference
to pure semanti cs and syntax, or, better still, about l i ngui sti c theory ,
wi thout prej udgi ng the i ssue whether some cl assi kati on of the rul es of
such a theory i nto phonol ogi cal , syntacti c, and semanti c rul es will turn
out to be appropri ate and useful . Whoever has acqui red, mastered, i nter-
nal i zed, the rul es of the l i ngui sti c theory of some natural l anguage, has
become competent inthat l anguage. (There i s, of course, no reason not to
tal k about degrees of compet ence j ust as we have no qual ms tal ki ng
about degrees of mastery of some l anguage. ) The actual speech behavi or
of a gi ven speaker will then be regarded as the resul t of the i nteracti on of
his l i ngui sti c compet ence wi th hi s i ntenti ons and with vari ous other
factors and pressures, soci al or i ndi vi dual , general (such as the al ready
ment i oned smal l size of the human short- ti me memor y) or speci al
(such as pressures resul ti ng f romthe parti cul ar si tuati onal context of
utterance).
If we adopt thi s vi ew, we can expl ai n the way by whi ch l i ngui sti c
theory may t hrowl i ght on what happens i nordi nary di scourse as f ol l ows:
the meani ng whi ch an expressi on recei ves in the expl anatory model
provi ded by the theory i s i ts basi c meani ng. The actual meani ng of an
utterance of thi s expressi on i s obtai ned t hrough modi fi cati on of i ts basi c
meani ng by the vari ous menti oned factors. I n pri nci pl e, i t coul d well
happen that for some expressi on the actual meani ng of all i ts utterances
i s never the same as its basi c meani ng, because for this parti cul ar ex-
122 Yehoslzua Bar-Hillel
pressi on the soci al pressures are al ways so hi gh as tocompl etel y mask the
basi c meani ng.
Looki ng at i t f r omthi s poi nt of vi ew, certai n wel l - known vexi ng
probl ems appear ina newl i ght. Though there i s nothi ng wr ong insayi ng
that the basi c meani ng of a symbol for (i ncl usi ve) di sj uncti on is such- and-
such, i t i s defi ni tel y concei vabl e that the Engl i sh wor d or7 will sel dom
have this meani ng in an actual utterance. To general i ze, what f ormal
l ogi c does, under thi s concepti on, i s to bri ng out the basi c (or depth)
meani ng of the so-cal l ed l ogi cal terms. For mal l ogi c turns out to be the
dept h l ogi c of ordi nary di scourse. Usi ng f ormal l ogi c does not entai l
modi f yi ng or regi menti ng ordi nary speech but i s rather an effort to
understand i t mor e deepl y and ful l y. Logi c, qua theoreti cal , expl anatory
model , i s not di scovered but i nvented. I ts rul es are used as nor ms with
whi ch to compare and by whi ch to eval uate actual di scourse as to i ts
l ogi cal force, inparti cul ar by whi ch to j udge the val i di ty of arguments
presented innatural l anguage. Looki ng at the rul es of l ogi c as nor ms
does not, of course, turn l ogi c i nto a systemof conventi ons, nor i nto a
systemof prescri pti ons. It is not the task of the l ogi ci an to l egi sl ate about
val i di ty of arguments, j ust as i t is not the task of the grammari an to
l egi sl ate about wel l - f ormedness of l i ngui sti c expressi ons or thei r
acceptabi l i ty . Theoreti cal gr ammar does not oppose descri pti ve gr am-
mar as bei ng prescri pti ve, but contrasts wi th i t as bei ng expl anatory,
j ust as theoreti cal physi cs stands inno opposi ti on to descri pti ve physi cs
but i s i ts expl anatory suppl ement whi ch no physi ci st woul d want to do
wi thout. The whol e prescri pti vi sm i ssue di sappears inthi n air as soon
as the theoreti cal character of a gr ammar i s recogni zed for what i t i s
( Chomsky, 1968, p. 13), and so do the equal l y spuri ous compl ai nts about
the fi cti ti ousness of some of the new gr ammar s that go beyond descri p-
ti on (Bar-Hi l l el , 1966).
7. LANGUAGE, DIALECT, IDIOLECT; COLLOQUIAL A N D SCIENTIFIC
LANGUAGE
Fewi ssues inthe study of l anguage have rai sed so many controversi es -
and, to many onl ookers, rather futileones - as those centered around the
di sti ncti ons between l anguage, di al ect and i di ol ect (Hockett, 1958,
p. 321f). Everybody knows i n his heart that some di sti ncti on must be
Lunguage 123
dr awn between the systemof l i ngui sti c rul es that govern the speech
behavi or of a gi ven i ndi vi dual and the systems of l i ngui sti c rul es that
govern the speech behavi our of the vari ous groups to whi ch this i ndi vi dual
bel ongs and that i n part turn these groups i nto l i ngui sti c communi ti es. It
i s onl y when one begi ns to theori ze about these wel l - known facts that one
qui ckl y runs i nto seemi ngl y unendi ng di fferences of opi ni on and f or m-
ul ati on.
I have no i ntenti on to enter these di scussi ons here. The onl y comment I
woul d like to make is to insist that these di fferences of opi ni on are no
mor e severe than one shoul d expect and pose no parti cul ar met hodol og-
i cal probl ems the l i ke of whi ch do not occur i nmany other si tuati ons.
It is to be expected that the systemof l i ngui sti c rul es a person has
acqui red at a gi ven age will be somewhat di fferent f r omthat of any other
person wi th whomhe comes i n communi cati ve contact. Neverthel ess,
these persons will often understand each other to vari ous degrees, and
someti mes perfectl y, withthe mi sunderstandi ngs that woul d sti l l ari se to
be traced excl usi vel y to extra-l i ngui sti c causes. Each i ndi vi dual systemis
an i di ol ect, each common systemis cal l ed a di al ect or a l anguage, depend-
i ng on vari ous other factors i nto whi ch i t woul d be too compl i cated togo
here. As usual wi th such theoreti cal t erms, there will be di sputes about
boundari es, whose acri moni ousness has nothi ng to do wi th any speci al
methodol ogi cal ni ceti es but rather withthe fact that certai n di al ects tend
to acqui re, for whatever hi stori cal reasons, certai n presti ge val ues whi ch
are absent f r omothers, thereby i ntroduci ng i ssues of prescri pti veness
whi ch have nothi ng to do wi th the sci enti fi c status of the probl em.
One mi ght al so want to depl ore the wi despread cust omof tal ki ng
about sci entSc l anguage , i.e. the l anguage used by sci enti sts, inopposi -
ti on to col l oqui al (or everyday) l anguage , i.e. the l anguage used by the
man-i n-the-street, presumabl y i ncl udi ng the sci enti sts themsel ves, when
off duty . This customi s potenti al l y mi sl eadi ng, t hough I am not sure
whether any actual har mis often caused by it. The systemof l i ngui sti c
rul es whi ch governs a sci enti sts speech when he tal ks (or wri tes) shop i s
inno seri ous sense di fferent f romthe one that governs his speech when he
tal ks about everyday affai rs. The di fference certai nl y does not liei nthat
these modes of speech are governed by di fferent rul es but that i n the
scientific ni ode of speech, the sci enti st uses addi ti onal theoreti cal t er ms
(and someti mes uses ordi nary t er ms ina speci al , theoreti cal sense) whose
meani ngs are theory- dependent, and hence ful l y comprehensi bl e onl y
124 Yehoshua Bar-Hillel
to those who master the rel evant theory, but the axi oms and rul es of
i nterpretati on of such a theory shoul d be careful l y di sti ngui shed f r omthe
rul es of the l anguage i nwhi ch thi s theory i s embedded. So, why not use
sci enti fi c (or theoreti cal ) mode of speech rather than sci enti fi c l anguage ,
and, if necessary, col l oqui al (or everyday) mode of speech rather than
col l oqui al l anguage .
8. CYBERNETICS AND LINGUISTICS
Fr omi ts very begi nni ng, cyberneti cs, as al ready i ndi cated i n the sub-
ti tl e of Norbert Wi ener s pi oneeri ng book (1948), was vitally i nterested
in natural l anguages, as the mai n vehi cl e of communi cat i on between
humans as well as incomput er l anguages as the mai n vehi cl e of control
whi ch humans exerci se over computers. Shortl y after the publ i cati on of
thi s book, another cl assi c deal i ng wi th the mathemati cal theory of
communi cat i on ( Shannon and Weaver, 1949) was publ i shed, and there
was at the timean ubi qui tous feel i ng that wi th these t wo works a new
era had begun and that a breakthrough had been achi eved towards a ful l
understandi ng of the compl exi ti es of communi cat i on i nthe ani mal and
the machi ne and, i n parti cul ar, of human l anguage. Among the many
appl i cati ons of thi s breakthrough, and as a resul t of a better under-
standi ng of the i ntri nsi c properti es of those marvel ous symbol - mani pu-
l ati ng mechani sms, the el ectroni c di gi tal computers, that expressed itself
inthe credo, i f a human bei ng can do it, a sui tabl y pr ogr ammed com-
puter can do i t too , one expected to make qui ck progress towards the ful l
automati on of hi gh-qual i ty transl ati on, of document retri eval , of
i nf ormati on retri eval , and even of deri vi ng gr ammar s f r omfi ni te corpuses.
Certai n i ni ti al successes in all these di recti ons wer e i ndeed qui ckl y
achi eved, but one tended greatl y to overesti mate the reach of these
l i mi ted achi evements and to bel i eve, or rather to hope, that the short-
comi ngs woul d be overcome through an i ncrease of the l earni ng abi l i ti es
of sui tabl y pr ogr ammed comput ers by the bootstrappi ng techni ques of
what became soon known as artificial i ntel l i gence and sel f-organi za-
ti on.
The expected breakthrough never arri ved. It is vi tal , at this poi nt, to
shed all i l l usi ons inthis respect. The hi gh expectati ons wer e based inpart
on a deep mi sunderstandi ng of the rel ati on between l anguage and
Languuge 125
speech, as a resul t of whi ch one fai l edto real i ze that the revol uti onary
i nsi ghts of cyberneti cs and the powerf ul techni ques of i nf ormati on
theory, t hough i ndeed of deci si ve i mportance to the study of speech, wer e
i rrel evant to the study of l anguage. I n l anguage, there i s no pl ace (beyond
trivialities) for probabi l i ty and statistics. Computati onal l i ngui sti cs,
unl ess i t i s based on theoreti cal l i ngui sti cs, i s, and inall l i kel i hood will
remai n, of l i mi ted i mpact. But the rol e of the computer, and of cyberneti c
types of thi nki ng, inthe devel opment of l i ngui sti c theori es, as of all other
theori es, i s at present extremel y restri cted. The mor e we understand the
nature of sci enti fi c theori es, the di mmer becomes the prospect of maki ng
i ncreased use of comput er facilities indevel opi ng such theori es, t hough
there seemto be good prospects for a much l arger use of comput ers inthe
future i n the testi ng of theori es devel oped by humans and in forci ng
human sci enti sts to i ncrease the formal i zati on of thei r theori es, in order to
al l owfor thei r testi ngby computers.
The si tuati on mi ght change if a l anguage coul d be constructed in
whi ch humans and comput ers coul d converse, ina mor e or l ess literal
sense of thi s word. But, unfortunatel y, here agai n, the better we under-
stand the nature of human l anguages, the slimmer becomes the outl ook
for the possi bi l i ty of devel opi ng an effecti ve quasi -uni versal man-
comput er conversati onal l anguage. What ever some of the mor e
specul ati ve mi nds i nthe artificial i ntel l i gence fieldmi ght thi nk or say, for
the t i me bei ng at l east, such a conversati onal l anguage will have to l ook
very much likea natural l anguage, if i t i s expected to have the power and
uni versal i ty of such a l anguage. Thi s has often been real i zed but unti l
recentl y the full i mpl i cati ons of this real i zati on wer e not real l y taken
i nto account, probabl y j ust due to wi shful thi nki ng. One knew that
natural l anguages are ful l of ambi gui ti es, and much i ngeni ous thi nki ng
was empl oyed to programcomputers todi scover the syntacti c ambi gui ti es
and perhaps, on the basi s of vari ous tri cks, to di sambi guate gi ven
utterances of syntacti cal l y ambi guous sentences by taki ng i nto consi der-
ati on the l i ngui sti c co-text, for i nstance, t hough thi s was done onl y after
one came to real i ze that the occurrence of syntacti c ambi gui ti es, for all its
annoyi ng character, i s one of the maj or sources of the powerf ul ness of
natural l anguages.
A si mi l ar devel opment, i nvol vi ng sti l l mor e agoni zi ng re-thi nki ng, took
pl ace wi th regard to semanti c ambi gui ti es. I n spi te of some mi nor
successes i n the devel opment of semanti c di sambi guati on procedures,
126 Yehoshua Bar- Hi l l el
based upon parti al sketches of semanti c theori es for natural l anguages and
agai n on taki ng i nto consi derati on l arger and l arger co-texts and
i nf ormati on sol i ci tedf r omthe human partner duri ng the conver sat i on^,
i t became soon cl ear that the successes wer e defi ni tel y of an i sol ated
character, with smal l chances for si gni fi cant general i zati ons. But the
si tuati on began to l ook al most hopel ess when the essenti al l y pragmati c
character of human l anguages was ful l y real i zed and when one began to
l ook at the rel evant facts strai ght inthe face (Bar-Hi l l el , 1970a). Human
communi cat i on in natural l anguages i s, in general , so successful and
powerf ul , j ust because i t rel i es to such a hi gh degree on extra-l i ngui sti c
channel s, on the fact that speakers and hearers have certai n background
knowl edge, have eyes, ears, and hands, and make use of themduri ng
conversati on, as well as on countl ess other factors. The fact that innormal
communi cat i on, utterances of syntacti cal l y and semanti cal l y ambi guous
sentences are most of the timeunderstood by the hearer inthe exact sense
i ntended by the speaker, is due mostl y to the ci rcumstance that all con-
cerned make effi ci ent use of these addi ti onal channel s. The di l emma
whi ch the pl anner of a man- machi ne conversati onal l anguage has to face
i s now cl ear: ei ther the computer has to be equi pped withthe equi val ents
of the addi ti onal channel s, turni ng i t i nto a quasi - human enti ty, or el se
some way has to be f ound to expand the powers of the onl y channel
avai l abl e so far, i ts huge memor y capaci ty and abi l i ty to understand
certai n programmi ng l anguages, so as to turn i t i nto a medi umthat will be
abl e to take over the capaci ti es of the other channel s. Though attempts i n
those di recti ons are bei ng made at present by research groups all over the
worl d, i t shoul d be unequi vocal l y stated that so far both horns of the
di l emma are ina l argel y programmati c stage and that onl y the perenni al
opti mi sts still bel i eve to be on thei r way towards a sati sfactory sol uti on.
J ust because natural l anguage has recentl y ceased to be a mystery, j ust
because we now understand i ts nature so much better than ever before,
we are i na posi ti on to appreci ate i ts uni queness as a systemof commu-
ni cati on and to get a realistic viewof the difficulties that stand i n the
way of any attempt to repl ace i t by some artifical devi ce.
REFERENCES
BAR-HILLEL, Y., (1962) Some recent results in theoretical linguistics, in E. Nagel,
P. Suppes and A. Tarski (eds.), Logic, Methodology, und Philosophy of Science;
Lunguuge 127
Proceedings of the 1960 International Congress, Stanf ord (Calif.), Stanf ord Uni v.
Press, 551- 557.
BAR- HI LLEL, Y. , (1964) Language and Information, Readi ng (Mass,), Addi son- Wesl ey.
BAR- HI LLEL, Y. (1966) On a mi sapprehensi on of the status of theori es i nl i ngui sti cs,
Fou?zdations of Linguistics, 2, 39- 399, repri nted i nAspects of Language.
BAR- HI LLEL, Y. , (1968) Cyberneti cs and l i ngui sti cs, in S. Moses (ed.), Information
und Kommimilcation, Muni ch- Vi enna, R. Ol denbourg, 29- 38. Repri nted inAspects of
Language.
BAR- HI LLEL, Y. (1969) Uni versal semanti cs of the phi l osophy of l anguage , inJ. Puhvel
(ed.), Substance anclStructiire of Language, Uni v. of Cal i f omi a Press, 1-21. Repri nted
i nAspects of Language.
BAR- HI LLEL, Y. , (1970a) Communi cat i on and argumentati on i n pragmati c l anguages ,
Linguaggi Nella Societ e Nella Tecnica ( Convegno I nternazi onal e Ol i vetti , Mi l an,
1968), Mi l an, Edi zi oni di Communi t .
BAR- HI LLEL, Y., (1970b) Aspects of Language, J erusal em, Magnes Press.
BRI GHT, W. , (ed.) (1966) Sociolinguistics, The Hague, Mout on.
CARNAP, R., (1937) The Logical Syntax of Language, London, Kegan Paul , Trench,
CARNAP, X., (1946) Introduction to Semantics, Harvard Uni v. Press.
CWERRY, C., (1957) On Human Communication, New Yor k, John Wi l ey.
CHOMSKY, N. , (1957) Syntactic Structures, The Hague, Mout on.
CHOMSKY, N. , (1959) Language, 35, 26- 58 (repri nted inJ. A. Fodor and J. J. Kat z,
CHOMSKY, N. , (1965a) Current Issues in Linguistic Theory, The Hague, Mout on.
CHOMSKY, N. , (1965b) Aspects of the Theory of Syntax, Cambr i dge (Mass.), M.I .T.
CHOMSKY, N. (1966) Cartesian Linguistics, New Yor k, Har per and Row.
CHOMSKY, N. , (1968) Language andlllind, New Yor k, Harcourt , Br ace and Wor l d.
CHOMSKY, N. and HALLE, M. , (1968) The Sound Pattern of English, New Yor k, Har per
Encyclopaedia of Philosophy, vol . VI I , New Yor k, Macmi l l an, 358- 406.
FODOR, J. A. and KATZ, J. J. (eds.), (1964) The Structure of Language, Engl ewood Cl i ffs
(N.J.), Prenti ce- Hal l .
GUMPERTZ, J. J. and HYMES, D. , (eds) (1969) Directions in Sociolinguistics. The
Ethnography of Communication, New Yor k, Hol t, Ri nehart and Winston.
HARRI S, Z. , (1957) Co- occurrence and transf ormati on i nl i ngui sti c structure , Language,
33, 283- 340; repri nted i n J. A. Fodor and J. J. Kat z (eds.) The Structure oflanguage,
Engl ewood Cl i ffs (N.J.) Prenti ce- Hal l , 1964, 115- 210.
HINTIKKA, J., (1968) Epi stemi c l ogi c and the met hods of phi l osophi cal anal ysi s ,
Australasian Journal of Philosophy, 46, 37- 51 (repri nted inJ. Hi nti kka, 1969, Models
for Modalities, Dordrecht / Hol l and, Reidel).
Trubner and Co.
The Structure of Language, 547-578. )
Press.
and Row.
HOCKETT, C. , (1958) A Course in Modern Linguistics, New York, Macmi l l an.
KATZ, J. J., (1966) The Philosophy of Language, NewYor k, Har per and Row.
KATZ, J. J. and FODOR, J. A. , (1963) The structure of a semanti c theory,Languuge,40,
170- 210 (repri nted as Chap. 19 of The Structure of Language).
128 Yehoshua Bar-Hillel
LENNEBERG, E. H. (1964) The capaci ty for l anguage acqui si ti on , inJ. A. Fodor and
LUCE, R. D., BUSH, R. R. and GALANTER, E., (eds.) (1965) Readings in Mathematical
LYONS, J., (1968) Introduction to Theoretical Linguistics, Cambr i dge (Mass.), Cambr i dge
MORRI S, C. (1938) Foundations of tJze Theory of Signs, Chi cago (Ill.), Uni v. of Chi cago
OLDFI ELD, R. C. and MARSHALL, J. C. , (eds.) (1968) Language, Pengui n Books.
PEI RCE, C. S., (1931-1935) Collected Papers, Harvard Uni v. Press.
PERELMAN, C. and OLBRECHTS- TYTECA (1958) Trait de largumentation, Pari s,
POPPER, K. R., (1962) Conjectures and Refutations, NewYor k, Basi c Books.
QUINE, W. V., (1960) Word and Object, Cambr i dge (Mass.), M. I . T. Press.
SAPORTA, S. (ed.) (1961) Psycholinguistics, New Yor k, Hol t , Ri nehart and Wi nst on.
SAUSSURE, F. DE, (1955) Cours de linguistique gnrale, Pari s, Payot , fifthedi ti on.
SCHI LPP, P. (ed.), (1963) The Philosophy of RudoZph Carnap, L a Salle(IU.), Open Court.
SHANNON, C. E. and WEAVER, W. (1949) The Mathematical Theory of Communication,
SKINNER, B. F., (1957) Verbal Behavior, New Yor k, Appl eton- Century- Crof ts.
TARSKI, A., (1956) Logic, Semantics, Mathematics, Oxf ord, Cl arendon.
ULLMAN, S., (1962) Semantics: an Introduction to the Science of Meaning, Oxf ord,
WIENER, N. , (1948) Cybernetics or Control and Communication in the Animal and the
WI TTGENSTEI N, L., (1922) Tractatus Logico-Philosophicus, London, Routl edge and
WI TTGENSTEI N, L. , (1953) Philosophical Investigations, Oxf ord, Bl ackwel l .
ZI FF, P. (1960) Semantic Analysis, I thaca (N.U.), Come11 Uni v. Press.
J. J. Kat z (eds.), The Structure of Language, 579-603.
Psychology, New Yor k, J ohn Wi l ey.
Uni v. Press.
Press.
Presses Uni versi tai res de France.
Ur bana (Ill.), Uni v. of Il l i noi s Press.
Bl ackwel l .
Machine, New Yor k, John Wi l ey.
Kegan Paul .
7 Information
SATOSI WATANABE
1. INFORMATION CONVEYED BY SENTENCE
What is the di fference between the fol l owi ng t wo sentences f r omthe poi nt
of vi ewof the i nf ormati on conveyed by t hem?
(1) J ohn reads the book.
(2) J ohn reads a book.
A meani ngf ul answer to the questi on can be gi ven if one noti ces that (1)
may be an answer to one of the three questi ons:
(1.1) Who reads the book?
(1.2) What does J ohn do wi th the book?
(1.3) Who does what wi th the book?
(2.1) What does J ohn read?
(2.2) What does J ohn do wi th what ?
(2.3) Who reads what ?
(2.4) Who does what wi th what ?
Sentence (2) may be an answer to one of the f our questi ons:
Of course, the questi on may be actual , taci t, potenti al or anti ci patory
dependi ng on the ci rcumstances.
(1)" It is J ohn ( among others) who reads the book.
It coul d be J ohn, Mar y, Di ck or Li nda, but i t i s J ohn who reads the book.
(2)" It i s a book ( among other thi ngs) that J ohn reads.
It coul d be a book, a newspaper, a l etter, but i t is a book that J ohn reads.
The functi on of the defi ni te ' and i ndefi ni te arti cl es here i s to suggest
what the possi bl e al ternati ves are and to tel l what the actual choi ce is.
The noun wi th a defi ni te arti cl e is i rrel evant to thi s choi ce- maki ng, whi l e
the one wi th an i ndefi ni te arti cl e is rel evant. I n other words, i t hel ps
i ndi cate what the message is, what the i nf ormati on i s, i.e. what the
I f (1.1) i s the questi on, statement (1) can be re-phrased as
I f (2.1) i s the questi on, statement (2) can be re-phrased as
130 Satosi Watanabe
underl yi ng questi on i s and what i ts answer is. Thi s i s preci sel y what the
l anguage - at l east i ni ts cogni ti ve aspect - i s used for. It shoul d answer a
questi on actual or i magi nary.
Seen f r omthis poi nt of vi ew, the f ormal l anguage used in symbol i c
l ogi c, P(a), meani ng that obj ect a affi rms predi cate P is defecti ve. Does i t
mean: 1) i t is (a) predi cate P that (the) obj ect a affi rms, or 2) i t i s (an)
obj ect a that affi rms (the) predi cate P? The i dea of i nf ormati on as a
choi ce among possi bl e al ternati ves i s so i mportant that each l anguage has
devel oped i ts own devi ce to express it. It i s surpri si ng to note the same
f uncti on is perf ormed by enti rel y di fferent grammati cal enti ti es indi fferent
l anguages. Compar e t wo J apanese sentences:
(3) Wat akushi ga yuku,
(4) Wat akushi wa yuku,
where ' watakushi ' means I or me, and ' yuku' means go. The t wo smal l
parti cl es ' gay and ' wa' both i ndi cate that the noun or pronoun precedi ng
t hemi s inthe nomi nati ve case. Hence, both (3) and (4) can be transl ated
as 'I go'.
If you aks a J apanese grammari an the di fference between the t wo, the
answer woul d perhaps be that the emphasi s is sl i ghtl y di fferent. A much
cl earer answer can be gi ven f romour poi nt of viewof i nf ormati on.
Sentences (3) and (4) answer respecti vel y the fol l owi ng questi ons :
(5) Who goes?
(6) What do YOLI do?
Hence, the meani ngs of (3) and (4) can be cl ari fi ed(though a bit awkwar d)
if we transl ate t hemrespecti vel y as
(3)" It is I who go.
(4)" What I do is goi ng.
(3)** Others do not go.
(4)** Others may remai n or go.
Those who know J apanese well enough will agree that (3) and (4) are
often used to mean (3)"" and (4)**.
(3)" means that there exi sts a possi bi l i ty that somebody el se, such as
J ohn, Sam or Mar y, shoul d go but the actual choi ce i s I . (4)" means that
I coul d remai n or go or perhaps do somethi ng el se but the deci si on is that
I go.
For this reason, (3) and (4) respecti vel y i mpl y al so:
Infornzation 13 1
2. QUANTITATIVE CONCEPT OF INFORMATION
The short anal ysi s of the precedi ng secti on will, i t is hoped, have shown
cl earl y that the essence of cogni ti ve l anguage is conveyance of i nf orma-
ti on and the essence of i nf ormati on i s choi ce- maki ng. I f that is the case,
and if sci ence shoul d ai mat an abstract - and if possi bl e, quanti tati ve -
formal i zati on of concrete i nstances, then the sci ence of l anguage, and in
parti cul ar the sci ence of i nf ormati on, has to take up as i ts maj or task a
mathemati cal f ormul ati on of the process of choi ce- maki ng. The t er m
choi ce has someti mes a connotati on of personal preference or aversi on,
but i t is used here of course ina mor e neutral sense of whi ch one of the
potenti al al ternati ves i s sel ected.
The fi rst thi ng we shoul d ask oursel ves is: what shoul d we understand
by a l arge amount of i nf ormati on and a smal l amount of i nf ormati on?
Let us take the case of questi on (1.1) and its answer (1). If there i s onl y
J ohn inthe r oom, then the answer (1) i s not conveyi ng any i nf ormati on.
For the questi on tacitly acknowl edges al ready that somebody in the
r oomi s readi ng the book, and if J ohn is the onl y person inthe r oom, then
the l ogi cal concl usi on al ready is that J ohn i s the answer. On the other
hand, suppose that there are many peopl e inthe r oom, then the si ngl i ng-
out of J ohn among many other competi tors has news val ue . Thi s
suggests that the l arger the choi ce, the l arger the amount of i nf ormati on.
So, if z1 i s the number of al ternati ves and if we denote the amount of
i nf ormati on by I, we shoul d assume that I i s an i ncreasi ng functi on of n,
that is to say, lis determi ned by n insuch a way that the l arger the n the
l arger the I. We know that the l ogari thmi c f uncti on is an i ncreasi ng
functi on. So i t may be all ri ght to defi ne I by:
I = l og Iz. (1)
The reason why we choose the l ogari thmi c f uncti on among all other
i ncreasi ng functi ons is as fol l ows. Suppose there are t hree boys, J ohn,
Sam, and Fred, i n the r oomThen the sentence (1) as an answer to
questi on (1.1) conveys an amount of i nf ormati on equal tolog 3. Next, l et
us regard sentence (1) as an answer toquesti on (1.2). Let us assume that
J ohn can do t wo thi ngs. Ei ther he reads the book or he falls asl eep wi th
the book. Then the number of al ternati ves i s 2, and sentence (1) as the
answer to questi on (1.2) conveys i nf ormati on l og 2. Next l et us come to
questi on (1.3) assumi ng that each of the three boys can ei ther read or fall
132 Satosi Watuizube
asl eep wi th the book. The number of possi bl e answers is obvi ousl y
6 = 2 x 3: J ohn reads the book, J ohn sl eeps, Sani reads the book, Sam
sl eeps, etc. Theref ore, the i nf ormati on carri ed by sentence (1) as the
answer to questi on (1.3) i s l og 6 = l og 2 + l og 3. This is exactl y equal to
the amount of i nf ormati on carri ed by (1) as the answer of (1.1) pl us the
amount of i nf ormati on carri ed by (1) as the answer of (1.2). Si nce
questi on (1.3) is the j uxtaposi ti on of t wo questi ons (1.1) and (l.2), thi s
equal i ty is extremel y conveni ent. Thi s conveni ent addi ti ve property st ems
f r omthe fact that we chose a l ogari thmi c functi on in(1).
3. GENERALIZATION OF THE QUANTITATIVE CONCEPT OF
INFORMATION
Suppose you are pl anni ng an excursi on t omor r owand you are worri ed
about the weather. Suppose the Weat her Bureau tel l s you that you shoul d
expect (1) fai r, (2) rai n, and (3) snow, wi th respecti ve probabi l i ti es
(1) p1 = 0.9, (2) p2 = 0. 09, and (3) p3 = 0. 01. Next morni ng, your chi l d
opens the wi ndow and tel l s you: i t i s snowi ng. Yov. are surpri sed because
there was a very smal l probabi l i ty of snowand yet i t happened. On the
other hand, i f your chi l d tel l s you that i t i s fai r, you are not surpri sed
because that is what you had expected (l arge probabi l i ty). Thi s means, the
smal l er the probabi l i ty you attach toan event, the l arger yous surpri se if
the event actual l y occurs. (Events can be transl ated int er ms of sentences
too.) We may then express the amount of surpri se of event i by SZ =
-l ogPt, where i may be 1, 2, .. . , E-; if there are i ntotal n al ternati ve events.
Si nce the probabi l i ty i s between O and 1, this quanti ty St becomes non-
negati ve. The choi ce of the negati ve of the l ogari thmi c functi on rather
than any other decreasi ng, non- negati ve functi on has i ts reason. It is
si mi l ar to the one gi ven i nthe l ast secti on, but will not be repeated here.
The probabi l i ti es, such as p1, p2 and p3 inthe above exampl e, represent
col l ecti vel y your state of i gnorance about the out come whi ch necessari l y
must be one of the possi bl e al ternati ves. For, if you were omni sci ent, you
woul d attach probabi l i ty uni ty (i.e. certai nty of occurrence) to one of the
al ternati ves and probabi l i ty zero (i.e. certai nty of non-occurrence)
to other al ternati ves. If you are not omni sci ent, you have probabi l i ti es
whi ch are nei ther one nor zero. After the chi l d has reported the actual
weather, however, the probabi l i ti es change and one of t hembecomes
Information 133
uni ty. If he says snow, then p1 and p2 become zero and p3 becomes one.
How coul d we express mathemati cal l y the degree of our i gnorance? We
may say that the l arger the i gnorance, the l arger will be the surpri se when
the actual choi ce is made known. I ndeed an omni sci ent person will not be
surpri sed, because he knows the answer bef orehand. We sai d that event
i will cause surpri se S4 = -l og pz if i t happens. Si nce event i happens,
accordi ng to our expectati on, with probabi l i ty pz, we can make the
average or expected val ue of surpri se by cal cul ati ng S =
We may use this quanti ty
n
i = l
Si pt.
as the expressi on of the degree of i gnorance when our state of i gnorance i s
gi ven by probabi l i ti es, pl, p2, ..., pn. We agree that O l og O is zero,
al though l og O i s - OO. If we i nterpret the ps not j ust as our expectati on
but as a rel ati ve f requency inan ensembl e defi ned by certai n condi ti ons,
we may consi der S of (2) al so as the expressi on of uncertai nty.
Now suppose we make an observati on (likethe report of the chi l d) and
fi nd out the actual choi ce, then one of the probabi l i ti es becomes one and
all the rest become zero. Then the degree of i gnorance S becomes zero.
The i gnorance S gi ven in(2) changes to zero. The decrease ini gnorance
(by observati on or report) can be i nterpreted as i nf ormati on obtai ned.
For this reason S can al so be consi dered as the amount of i nf ormati on
provi ded by i ts observati on or report. Thi s will expl ai n why the apparent
opposi te concepts of i gnorance and i nf ormati on are expressed by the
same f ormul a.
Thi s defi ni ti on (2) i s actual l y a general i zati on of our earl i er defi ni ti on
(1). I ndeed, if we have n al ternati ves and if we do not know whi ch
al ternati ves are mor e probabl e than others, we may putpt = l /n for all i.
Then (2) becomes l og n as i n (1). Engi neers liketo attri bute (1) to Hartl ey
(1928) and (2) to Shannon (1948), but of course both expressi ons wer e
used al ready i n the l ast century by Bol t zm- ann (1896-1898) as the ex-
pressi ons of physi cal entropy. The i dea that the physi cal entropy corres-
ponds, i na certai n sense, to theuncertai nty about the preci se state of the
obj ect must have been known to Bol t zmann, but i t was Szi l ard (1929)
who poi nted out with emphasi s that a decrease of the physi cal entropy
shoul d i mpl y the obtai ni ng of i nf ormati on by the observer.
134 Sutosi Wat unabe
4. PART AND WHOLE - THEORY OF STRUCTURE
If the bi rth of i nf ormati on theory shoul d be mar ked by i ts i ndependence
f r omt hermodynami cs, the present author s paper of 1939 may be regar-
ded as the fi rst paper inthe newdi sci pl i ne. Usual l y, i nf ormati on theory
i s consi dered tohave been born in1949 ( Shanon, 1949). The paper of 1939
had t wo messages to convey: (A) the quanti ty expressed by the functi on
(2) can be used as a measure of uncertai nty inany area di sconnected f r om
t hermodynami cs; (B) the same functi on can be used as a measure of
structure i n an assembl y of i ndi vi dual obj ects. The theory used in the
paper of 1939 was f ormul ated inthe f r amewor k of the non- Bool ean l ogi c
( quant uml ogi c) and therefore was much mor e general than (hence
i ncl uded as a speci al case) the usual i nf ormati on theory, but we shal l not
di scuss thi s last poi nt inthe present paper. Poi nt (A) may have been made
cl ear by the expl anati on gi ven inthe fi rst three secti ons of the present
paper. We shal l now expl ai n poi nt (B) bri efl y inthis and the next secti ons.
The i dea of redundancy di scovered inthe 1950 s by the communi cat i on
engi neers i s a si mpl e case of what is meant by structure here.
As the assembl y of i ndi vi dual obj ectsy, you can i magi ne, for i nstance,
a human soci ety or a pi ece of matter consi sti ng of mol ecul es. Nobody can
predi ct the exact behavi our of i ndi vi dual persons i nsoci ety or the exact
posi ti ons and vel oci ti es of mol ecul es. The state of each i ndi vi dual can
therefore be consi dered as a chance phenomenon. The assembl y, bei ng a
col l ecti on of these stochasti c i ndi vi dual s, al so behaves stochasti cal l y.
When woul d we then say that an assembl y is structured or organi zed?
If the behavi ours of i ndi vi dual s are correl ated, and i nterdependent, we
woul d say that there i s some structure. The resul t of such correl ati on
woul d be that inspi te of i ndi vi dual di sorder or uncertai nty, the beha-
vi our as a whol e will show some regul ari ty or predi ctabi l i ty and l ess
di sorder or uncertai nty. If t hi s i s the case, the degree of structure may be
measured by the bal ance between the uncertai nty of i ndi vi dual s and the
uncertai nty of the whol e. For, the l arger the i ndi vi dual di sorder and the
smal l er the total di sorder, the l arger the structure.
The above consi derati on l eads us to the fol l owi ng i dea. The degree of
structure K i s expressed by the sumof the entropi es (the S- f uncti on gi ven
in(2)) of i ndi vi dual s mi nus the entropy of the whol e.
K = Z S(part) - S(who1e)
(3)
I nformati on 135
Let us test thi s i dea wi th a very si mpl e exampl e. Let us consi der t wo
girls a and b i na dormi tory. We observe the l ounge and if a is there, we
put 1 in the r ow correspondi ng to a and if a is not there we put O.
Si mi l arl y for girl b. Suppose we have obtai ned the fol l owi ng l i st as the
resul t of ei ght observati ons:
1 2 3 4 5 6 7 8
u 1 1 0 0 1 1 0 0
b O l l O l O O l
Thi s shows that girl a has probabi l i ty 3 of bei ng there and probabi l i ty
of not bei ng there. Hence, her entropy is S(a) = log 2. Thi s cambe
obtai ned f r om(l), as there are t wo cases (presence and absence) wi th an
equal probabi l i ty. Si mi l arl y, the entropy of girl b is S(b) = l og 2. No w
how about the state of the assembl y of t wo girls together? The are f our
cases: bot h there (l,l), both not there (O,O), a is there and b is not there
(l,O), and a is not there and b i s there (0,l). Each of the four cases appears
wi th equal probabi l i ty i nthe above l i st, hence the entropy of the whol e is
S(who1e) = l og 4. Hence, the strength of structure is K = 2 l og 2 - l og
4 = O, i.e. there is no structure.
Thi s resul t agrees well wi th our i ntui ti ve understandi ng of the si tuati on.
The above l i st will be obtai ned i ndeed if t wo girls come to the l ounge
randoml y and i ndependentl y wi thprobabi l i ty 3.
Now consi der the l i st:
1 2 3 4 5 6 7 8
u 1 0 1 1 0 0 1 0
b l O l l O O l O
Thi s l i st shows that each girl comes to the l ounge wi th probabi l i ty 4 as
before, but they are such good fri ends that they come al ways together.
S(a) and S(b) remai n the same, but S(who1e) is di fferent. I n fact there are
onl y t wo cases, both there (1,l ) and both not there (O,O), wi th equal
probabi l i ty 3. Hence S(who1e) = l og 2. The resul t i s that the strength
of structure i s K = 2 l og 2 - l og 2 = l og 2. The poi nt i s that, i ndi vi dual l y
seen, the girls came to the l ounge randoml y wi th probabi l i ty 3, but seen
col l ecti vel y they showsome predi ctabl e regul ari ty, namel y, they come
al ways together.
136 Sutosi Wat unube
5. EMERGENT STRUCTURE
The use of entropy functi ons as a measure of structure was started i n
the paper of 1939, but a mor e systemati c study al ong these l i nes had to
wai t another twenty years ( Wat anabe, 1959; 1960; Garner, 1962). These
l ater papers of the present author showed among other thi ngs that the
entropi c measure of structure had a power to reveal emergent properti es
of an assembl y. By an emergent property i s meant here a property whi ch
does not exi st when i ndi vi dual s are taken si ngl y or i n pai rs but appears
when i ndi vi dual s are taken as consti tuti ng a group of three or more.
Let us take an exampl e. There are f our girls l i vi ngina dormi tory, and
we observe as bef ore thei r presence and absence inthe l ounge. The resul t
of ei ght observati ons was as fol l ows:
1 2 3 4 5 6 7 8
a 1 1 1 1 0 0 0 0
b l l O O l l O O
c o o 1 1 1 1 0 0
d O l O l O l O l
The probabi l i ty of each girl bei ng in the l ounge is 3. Theref ore thei r
i ndi vi dual entropy i s l og 2. Take, next, a pai r of gi rl s, say, a and b.
We fi nd the si tuati on exactl y the same as the fi rst l i st of the last secti on,
i.e. there are f our cases (O,O), (OJ), (l,O), and (1,l ) wi thequal probabi l i ty.
Hence, thei r pai rwi se entropy S(u,b) = l og 4. If we consi der the pai r as the
whol e and the i ndi vi dual girls as the parts , the structure Kf or these t wo
i s K(u,b) = S(u) + S(b) - S(u,b) = O. There i s no structure. Thi s si tua-
ti on i s the same for any of the sixpossi bl e pai rs you may take among t hem.
The pai rwi se structure does not exi st among these four girls.
Next , take t hree girls together, say, a,b,c. We di scover f r omthe above
list, they show f our di fferent patterns (l,l,O), (l,O,l), (O,l,l), (O,O,O) wi th
equal probabi l i ty. Theref ore, the entropy for the t hree as a group i s
S(u,b,c) = l og 4 = 2 l og 2. Appl yi ng our f ormul a tothi s group of three we
get K(u,b,c) = S(a) + S(b) + S(c) - S(a,b,c) = 3 l og 2 - l og 4 = l og 2.
So, there is some structure i nthe group of (u,b,c).
Next , take another group of three, say, (a,b,cE). We di scover that they
show ei ght di fferent patterns. (l,l,O), (l,l,l), (l,O,O), (l , O, l ~, (O,l,O),
(O,l,l), (O,O,O), and (O,O,l) wi th equal probabi l i ty. Hence S (a,b,d) =
l og 8 = 3 l og 2. The structure for these three is K(a,b,d) = S(a) +
Information 137
S(b) + S(d) - S(a,b,d) = 3 l og 2 - l og 8 = O. Thi s means that the
group of (a,b,d) shows no structure. Ther e are t wo mor e ways of taki ng
a group of three girls, and they do not showany structure ei ther.
Why does the group of (a,b,c) have structure whi l e the other t hree do
not? The reason i s because the group of (a,b,c) l acks the patterns (l,O,O),
(O,l,O), (0,0,1) and (l ,l ,l ) whi ch exi st i n the other groups of three. Such
a si tuati on can very well happen if the three girls a,b,c are emoti onal l y
l i nked by fri endshi p and j eal ousy insuch a way that each of t hemdoes
not want tobe al one i nthe l ounge but does not want to see the other t wo
together at the same ti me. Such a property di sappears if we take girls
pai rwi se but emerges if we take t hemi n a group of three. The groups of
three other than (a,b,c) do not have such an emoti onal tie, hence they
come i nto and l eave the l ounge ina r andomfashi on.
To si mpl i fy the expl anati on, we t ook si mpl e and concrete exampl es, but
the met hod of structure anal ysi s based on the i dea of i nf ormati on i s .
appl i cabl e al so to mor e abstract i t ems such as grani mati cal enti ti es,
proposi ti ons, etc., to reveal sti uctures underl yi ng t hem. We may di scern
t wo di fferent streams of thought i nthe anal ysi s of structure. One i s the
now popul ar grammati cal approach and the other is the stati sti cal
approach. The fi rst tri es to f ormul ate the generati ve rul es f r omwhi ch all
i ndi vi dual cases are supposedl y deri ved by deducti ve i nstanci ati on,
whilethe second tri es to di scover the underl yi ng common general i ty or
regul ari ty i nducti vel y f r oma l arge number of i ndi vi dual cases. I f the
f ormer i s rati onal i sti c the latter is empi ri ci sti c. These t wo, however,
woul d not consi der t hemsel ves as mutual l y excl usi ve but they shoul d
compl ement each other. Our met hod of structural anal ysi s bel ongs of
course tothe l atter stream.
K(b,c,d) = K(a,c,d) = o.
6. NOISELESS CODING
In the Uni ted States, the t er mi nf ormati on theory i s understood general l y
inthe sense of its technol ogi cal appl i cati ons intel ecommuni cati on and, i n
parti cul ar, i n the probl ems of codi ng . After the second worl d war,
research produced a certai n smal l number of real l y outstandi ng di s-
coveri es and a l arge number of sci enti fi c heroes, and i nevi tabl y, i nnumer-
abl e di ssertati ons, mostl y l acki ng sci enti fi c ori gi nal i ty and techni cal
138 Satosi Watanabe
utility. The f amous cl assi c is a book publ i shed in 1949 by Shannon and
Weaver enti tl ed The Mathematical Theory of Communication.
Codi ng i s usual l y di vi ded i nto t wo categori es: 1) codi ng for noi sel ess
channel s, and 2) codi ng for noi sy channel s. The fi rst ai ms at the most
economi cal , i.e. statistically shortest way of encodi ng the message for
transmi ssi on through a communi cat i on channel , di sregardi ng the probl em
of di storti on caused by possi bl e errors duri ng the transmi ssi on. The
second ai ms at the most economi cal way of encodi ng the message insuch
a way that we can retri eve the correct message when i t arri ves, wi th
certai n errors at the recei vi ng end of the channel . The latter i s somet i mes
cal l ed error-correcti ng code . There i s of course a thi rd area of codi ng
whi ch has the ol dest hi story, namel y, cryptography whose ai mis to make
decodi ng feasi bl e onl y by those who are provi ded witha certai n secret
and very speci fi c i nf ormati on. Ther e may have been a consi derabl e
progress made i nthis area inrecent years, but not much has been made
known to the general sci enti fi c publ i c.
The i dea of noi sel ess codi ng, or of the stati sti cal l y shortest codi ng
i nthe absence of errors is not newei ther. Everybody knows that the most
frequentl y used l etter inEngl i sh is e , and that the Mor se code assi gns
the shortest code to i t, namel y, a si ngl e dot. To i nfrequent l etters such as
q and x , are assi gned l onger codes. Thi s i ngeni ous i dea seems to be
attri butabl e to Al fred Vai 1 who worked for Samuel Mor se i nthe fi rst hal f
of the ni neteenth century. Of course, the natural l anguages have al ready
thi s tendency. Mor e f requent words have usual l y shorter temporal
durati on when spoken and shorter spel l i ng when wri tten out.
The contri buti on of i nf ormati on theory to this probl emi s a quanti ta-
ti ve rul e that shoul d exi st inthe i deal codi ng between the l ength of the
code for a l etter and the probabi l i ty of i ts occurrence. Suppose we want to
express (encode) four l etters LI, Lp, L3, and L4 as seri es of zeros and ones.
One way of doi ng that may be to assi gn (00) (Ol), (lo), and (11) respec-
ti vel y to LI, L2,4, and L4. There are many other ways, but one of them
may be to assi gn (000), (OOl), (OI), and (1) respecti vel y to LI, Lp, L3,
and L4. These t wo codi ng schemes can al so be expressed as di chotomi c
trees as gi ven on the opposi te page.
We can thi nk of these trees as if they were real trees, and assume that
the thi ckness of the branch becomes one hal f every timea branch spl i ts
off i nto t wo branches. Thi s way, we can assi gn to l etters their thi ckness
41, qp, y3, q4, assumi ng the trunk of the tree to have thi ckness uni ty. For
I nformati on 139
-
O
1
L3'
~
1
IL. L4
i nstance inthe fi rst codi ng menti oned above, each q will be 4, for each
branch has to go through t wo branchi ng poi nts. I n the second codi ng, we
have q1 = 8, q2 = Q, 43 = 4, and q4 = 1- 2'
We note t wo i mportant properti es about the thi ckness. One thi ng is
that if we sumup the q's, i t becomes uni ty. Zqt = q1 + q2 + 43 + y4 = 1 ,
as if each q wer e a probabi l i ty. To assume thi s, we shoul d not l eave any
branch unused. The second thi ng is that the l ength of each code i s propor-
ti onal to 16 = -l og qt. Thi s must be so because qr i s the r-th power of
(4) where r is the number of branchi ng poi nts. Hence, if we take the
negati ve of the l ogari thmof qi, we shoul d get r t i mes l og 2. For i nstance,
inthe second codi ng we have 11 = 12 = -l og (&) = 3 l og 2, 13 = -l og
(&) = 2 l og 2,14 = -l og (+) = l og 2. I f we use l og 2 as the uni t of l ength,
11,12,l3,14 become respecti vel y 3, 3,2, 1 as they shoul d,
Let us assume that each l etter Lr appears wi th probabi l i ty pz in the
l anguage. Then the expected (average) l ength of the encoded message will
become, per l etter, Zpr li = - Cpz l og qr. What are the best val ues of q's to
make thi s expected l ength as short as possi bl e? Ther e is an ol dmathemati cal
theoremattri buted to W. Gi bbs used often by physi ci sts f romthe turn
of the century, whi ch says that the above expressi on becomes mi ni mum
if pt = 4%for all i. Thus, the strategy is very si mpl e: make the l ength of
the code of a l etter proporti onal to -1ogpr. Thi s is the basi c i dea of the
' noi sel ess' codi ng. For i nstance, i fpl = pz = p3 = p4 = 4, the fi rst codi ng
above will be the i deal . If p1 = p2 = 9, p3 = 4, and p4 = 3, then the
second codi ng will be i deal . In other words, the l ength shoul d be -l og pt
di vi ded by l og 2.
But what shoul d we do if --logpi does not become a mul ti pl e of l og 21
A possi bl e strategy insuch a case woul d be to take an i nteger cl ose to
-l og pz/l og 2, but there i s no guarantee that such a codi ng i s feasi bl e or
successful . A general al gori thmfor the opti mal codi ng for an arbi trary
3
140 Satosi Wutartabe
number of symbol s in the code l anguage and for an arbi trary set of
probabi l i ti es was gi ven by Huf f man (1952) and l ater redi scovered
i ndependentl y by Zi mmer mann (1959) inconnecti on wi th the probl emof
searchi ng .
7. ERROR-CORRECTING CODING
Tel ecommuni cat i on i mpl i es transmi ssi on of encoded messages through
some ki nd of physi cal medi umcal l ed a channel (such as copper wire),
pl us emi tti ng, recei vi ng and ampl i f yi ng apparata such as those used in
tel egraphy. It is i nevi tabl e that the symbol emi tted at the i nput and the
symbol recei ved at the output of the channel are not al ways i denti cal ,
because of di sturbances, mal f uncti ons, etc. That i s, there are non- zero
error probabi l i ti es. The i dea of an error-correcti ng code is that by
cl everl y encodi ng the message we can retri eve the ori gi nal message cor-
rectl y i n spi te of the errors whi ch occur inthe channel .
This mi ght mysti fy the l ayman, but he will understand how this can be
done f romthe f ol l owi ng si mpl e exampl e. Assume that we have 12 possi bl e
symbol s avai l abl e for transmi ssi on. Assume further that the errors that
occur are such that the number recei ved may be the ori gi nal number, or
the ori gi nal number pl us one, or the ori gi nal number mi nus one. To make
the probl emeasi er, l et us i magi ne that the numbers are arranged l i ke the
di al of a cl ock, so that 12 pl us 1 means 1 and 1 minus 1 means 12. Under
such ci rcumstances, we may gi ve up sendi ng 12 di fferent numbers and
i nstead send onl y 12, 3, 6 and 9. Then, if 11, 12 or 1 is recei ved i t must have
been ori gi nal l y 12 and if 8,9 or 10is recei ved i t must have been ori gi nal l y
9. I n this way we can reconstruct the ori gi nal message wi thout error in
spi te of the errors inthe transmi ssi on channel .
Now, we have to note that thi s error-correcti ng operati on inthe above
exampl e was not possi bl e wi thout payi ng for i t, i.e., we have restri cted the
number of symbol s avai l abl e f rom12to onl y 4. I f we used all the 12
symbol s wi th an equal probabi l i ty we coul d transmi t i nf ormati on inthe
amount l og 12 accordi ng to f ormul a (1) of Secti on 3. But now by usi ng
onl y 4 symbol s, we can send onl y i nf ormati on inthe amount of l og 4. We
have lost l og 3 = l og 12 - l og 4 of the capaci ty of i nf ormati on trans-
mi ssi on.
The reader must have now seen that inspi te of the errors i nherent in
I nformati on 141
the channel , a correct deci pheri ng of the recei ved message i s possi bl e at the
expense of l oss of i nf ormati on capaci ty of the channel . Lengtheni ng of an
encoded message i mpl i es al so a l oss of capaci ty ( per uni t ti me). Shannon s
f amous theory determi nes ina great general i ty the amount of the mi ni mum
l oss of channel capaci ty for i deal error-free deci pheri ng int er ms of the
error probabi l i ti es of the channel .
There are practi cal l y i nnumerabl e ways of constructi ng error-correcti ng
codes, dependi ng on the types of channel , the error types, and other
practi cal constrai nts. Each, however, wants tocome as cl ose as possi bl e to
the upper limit of channel capaci ty set by Shannon s theorem. An
el ementary precauti on inconstructi ng an error-correcti ng code is not to
use a symbol often if i t i s very l i kel y to i ncur an error. But not to use such
a symbol at all may resul t inundul y l i mi ti ng the channel capaci ty. Many
compl i cated quanti tati ve consi derati ons are i nvol ved. Actual codes
usual l y do not correct all possi bl e errors. They correct onl y the mor e
frequent types of errors and l eave some smal l probabi l i ty of remai ni ng
errors. Usual l y the error-correcti ng operati on i s done on a bl ock of a
certai n number of transmi tted symbol s. A l onger bl ock tends to decrease
the l oss of capaci ty, but becomes mor e and mor e awkwar d for practi cal
handl i ng. The deci pheri ng becomes mor e and mor e compl i cated and we
have to wai t l ong before we can start deci pheri ng. Thus, each codi ng
scheme is characteri zed by the l oss of capaci ty, the l ength of a bl ock and
the remai ni ng error probabi l i ty. Usual l y the timerequi red in deci pheri ng
i s not i ndi cated as one of the characteri sti cs of a codi ng scheme, but if the
schemei s l arge, the very best error-correcti ng code will not do much good
inpracti ce.
So far we have l i mi ted oursel ves to a f ewnarrowtopi cs of conspi cuous
i mportance rel ated to the quanti ty cal l ed i nformati on . Thi s noti on is a
speci al expl i cati on of the broader noti on of i nf ormati on inthe general
field of i nf ormati on sci ences. In the remai nder of thi s paper, we shal l
gl ance qui ckl y at the general background f romwhi ch thi s newgroup of
scientific devel opments ini nf ormati on sci ences has emerged.
8. INFORMATION IN BROADER CONTEXT
The hi stori cal devel opment of sci ence has i ts own i nternal causal chai n,
or at l east i t can so be descri bed. But at the same time, we shoul d r emember
142 Satosi Wataitabe
that i ndustri al probl ems have al ways gi ven i ncenti ves to starti ng a new
wave of sci enti fi c devel opments, and a new sci enti fi c di scovery has often
i mmedi atel y caused new i ndustri al progress. I ndustri al changes, on the
other hand, are i nseparabl e f r omsoci al and i deol ogi cal changes, whi ch
i n turn have i nteracti ons with sci enti fi c thought. In thi s respect, the i n-
f ormati on sci ences are no excepti on.
The ni neteenth century was characteri zed by energy-converti ng
machi nes whi ch converted chemi cal energy i nto heat energy and heat
energy i nto mechani cal energy. Thi s caused an i ndustri al revol uti on in
soci ety i ngeneral and gave bi rth to t hermodynami cs on the theoreti cal
si de. The twenti eth century has seen the appearance of an enti rel y new
breed of machi nes whose functi ons had nothi ng to do with conversi on
of energy - tel ephone, radi os, f i l ms, gramophones, tape recorders,
tel evi si on, computers, etc. They are concerned wi th the conveyance, sto-
rage and processi ng of i nf ormati on. There had been predecessors
( megaphone, pri nti ng, tel egraphy), but the i nf ormati on machi nes are
products and symbol s of thi s century. They are causi ng an i nf ormati on
revol uti on in soci ety i n general and gave bi rth to the i nf ormati on
sci ences on the theoreti cal si de. If the energy-converti ng machi nes can be
compared to human muscl es, the i nf ormati on machi nes are comparabl e
to the human nervous system.
Some apol ogi sts of the i nf ormati on revol uti on cl ai mthat, because
of the effi ci ent and cl ose i nf ormati on transfer that is made possi bl e by
newcomni uni cati on medi a, soci ety is regai ni ng the characteri sti cs of
a tri be of a vi l l age. But thi s i s a grossl y mi sl eadi ng i nterpretati on of the
present century. The bi ggest novel ty of this century is obvi ousl y the mass
medi a, and the mass medi a are not a means of person- to- person contact.
They provi de a tremendous i nstrument for person- to- mass contact, an
i nstrument most sui tabl e for total i tari an governments, i ndustri al
monopol i es, and mass adverti sement. The vi l l age is characteri zed by
person- to- person contact and does not i nterfere wi th i ndi vi dual i sti c di f-
ferences. The person- to- mass contact on the other hand l eads to a uni -
f ormi zati on of the masses, and kills i ndi vi dual i st di fferences or i ndi vi dual
opi ni ons. Thi s l eads to a soci ety of al i enati on, where each person is j ust a
gear i na transmi ssi on box or a di ode ina tel evi si on set ( Wat anabe, 1968a).
We shoul d now face the probl emof what the sci ence of i nf ormati on
shoul d be. It may be recal l ed that i nf ormati on machi nes are pl ayi ng the
rol e of the nervous systeminthe human body. The nervous systemrecei -
Infornzation 143
ves i nf ormati on f romthe receptors, conveys i t through nervous fi bres,
stores i t i nthe memor y, processes i t inthe brai n, makes deci si ons, traas-
mits the deci si on to the effectors. Conversi on of physi cal sti mul i i nto
percepti on, pattern recogni ti on, f ormati on of concepts, deducti ve and
i nducti ve processes, theory- maki ng are all parts of i nf ormati on processes.
The i mportance of sel ecti ng what is useful out of an i mmense sea of
worthl ess i nformati on can never be over- emphasi zed. Ther e must be a
theoreti cal study of all these processes and al so of thei r mechani cal
substi tutes. The f ormati on of thoughts and thei r communi cat i on are
geneti cal l y i nseparabl e. I n that sense, communi cat i on and hi gh-l evel
l i ngui sti cs have sti l l to pl ay an i mportant part, and a l arge part of the
ol d terri tory of psychol ogy, coveri ng cogni ti ve processes, must be an-
nexed. Epi stemol ogy, whi ch was part of phi l osophy, has to be re-
f ormul ated ina quanti tati ve and scientific f rame and al so handed over.
Thi s re-f ormul ated future versi on of epi stemol ogy i s what the present
author cal l ed epi stemometri cs and shoul d become the centre of the
i nf ormati on sci ences ( Wat anabe, 1969).
It is a mi sunderstandi ng to bel i eve that a physi cal obj ect or physi cal
process carri es a defi ni te pi ece of i nf ormati on. I nf ormati on is defi nabl e
onl y by the non- physi cal i nterpretati on of the physi cal i nteracti on
between such a physi cal systemand an external system. A fal l i ng l eaf
may gi ve an observer i nf ormati on about the tree f r omwhi ch i t has come,
or about the wi nd condi ti on of the atmosphere, or about the chemi cal
condi ti on of the soi l , or the fact that i t is aut umn, or that every l i vi ng
organi smand i ts parts are doomed to decay.
The nervous systeminan ani mal does not pi ck up all the i nf ormati on
that is avai l abl e, but onl y those pi eces that are rel evant to the survi val
and growt h of the organi sm. Mor e general l y, we may say that the enti re
i nf ormati on syst emi nan ani mal i s an i nstrument hel pi ng toadapt to and
control the envi ronment. For thi s reason, i nf ormati on has meani ng onl y
inthe cyberneti c context of life. Li fe will l i ve, and for this i t has to adj ust
itself and control the mi l i eu. Thi s can be made possi bl e onl y by col l ecti ng
perti nent i nf ormati on about the outsi de worl d and processi ng i t toobtai n
appropri ate deci si ons. Cyberneti cs is a sci ence of control l i ng the mi l i eu
for the survi val and devel opment of life. As such, cyberneti cs i mpl i es
i nf ormati on and i nf ormati on i mpl i es a cyberneti cal system. The
mechani cal si mul ati on of i nf ormati on processi ng is done by computers,
and for thi s reason, i t is natural that the comput er shoul d occupy an
144 Satosi Wat anube
i mportant pl ace incyberneti cs and the i nf ormati on sci ences. But i t must
be remembered that the comput er wi thout appropri ate i nterpretati on is
merel y a physi cal system, and by i tsel f cannot be a subj ect of cyberneti cal
or i nformati c study. What is i mportant is t he purpose for whi ch human
bei ngs use i t ( Wat anabe, 1968a).
What i s cal l ed i nf ormati on in thi s paper, even ini ts broadest i nter-
pretati on, concerns onl y the cogni ti ve aspect of our thought and com-
muni cati on. It is a common dai l y experi ence that a mer e gl ance at a
person s face i s suffi ci ent to evoke in our mi nd an emoti onal empat hy
for the person. We are recei vi ng some ki nd of i nf ormati on inthis process,
but i t is not cogni ti ve i n nature. Language can convey a great deal of
affecti ve i nf ormati on by i ntonati on, styl e, speed, l oudness, choi ce of
vocubul ary, etc. I t may even be suspected that what we usual l y call a
cogni ti ve process i nvol ves the parti ci pati on of certai n non-cogni ti ve
i nf ormati on ( Wat anabe, 1968b). It is doubtf ul whether such non-
cogni ti ve i nf ormati on can be represented by abstract symbol s and treated
wi th the hel p of l ogi c and mathemati cs. Taken in so wi de a sense, the
noti on of i nf ormati on is at the present day enti rel y outsi de the reach of
sci enti fi c i nqui ry.
REFERENCES
BOLTZMANN, L., (1896-1 898) Vorlesungen ber Gastheorie, Leipzig.
GARNER, W. R., (1962) Uncertainty and Structure as Psychological Concepts, N e w York,
HARTLEY, R. L. V., (1928) Bell System Technical Journal, 7, 535.
HUFFMAN, D. A., (1952) Proceedings of the Institute of Radio Engineers, 40, 1098.
SHANNON, C. N. and WEAVER, W., (1948) Bell System Technical Journal, 27,279, 623.
SHANNON, C. N. and WEAVER, W., (1949) The Mathematical Theory of Communication,
SZILARD, L. (1929) in: Zeitschrgt fr Physik, 53, 840.
WATANABE, S. (1939) in: Zeitschrqt fr Physik, 113, 482.
WATANABE, S., (1959), in: Nuovo Cimento, Ser. X, 13, Suppl., 576.
WATANABE, S., (1960), in: ZBM Journal of Research and Development, 4, 66.
WATANABE, S., (1968a) in S. Dockx (ed.), Civilisation technique et humanisme, Brussels,
Office International de Librairie.
WATANABE, S., (1968b) Mi nd- body problems, in Proceedings of the Hawaii Conference
on Philosophical Problems in Psychology, 1968, Hawaii, Univ. of Hawaii Press, (1972).
WATANABE, S., (1969) Knowing and Guessing, New York, John Wiley.
ZIMMERMANN, S. (1959), in: American Mathematical Monthly, 66, 690.
John Wiley.
Urbana (Ill.), Unk. of Illinois Press.
8 Models
HERBERT STACHOWIAK
1. FUNDAMENTALS
1.1 Attributes and predicates
By attri butes strictly speaki ng are meant properti es (characteri sti cs,
qual i ti es) of i ndi vi dual s, rel ati ons between i ndi vi dual s, properti es of
properti es, etc. The di sti ncti on between i ndi vi dual s and attri butes is not
based upon any metaphysi cs of substance . Accordi ng to sui tabi l i ty, any
obj ect-consti tuti ng el ements can, inone context, stand for attri butes, in
another, for i ndi vi dual s.
The i ndi vi dual s of an obj ect may al so be cal l edzero- order attri butes or
i mproper attri butes. By contrast, proper attri butes woul d be fi rst-,
second- , etc. order attri butes. Fi rst-order attri butes are properti es (of
the i ndi vi dual s in questi on) and rel ati ons (between the i ndi vi dual s in
questi on), second- order attri butes are properti es of properti es, properti es
of rel ati ons, rel ati ons between properti es, rel ati ons between rel ati ons,
etc. (accordi ng to the type-theoreti cal di sti ncti ons of logic). No attri bute
(proper or i mproper) can be sai d to bel ong to an obj ect as such. All i ts
attri butes are attri buted to the obj ect. W e may say: The obj ect is con-
sti tuted, or produced, by attri buti on.
I n the general theory of model s onl y such attri butes are admi tted for
whi ch there i s at l east one suffi ci entl y conventi onal i zed met hod accordi ng
to whi ch the exi stence or non-exi stence of these attri butes can be deci ded
upon.
Attri butes are pri mari l y products of percepti on and cogi tati on. Hence
they must be careful l y di sti ngui shed f romthei r l i ngui sti c and semi oti c
arti cul ati ons - spoken, wri tten, or otherwi se symbol i zed. It must be sai d,
however, that those i nternal f ormati ons are not meant to be a worl d of
spi ri tual enti ti es, i sol ated meani ngs , or of Wesenheiten, Sinngehalte,
146 Herbert Stachowiuk
and the l i ke, exi sti ng by i tsel f besi de or beyond the worl d of publ i cl y
percepti bl e si gns. The attri butes represented by si gns are consi dered
rather as bei ng pragmati cal l y consti tuti ve i n the sense of expressi ng
psychi cal processes, i nferabl e onl y f romthe di fferent connecti ons of the
uses of si gns, the latter bei ng stated by external observers.
The symbol i zati ons j oi ned to the attri butes and representi ng t hem
l i ngui sti cal l y are cal l edpredi cates. Accordi ngl y, we are enti tl edto speak
of fi rst-, second- , etc. order predi cates.
The f ormal characteri zati on, as gi ven here, is mai nl y one by l ogi c of
cl asses; onl y occasi onal l y, for the sake of conci seness, i t makes use of
t er ms and descri pti ons proper to the l ogi c of predi cates.
The f ol l owi ng exposi ti ons requi re a type-theoreti cal l y structured
obj ect l anguage 2 (as well as, for the descri pti on of i ts syntax, a syntacti c
meta- l anguage ilJ3 of S). A predi cate, consi dered extensi onal l y, is a cl ass.
The el ements of such a predi cati ve cl ass are cal l ed arguments of the
predi cate. If the arguments of a predi cate are i ndi vi dual s, the predi cati ve
cl ass shal l be cal l ed a one- pl ace predi cate, if they are ordered pai rs, a
two- pl ace p edi cate, etc. I ndi vi dual s shal l be consi dered zero- order
predi cates. Each predi cate whose arguments are i -order predi cates then
i s an (i + 1)-order predi cate for i = 0,1,2, ... , n. As f r oma consi stentl y
extensi onal formal i zati on of the concept of predi cate there woul d soon
resul t heavy f ormul ati ons for many- pl ace rel ati ons, we may temporari l y
use t er ms proper to the (i ntensi onal ) l ogi c of predi cates.
Let then be x and y any predi cates (i ntensi onal l y i nterpreted) of the
type of predi cati ve sententi al f ormul as x(.), y(. , .), etc. withone, t wo,
etc. arguments. Then by x E y shal l be desi gnated, now cl ass-l ogi cal l y
agai n, the fact that the predi cate (predi cati ve cl ass) x bel ongs to the
predi cate (predi cati ve cl ass) y, x bei ng of a l ower order than y. Accord-
i ngl y, by x = y shal l be desi gnated the i denti ty of the predi cates x and
y; i nthis case x and y are of the same order. It will be remembered that
a predi cate withi -order arguments bel ongs to the (i + 1) order, i ndepen-
dentl y of i ts number of pl aces (and that i ndi vi dual s are zero- order
predi cates).
I n Z an i -pl ace I c-order predi cate a shal l be desi gnated by I&. All
the attri butes of ' attri buted' obj ects shal l be f ormal l y characteri zed by
means of such predi cates. I f necessary, di sti ncti ve mar ks (i ndex numbers)
are to be used.
Models 147
1.2 Attri bute cl asses andpredi cate cl asses
For the vari ous consti tuti ons of obj ects each cogni ti ve subj ect di sposes
of a repertoi re, however pre-structured, of possi bl e attri butes (attri bu-
tabl e qualities). The subj ect pi cks out of t hi s repertoi re the el ements
seemi ng sui tabl e to hi mrespecti vel y, and inan order seemi ng sui tabl e to
hi mrespecti vel y.
Suppose for each attri bute cl ass O there is a predi cate cl ass P des-
cri bi ng O. Let O be taken f romthe repertoi re of attri butes < O ) , and
P f r omthe respecti ve repertoi re of predi cates < P ) . All the attri bute and
predi cate cl asses consi dered here are fi ni te.
I n order to real i ze all the predi cates of a repertoi re that are not l ogi cal
vari abl es, they are enumerated after thei r order - and pl ace - number .
The i ndi vi dual s, as zero- order predi cates, are col l ected i n one general
cl ass of i ndi vi dual s; next, the fi rst-order predi cates are enumerat ed as
thei r pl ace numbers i ncrease; second- , thi rd-, etc. order predi cates are
treated accordi ngl y.
Any obj ect of percepti on and cogi tati on can be descri bed by the uni on
cl ass of those among the predi cates of the compl ete repertoi re of an
attri buti ng subj ect whi ch are the %l anguage representati ons of' the
obj ect-produci ng attri butes. Let predi cates correspondi ng to obj ect-
produci ng attri butes general l y be named obj ect-descri bi ng predi cates.
By O. t/. = i s desi gnated the cl ass of obj ect-descri bi ng i ndi vi dual s
(zero-order predicates), by 1 a1 the cl ass of obj ect-descri bi ng one- pl ace
predi cates of the fi rst order (properti es), by 1x2 the cl ass of obj ect-
descri bi ng two- pl ace predi cates of the fi rst order (relations), etc., l i kewi se
by 2d the obj ect-descri bi ng one-pl ace predi cates of the second order
(rel ati ons between properti es, rel ati ons between relations), etc. Then the
obj ect under consi derati on, O, is descri bed by a predi cate cl ass (uni on
cl ass) of the general f or m
nl n2 nk
P(k3n) = Df tl. V u 1 dl V u 2 E% U ... V u k d k ,
i l = 1 i z=1 ik= 1
where n = Max(n1, nz, ..., nk).
Predi cate cl asses i n whi ch occur, besi des zero-order predi cates, onl y
fi rst-order predi cates, are cal l ed predi cate cl asses of the fi rst order.
Accordi ngl y, a predi cate cl ass of the second order i s understood to be a
predi cate cl ass inwhi ch occur, besi des of zero- and fi rst-order predi cates,
148 Herbert Stachowiak
onl y predi cates of the second order, etc. Predi cate cl asses of second and
hi gher orders may al so be cal l ed order-heterogeneous.
As each predi cate x ha-s a predi cati ve asserti on f ormul a, we can say:
if x = KU$ bel ongs to 5, at l east the i argument s of x bel ong to X as well.
2, GENERAL NOTION O F MODEL
2.1 The icostructural mapping
DeJinition 1. Let Pl and PZ be t wo predi cate cl asses. Then a one- t o- one
mappi ng Fis cal l ed an i costructural mappi ng of Pl onto Pa, if the domai n
(pre- domai n) of F i s any subset of P1 and the range (or post- domai n)
of F any subset of P2. F is especi al l y cal l ed i comorphi c, if the pre-
domai n of F coi nci des wi th Pl, and the post - domai n of F withP2. I n the
l atter case, i t can al so be sai d: F i s an i comorphi smof Pl and P2. For
Fi s an i costructural mappi ng of Pl onto Pz may be wri tten I cop Pl, P2.
The inverse of I cop P1,P2, where U;! c PZ is mapped onto Ul c Pl
(U1 = F-1 (Uz)), shal l be symbol i zed by [I cop Pl, P2]-1.
2.2 Encoded predicate classes
The %-l anguage predi cates may now be formal l y i nterpreted. Thi s
requi res a semanti c metal anguage 6 of X. Let i ts onl y pri mi ti ve rel ati on
be the dyadi c rel ati on of desi gnati ng, symbol i zed by Des (Desi gnati on,
desi gnati on). By Des , the atomi c sententi al functi on
Des ~ d l ,
reads: The (5- term desi gnates the X- predi cate K~Z, is consti tuted in 6.
G contai ns both i and (a transl ati on of) X. In accordance wi th Marti n
(1959, p. 30), certai n rul es of desi gnati on are to be observed, e. g. the one
that a constant int ermof the f orm Des Ka6 desi gnates no mor e than
one obj ect-l anguage predi cate KU$. These rul es shal l not be quoted here
indetai l .
DeJinition 2. A %l anguage predi cate Ka( is cal l ed coded, if there is an
G- l anguage t erm wi th Des KU{. I n thi s case 5 i s cal l ed the code si gn
of the predi cate. Predi cates that are not coded shal l be cal l ed non- coded.
Models 149
DeJinition 3. The compl ete cl ass of the code si gns of a predi cate cl ass is
cal l edencodi ng cl ass of the predi cate cl ass (of the systemaggregate, of the
system).
Definition 4. Let P be a predi cate cl ass and C an encodi ng cl ass of P.
Then by an encodi ng of P f r omC, symbol i zed by: Codc P, is meant a
coordi nati on functi on, whose domai n is P and whose range i s C. It hol ds:
each predi cate cl ass P' coded by means of an encodi ng cl ass C is struc-
tural l y equi val ent to a non- coded predi cate cl ass P, whi ch besi des P'
contai ns as subcl asses a number of one-pl ace predi cates equal to the
number of el ements of the cl ass Codc P'.
If i nmany cases code si gns are treated as somethi ng quasi -materi al , not
merel y formal -structural , thi s is done mai nl y by reasons of economy
of descri pti on.
DeJinition 5. Let P1 and P2 be t wo predi cate cl asses, and I: an i costructural
mappi ng of P1 onto P2, wi th the domai n U1 and the range U2. Moreover,
l et CI be an encodi ng cl ass of Pl, and C2 an encodi ng cl ass of P2. Then
a cl ass of ordered code si gn pai rs (l, &) i s cal l ed a transcodi ng cl ass of
P1 and Pz, (symbol i zed by: Tel, C, (P, U1; P2, U2), if: 1) 81 E Cl and
i2 E CZ, and 2) i t hol ds: if1 codes the predi catepl E U1, and z the pred-
i cate p2 E U,, then p1 and p2 are (structural l y) i denti cal , hence, bei ng
enumerated, they coi nci de i n thei r number (and, of course, especi al l y in
thei r numbers of order and place).
Definition 6. Let P1 and P2 be t wo predi cate cl asses, and Cl and CZ
thei r correspondi ng encodi ng cl asses. Then the transcodi ng of P1 i nto P2
f romthe transcodi ng cl ass T, symbol i zed by: TranscodT Pl , P2, means a
coordi nati on f uncti on, whose domai n is (the Cartesi an product) PZ x P2,
and whose range i s T.
Definition 7. If in the transcodi ng Transcodr Pl, P2 the predi cate cl asses
P1 and P2 are repl aced by the predi cate cl asses P: and Pz respecti vel y,
then by the recodi ng of the transcodi ng TranscodT Pl , P2 withregard to
Pf and P;, symbol i zed by: RecodT Pl, Pf; P2, P;, i s meant the trans-
codi ng TranSCOdT PI n Pf, P2 n Pz witha si mul taneous i nversi on of the
el ements of each pai r of the transcodi ng subcl ass of T remai ni ng after the
substi tuti on of Pl by PI n Pf, and of PZ by PZ n P;.
150 Herbert Stachowiak
For Pl n P,* = 8, or Pz n P; = @we say empt y recodi ng, for Pl = Pf
and Pz 3 P$ predi cate-cl ass-i denti cal recodi ng.
2.3 The general concept of model
(A) Characteristic of mapping. Model s are al ways model s of somethi ng,
namel y representati ons of certai n ori gi nal s (or prototypes ), natural or
artificial, whi ch t hemsel ves can be model s agai n.
(B) Characteristic of shortening (reducing, abbreviation). Model s do not
general l y map all the attri butes of the ori gi nal represented by t hem, but
onl y those that are rel evant for the model l er or model - user.
Wi th the pri nci pal characteri sti c (B), the pragmati c di mensi on of vi ew
in a wi der sense i s attai ned. Specscal l y operati onal poi nts of vi ew,
accordi ng to whi ch the sel ecti on of the attri butes of the ori gi nal i s made,
do not yet enter i nto it. Pri nci pal l y, the attri butes of the ori gi nal can, of
course, be sel ected i rregul arl y and stochasti cal l y as well as strictlypurpo-
si vel y -wi th all i ntermedi ates possi bl e. I n a narrower sense, the sel ecti on
of attri butes becomes pragmati c onl y when i t fol l ows defi ni te operati onal
ai ms, and, moreover, when i t is cl ear at what t i mes the model does
represent the ori gi nal , and, above all, for whom.
(C) Characteristic of pragmatical model-function. Model s are not in
themsel ves coordi nated to thei r ori gi nal s. They al ways fulfil thei r
functi ons of substi tuti on onl y for subj ects withgoal - dependent mental or
factual operati ons wi thi n certai n l apses of ti me.
On account of the precedi ng verbal exposi ti ons, a formal i zati on of the
general concept of model ina pragmati cal l ogi c seems to be advi sabl e. To
sketch i t out, Marti n s outl i ne of such a l ogi c may be used once more. The
metal anguages %J I and 6, based upon X, are conti nued by a pragmati c
metal anguage Fp.
Before the latter will be characteri zed, the semi oti c structure of
l anguages < 2, %J I , 6, !@} shal l be compl ement ed to the left, beyond X,
by a semi oti c zero-l evel D. O is the l evel of the non-l i ngui sti c, hence non-
semi oti c obj ects, attri buted accordi ng to 1.1.
In the pragmati cal metal anguage Fp, t wo vari abl es are added: k and t.
k vari es over a gi ven finitecl ass of Kybi ak- organi sms, i.e. of natural or
artificial organi sms di sposi ng of functi on uni ts perceptor, moti vator,
Models 151
operator, and effector (Stachowi ak, 1965; 1969); t vari es over a gi ven
fi ni te cl ass of sl abs of ti me.
First shal l be bri efl y sketched out the temporal l ogi c (Marti n, 1959,
p. 36f.), i ncorporated i nthe pragmati cal l ogi c. The pri mi ti ve rel ati on is
the rel ati on of the temporal antecedens, symbol i zed by: An (Antecedens,
antecedence), withthe atomi c sententi al functi on: An tl , tz, readi ng: the
sl ab of time t l l i es compl etel y before the sl ab of time t2. Deduced
rel ati ons are the rel ati ons of temporal overl appi ng, symbol i zed by: Co
(Contegens), and the rel ati on of bei ng a temporal part, symbol i zed by:
Pa (Parti ci pans), for whi ch hol d the defi ni ti ons: Co ti, tz = Df N An
t l , tz p, - An tz, t l and Pa tl , t2= Df v t An t , t l nAn t,t2. The concept
of the (temporal ) moment , symbol i zed by Mom, shal l be defi ned as
fol l ows: Mo m t = t l Pa t , t n P a t, tl (a temporal moment then is
part of all i ts temporal parts).
Primitives i n ' @shal l be: the tri adi c rel ati on of the semi oti cal accep-
tance, symbol i zed by: Ac (Accepti on, acceptancel ), the doubl y speci fi ed
fi ve-pl ace rel ati on of extended acceptance, symbol i zed by: Acpt (Accep-
tati on, acceptancez), the foui -pl ace rel ati on of' (semi oti c) preference,
symbol i zed by: Praef (Praeferenz, preference), and the sevenfol d speci fi ed
fi ve-pl ace rel ati on of perf ormance, symbol i zed by: Perf (Perf ormati on,
performance).
The atomi c sententi al functi ons, bel ongi ng to the fi rst t wo pri mi ti ves,
are
k Ac o,t,
to be read as: The Kybi ak- organi smk accepts the X- l anguage enti ty o at
t i me t;
k Acpt q o,o', t,
to be read as: Kybi ak k accepts the rel ati on q exi sti ng between the (se-
mi oti c or non-semi oti c) obj ects o and o' at timet.
q vari es here over the repertoi re of cl asses of the acceptance rel ati on
( Descr, Subst.)
wi th the meani ngs:
152 Herbert Stachowi ak
sentence function meaning
(o and o are attri bute- or predi cate
cl asses f r omthe correspondi ng total
repertoi res)
k accepts at timet:
k Acpt Descr o,o, t
o as a descri pti on of
o (consequentl y o i s a predi cate
cl ass)
the substi tuti on o by o
_._..... .............. .I. , _.._............. ..._ ..... _. . ____ _................ ... ........................
k Acpt Subst o,o, t
Moreover:
k Praef o,o, t,
readi ng: Kybi ak k at ti me t prefers the %l anguage enti ty o to the
%- l anguage enti ty o.
k Ac o,t and k Acpt Descr o,o, t can be extended to t- l anguage
enti ti es. For k Ac (a Des K a$), t, for i nstance, is then needed a pragmati cal
meta- metal anguage. Such extensi ons bei ng admi tted, as i t shal l pri nci pal l y
be the case here, i t is suffi ci ent for the moment to postul ate that i n
k Ac o,t k, AC , and Y al ways bel ong to an order that i s hi gher than o
by exactl y one.
( Ac and Acpt are obvi ousl y cl assi fi catory, whereas Praef i s compar -
ati ve. So by the i ntroducti on of the orderi ng rel ati on Praef , the cl assi -
fi catory pragmati cal l ogi c changes i nto a comparati ve one, whi ch may be
consi dered to be a prel i mi nary stage of a quanti tati ve l ogi c (cf. al so
Marti n, 1964)).
Among the X- l anguage entities (%l - l anguage respecti vel y) accepted or
preferred by Kybi ak- organi sms, there may be l ogi cal as well as empi ri cal
sentences (conj uncti ons of sentences).
kAcL, t
i ndi cates that k accepts (all the axi oms and deduced theorems of) the
l ogi c of the l anguage-l evel Z (i ncl udi ng the type-theoreti cal structure of
orders) and al so the l ogi c of the l anguage-l evel %JI. The empi ri cal sentences
accepted by k will have to fulfill condi ti ons of veri fi cati on (inthe wi dest
sense).
Next , i n order to defi ne mor e exactl y the model -theoreti cal l y i mportant
Models 153
concept of the rati onal preference structure of a Kybi ak- organi sm, the
deduced rel ati on of i ndi fference (symbol i zed by I ndi ff) must be i ntro-
duced first. Let k I ndi ff a,b,t = Df - k Praef a,b,t A - k Praef b,a,t.
Then the defi ni ti on shal l hol d (Marti n, 1964, p. 158): k has a rati onal
preference structure at timet, symbol i zed by: Rat Praef k,t, if for any t wo
%- l anguage enti ti es (sentences) a and b ( m- l anguage respecti vel y) hol ds:
1) k Praef a,b,t A k Praef b,c,t CV k Praef a,c,t (transi ti vi ty of the
preference relation), 2) k I ndi ff a,b,t A k I ndi ff b,c,t f3 k I ndi ff a,c,t
(transi ti vi ty of the i ndi fference relation), and 3) k Praef a,b,t \j. k Praef
b,a,t \j. k I ndi ff a,b,t (unambi guousness of preference with non-
i ndi fference; \j. desi gnates excl usi ve di sj uncti on).
Definition 8. A Kybi ak- organi smk is cal l edrati onal at timet, symbol i zed
by: Rat k,t, if at thi s timek accepts the l ogi c of the l anguage l evel s X and
%R and possesses a rati onal preference structure.
Fi nal l y, the atomi c sententi al functi on bel ongi ng to the f ourth
pri mi ti ve i n I p is
k Perf q o,o, t,
readi ng: Kybi ak k perf orms q at timet concerni ng the non- semi oti c or
semi oti c obj ects o and o.
q vari es over the repertoi re
(Descr, Subst, ICOP, ReVF I COP PI, P2, Op$,
whose 7 el ements are cl asses of perf ormance withthe fol l owi ng meani ngs:
TranscodT, RecodT Pl, Pz),
sentence .function meaninE
(o and o are attri bute- or predi cate
cl asses f r omthe correspondi ng total
repertoi res)
k perf orms at t i me t:
k Perf Descr o,o, t the descri pti on o . of o
(consequentl y o is a predi cate
class)
_.._._.._ .......... _. _ ..__... ~ ...._
k Perf Subst o,o, t
.................... _. _ ._... ~ .............................................. _..
154 Herbert Stachowiak
k Perf I cop o,o, t the i costructural mappi ng F of P1
i nto P2 (consequentl y o and o are
predi cate cl asses)
the reverse F* of F (wi th I cop
PI, Pz), i.e. k fi nds out, or works
out, a one to one mappi ng F?
withF-1 E F*, and composes the
predi cate cl ass o = F-1 (P2 n o)
u F* (o-(P2 n o))
..........................................................................................................................................
k Perf (Revp, I COF PI , P2) o,o, t
........................................................................................................................................
k Perf Opt o,of, t certai n operati ons i transformi ng
o i nto o
the transcodi ng TranscodT o,o
(consequentl y o and of are pred-
i cate cl asses)
.........................................................................................................................................
k Perf TranscodT o,of, t
..........................................................................................................................................
k Perf ( RecodT Pl, P2) o,o, t the recodi ng RecodT PI, o; P2, of
of TranscodT Pl,P2 (consequent-
lyo and o are predi cate cl asses)
It woul d transgress the limits of thi s paper to enumerate, however
i ncompl etel y, the di fferent axi omati c or deduced temporal -l ogi cal
rel ati onshi ps appl yi ng to the above types of perf ormance. A hi nt con-
cerni ng Perf Opt may suffi ce. For this sort of perf ormance, for i nstance,
hol ds the (trivial) rel ati on: k Perf Opi o,of, tl A k Perf Opi ofy o, E2 n
An tl, t2. Af t er whi ch the general concept of model can be defi ned as
fol l ows, accordi ng to the three pri nci pal characteri sti cs:
Definition 9. Let 01 and 0 2 be any (semi oti c or non-semi oti c) obj ects.
Then, for a Kybi ak- organi smk, the obj ect 0 2 , at timet, withregard to
certai n operati ons Opt , i s a model of the obj ect 01.
(1) Rat k,t
(2) 3P1 3P2 k Perf Descr Pl, 01, t A k Perf Descr P2, 0 2 , t
(3) 3F k Perf I COF Pl, P2, t
(4) 3T k Perf TranscodT Pl, P2, t
Models 155
(5) k Perf Subst 0 1 , 0 2 , t
(6) 36 k Perf Opi 0 2 , O*, t
(7) 3P* k Perf Descr P::, O*, t
(8) 3 F k Perf ( Rem, I COP PI , P2) P;, P;, t
(9) IC Acpt Descr PT, O;, t A k Acpt Subst O;, O,*, t
(10) k Perf ( RecodT Pl, P2) PT, P;, t.
I n Defi ni ti on 9, there are to be di sti ngui shed: one condi ti on of rati on-
al i ty, 8 condi ti ons of perf ormance ( among whi ch one doubl e condi ti on)
and one (doubl e) condi ti on of acceptance. Among the condi ti ons of
perf ormance, condi ti ons (5), (6), and (8) are the heuri sti cal l y deci si ve ones;
wi th an operati onal concepti on of Defi ni ti on 9, they express the mai n
cl asses of operati ons of the process of model l i ng. Among t hem, condi ti on
(6) i s creati vel y to be val ued mor e hi ghl y than the others. The creati ve
perf ormances expressed by i t i ncl ude the determi nati on of i -opti mi zi ng
target-functi ons.
Though condi ti on (9) expresses but an acceptance, i t means nothi ng
l ess than the producti on of the new ori gi nal , i mproved in some
operati ve sense i n compari son wi th 01. Condi ti on (9) especi al l y makes
possi bl e an i ncrease of i nf ormati on for the model l er k on the model l ed
ori gi nal 01.
Any obj ects 01 and O2 shal l , ina mor e comprehensi ve sense, be cal l ed
semanti cal (symbol i cal ), or el se natural , accordi ng towhether they bel ong
to the semi oti c zero-l evel D (cJ: p. 150), or are semi oti c enti ti es produced
out of { 2, %Il, 6, @> . After thi s, i t will i mmedi atel y be cl ear whether i t
i s a semanti c or techni cal model , and whether a semanti c or natural , or
techni cal ori gi nal is bei ng di scussed i na parti cul ar case.
The operati ons transf ormi ng O2 i nto O,* at fi rst furni sh changes of 01
that are onl y possi bl y desi red, especi al l y newi nf ormati ons about 01. The
model l er k will onl y accept the transferabi l i ty to the represented ori gi nal
of the data obtai ned on the representi ng model , especi al l y of the new
i nsi ghts i nto rel ati onshi ps of the model , and will onl y hol d themrel i abl e if
he has been abl e to ascertai n that certai n transferenti al cri teri a have been
sati sfi ed. Wi th techni cal model s of natural or techni cal ori gi nal s, such
cri teri a are e.g. rel ati ons of scal es for the compari son of l engths, forces,
and energi es between ori gi nal and model , for the compari son of e.g.
modul i of elasticity between ori gi nal and model ( Hooke s l awof model ,
Cauchy s rul e of scale), of densi ti es and vi scosi ti es of l i qui ds (Reynol ds
rul e of scale), and the like. Parti cul ars may be f ound in an essay by
156 Herbert Stachowiak
P. Fi i sgen (1959). Wi th semanti c model s of natural or techni cal ori gi nal s,
e.g. wi th theori es of the empi ri cal sci ences, the transferenti al cri teri a
become cri teri a of conf i rmati on or veri fi cati on, cri teri a of fal si fi abi l i ty,
etc., wi th semanti c model s of semanti c ori gi nal s, especi al l y if the l atter
insome way or other transcend the structure of l anguages ( 2, !IR, 5, !$) ,
they become cri teri a of frui tful f ormati on f r omanal ogy , of l egi ti mate
general i zati on , and so forth. Fi nal l y, transferenti al cri teri a for techni cal
model s of semanti c ori gi nal s are so mul ti fari ous that suggesti ons about
t heminthe present connecti on woul d hardl y be inpl ace.
The pragmati cal character of the concept of model , as determi ned
above, extends beyond the reference of thi s concept to a pragmati cal
l ogi c. It al so concerns the utility of a model for the model er. Onl y a f ew
suggesti ons may be made to this: A model er k will accept his model
Mod (02, 01, k, t, Opt) at timet as usabl e, if i t satisfies certai n moti ves
of k wi thi n t (if i t reduces the strength of the moti ve pressure of k),
and of t wo model s of the same ori gi nal k will prefer the one that perf orms
a greater sati sfacti on of moti ves (agreater reducti on of moti ve pressure)
wi thi n t, or an equal sati sfacti on of moti ves (equal reducti on of moti ve
pressure) wi thi n onl y a part of t.
3. MEASURES
3.1 Strctural measurs
Let Pl and PZ be t wo arbi trary predi cate cl asses f romX and UI E PI
and U2 E P2. Bet ween PI and P2 there shal l exi st the i costructural
mappi ng F with U1 as its domai n, and U2 as i ts range. For any bi t e
cl ass Q, [QI shal l , i n the fol l owi ng, desi gnate the number of the el ements
of Q.
Definition 10. By the (i co)structural (or numeri cal ) adequati on (or
approxi mati on) of P2 to Pl (withi costructural mappi ng F of Pl onto P2)
(cf. 2. 1, p. 148) is meant the not-zero quoti ent
( a m F (Pi, Ui; p2, u21 =) mc0 =
Models 157
DeJinition II. The di fference P1 - U1 shal l be cal l ed the preteri ti on cl ass,
the di fference P2 - U2 the abundance cl ass of the i comorphi c mappi ng F
of PI onto P2. Then the correspondi ng quoti ents
[Pl - u11 P 2 - 7721
[Pl 1 [P21
, @AbUn = apraet =
are cal l ed the preteri ti on measure and the abundance measure of I COF
PI, 2 2 respecti vel y.
Fr omDefi ni ti ons 10 and 11 fol l ows that O < aIco < 1, O < apraet tl,
0 < aAbun < 1. Moreover, i t i s easy to see that for an i comorphi c (p. 148)
mappi ng Iimc0 = 1, and consequentl y @raet = UAbun = O. In whi ch case
U1 shal l be cal l ed non- pretered, and U2 non- abundant wi th regard to the
(i comorphi c) mappi ng F of PI onto P2.
Accordi ng to the characteri sti c of shorteni ng (B) (p. ISO), withmodel -
l i ngs where Pl and Pz f uncti on as descri bi ng predi cate cl asses,
general l y axco < 1, and i t will, with (physi co-, bi o- , and psycho- )
techni cal model s of natural ori gi nal s, lieeven far bel ow1. This hol ds for
sci enti fi c-semanti c model s of natural ori gi nal s; thus, any sci enti fi c
theory will al ways cut but certai n aspects out of the gi ven fields of pre-
sci enti fi cal l y, or at l east pre-di sci pl i nari l y attri buted physi cal , psychi cal ,
soci al , hi stori cal , etc. obj ects. As to the model i ng of natural obj ects,
onl y some phi l osophers will try to construct semanti c model s of real i ty
withthe most total ori gi nal -adequati on (-approxi mati on) possi bl e, namel y
such phi l osophers as adhere to a realistic epi stemol ogy and, perhaps, to a
theory of pi cture-l i ke manner of cogni ti on. If, on the other hand, inthe
physi co-techni cal sphere there are to be found model s with an ori gi nal -
adequati on approachi ng 1 or equal to 1, they ori gi nate wi thout excepti on
f romtechni cal ori gi nal s. a~~~ = 1 is obtai ned especi al l y with techni cal
model s of geometri c (three-di mensi onal ) obj ects, if these model s, no
matter on whi ch spati al or temporal scal e of mappi ng, only structural l y
dupl i cate thei r ori gi nal s.
To get mor e model -theoreti cal l y i nteresti ng measures for structural
adequati ons, submappi ngs of the i costructural mappi ng for order-
heterogeneous (p. 148) predi cate cl asses out of X will be i ntroduced fi rst
bel ow.
I n the f ol l owi ng four defi ni ti ons l et PI and P2 be order- heterogeneous
predi cate cl asses, between whi ch there exi sts the i comorphi c mappi ng F.
158 Her bevt Stachowialc
Defi ni ti on 12. Let KI be the i ndi vi dual cl ass of PI = Pl(kllnl), and a2 the
one of P2 = P2(762)nz). Now, if I cop PI, P2 is parti al l y i ndi vi dual -
preservi ng, namel y, if F maps the i ndi vi dual s (zero-order predi cates) of a
subcl ass Ual c a1 onto the i ndi vi dual s of a subcl ass Uq. c u2, then F
i s cal l ed a di chostructural mappi ng of PI onto P2, and
[UOr.ll [Ua2l
[Ell - 1x21
aDicho =
the di chostructural adequati on of P2 to Pl.
Defi ni ti on 13. Let ~ ~ l f be the cl ass of i -pl ace predi cates bel ongi ng to
PI = P~(~l,nl), and the cl ass of i -pl ace predi cates bel ongi ng to
P2 = P2(Jc27fi2). Now, if I COP PI, P2 is parti al l y pl ace- number- preservi ng,
namel y, if F, for K = O, 1, 2, . . , Max (kl, k2), maps the predi cates of a
subcl ass UKE; G ~ a f onto the ones of a subcl ass U KU; c KU;, then Fis
cal l ed a hypostructural mappi ng of PI onto P2. The measure
where r desi gnates the number of empt y predi cate subcl asses f r om
Pl u Pp withY < Max (kl, k2), where the t erms of the sumfor (each i =
O, 1, 2, . . , Max (nl, n2) and) K = O are repl aced by
[ U 4 - [UU21
[E11 b 2 1
wi th Uc2 = F(Ua1), and where each termof the sumwitha vani shi ng
denomi nator is wri tten zero, is cal l ed the hypostructural adequati on of
P2 to Pl.
DeJ i ni ti on 14. Let be the cl ass of K- order predi cates bel ongi ng to P1
= P1(~1*~1)~ and KU; the cl ass of K- order predi cates bel ongi ng to P2
= Now, if I cop Pl, Pz i s parti al l y order- number- preservi ng,
namel y, if F, for i = O, 1, 2, . . , Max (nl, nz), maps the predi cates of a
subcl ass U,KZ~ E KU: onto the ones of a subcl ass U, c iccciof the same
order number , then Fis cal l ed an orthostructural mappi ng of Pl onto P2.
The measure
Models 159
where s desi gnates the number of empt y predi cate subcl asses f r om
PI v P2 withs < Max (nl, nz), where the t er ms of the sumfor K = O
and i = O, 1, 2, . . ) Max (81, n2) are repl aced by
E . 1 1 * L . 2 1
wi th Uaz = F(Ual), and where each termof the sumwi th a vani shi ng
denomi nator is wri tten zero, i s cal l edthe orthostructural adequati on of
P2 to Pt.
DeJi ni ti on 15. Let ~ a f be the cl ass of K- order and -pl ace predi cates
bel ongi ng to PI = P I ( ~ ~ I ~ ~ ) , and KC& the cl ass of K- order and i -pl ace
predi cates bel ongi ng to P2 = P~(kz7"2). Now, if I cor Pl, P2 is parti al l y
both pl ace- number- and order- number- preservi ng, namel y, if F, for
K = O, 1, 2, . . , Max (kl, kz), and for i = O, 1, 2, . . , Max (al, nz), maps
the predi cates of a subcl ass U U ~ ~ i c K Q ~ onto the ones of a subcl sl ss
U,KU.$ c K Q ~ of the same order- and pl ace- number, then F is cal l ed an
i sostructural mappi ng of P1onto P2. The measure
MaX(nI , K, n2,~) 2
)
Max(kl Yk2)
wi th a = z
( K = l
where nl+ desi gnates the maxi mal number of pl aces of the order K i n
PI , nz,K the maxi mal number of pl aces of the order K i n Pz, and t the
number of empt y predi cate subcl asses with the number of order 6
Max(k1, kz) and the number of pl ace < Max(nl , K, ~ZZ,~), where the t er ms
of the sum for i = O (and each K) together wi th the t er ms for K = O and
i = 1 , 2, . . , MaX(nl , K, nz,,) are repl aced by
EU.11 . 1U.21
160 Herbert Stuchowiuk
withUua2 = F(Uual ), and where each t er mof t he sum witha vani shi ng
denomi nat or i s wri tten zero, is cal l ed the i sostructural adequati on of
Pz to Pl.
It is obvi ous that Defi ni ti on 12 is contai ned as a speci al case i n Defi ni -
ti ons 13 to 15. Moreover, i t is easy to see that both Defi ni ti on 13 and
Defi ni ti on 14 are contai ned as speci al cases i nDefi ni ti on 15.
Obvi ousl y, accordi ng to the i costructural adequati on (p. 156)
O < UDicho < I, O < UH,, < 1, O < aortho < 1, and O < urso < 1.
Definition 16. A di chostructural , hypostructural , orthostructural ,
i sostructural mappi ng F of a predi cate cl ass P1 onto a predi cate cl ass Pz,
whose adequati on measure equal s 1, is cal l ed di chomorphi c, hypomor -
phi c, orthomorphi c, i somorphi c respecti vel y. I n this case, we can al so
say: F i s a di chomorphi sm, hypomor phi sm, ort homorphi sm, i somor-
phi smof P1 and Pz.
For finite domai ns and ranges the model -theoreti cal concept of
i somorphi smgeneral i zes the usual mathemati cal one. The total ori gi nal -
adequati on, as menti oned on p. 157, woul d l ead to fi ndi ng an i somorphi sm
of real i ty, and to i ts theoreti cal representati on, mor e preci sel y: to an
i somorphi c dupl i cati on of the predi cati ve descri pti on of real i ty.
Opposi te to the case that, i n a model l i ng, the i costructural mappi ng F,
exi sti ng between the descri bi ng predi cate cl asses, or one of i ts vari eti es,
reach the val ue of structural adequati on 1, there i s the other extreme of a
maxi mal structural shorteni ng of the ori gi nal . Thi s is the case, if the
model consi sts of one, and onl y one, attri bute, i ts descri pti on then con-
si sti ng of one, and onl y one, predi cate (encoded or not). Such model s
shal l be cal l ed poi nt- model s.
Definition 17. Let F be an i somorphi c mappi ng between t wo predi cate
cl asses PI and Pz. If the i ndi vi dual cl asses al c Pl and MZ E PZ are
i denti cal and if F maps each i ndi vi dual onto itself, then Fis cal l ed an
auto-i costructural mappi ng of PI onto PZ (wi th the l i mi ti ng case of an
auto- i comorphi c mappi ng). Accordi ngl y, upon the equal condi ti on of
i ndi vi dual -i denti cal mappi ng, one will speak of an autodi chostructural
(in the l i mi ti ng case autodi chomorpl Gc), autohypostructural , etc.
mappi ng of Pl onto P2. Fr omthe above i t will be cl ear, too, when one
is al l owed to speak of an auto- i comorphi sm, aut odi chomorphi sm, etc.
of Pl and P2.
Models 161
The f ormul ae for measures of structural adequati ons, as i ntroduced for
the i comorphi c submappi ngs, can i npart be si mpl i fi ed. Thus aDicho i s
transf ormed
[ Uv.112
[KI12
aAuto Dicho = - .
The structural measures may be speci al i zed to the case that part of the
predi cates of the predi cate cl asses descri bi ng the ori gi nal and the model
are of spati al or ti me-metri cal nature, concerni ng e.g. l i near measures,
areas, vol umes, curvatures, etc. inthe sense of Eucl i di an geometry. Wi th
such predi cate cl asses, whi ch may be cal l ed (parti al l y) metri zed, the
structural adequati ons obtai ned after Defi ni ti ons 10and 12 to 15 are
often transf ormed i nto compari sons of scal e.
I f between the metri zed predi cate cl asses PI and P2, descri bi ng t wo
space-metri c confi gurati ons KI and K2 respecti vel y, there exi sts a (model -
theoreti cal ) i somorphi c mappi ng, then ei ther of the t wo confi gurati ons
shal l be cal l ed a structure copy of the other. If inthi s mappi ng all space-
metri cal proporti ons are preserved, i f, therefore, the mappi ng i s equi f orm
and di stance-preservi ng, then the structure copy shal l speci al l y be cal l ed a
space copy. A space copy K2 of a space-metri c ori gi nal KI maps this on a
scal e inr = 1 : 1 = 1. If, however, KI i s equi f orml y mapped onto K2, the
latter becomi ng proporti onal l y smal l er or l arger than KI, so that m, 5 1
respecti vel y, KZ is cal l ed a space-metri cal contracti on model of KI or a
space-metri cal di l atati on model of KI respecti vel y.
Beyond thi s, as many speci fi c metri cal contracti on model s, or di l atati on
model s respecti vel y, can of course be deal t with as there are metri zi ng
parameters.
Among the non- space- metri cal parameters, the cl assi c-physi cal ti me-
parameter is of speci al i mportance. It will be qui te cl ear what is meant by
a ti me- copy, a ti me-metri cal contracti on model , and a ti me-metri cal
di l atati on model of a ti me- dependent ori gi nal wi th the respecti ve
rel ati ons of ti me-scal es mt 2 1.
Of course, space-metri cal ori gi nal -model mappi ngs need not be
equi f orm. I n geometry, certai n ki nds of mappi ng are i nvesti gated,
whi ch are of model -theoreti cal i nterest as well: affi ne, proj ecti ve;
conf ormal (angl e-preservi ng), di stance-preservi ng; coni c, cyl i ndri cal ,
azi muthal , pol yconi cal ones, etc.
162 Herbert Stachowiuk
3.2 Coding measures
Dejinition 18. Let P be a predi cate cl ass wi th the encodi ng Codc P (p. 149).
Then the measure
i s cal l ed the encodi ng degree of P wi th regards to
O < acodgr < 1. Accordi ng to whether for a predi cate
C. It hol ds that
cl ass acoagr = O,
or 0 < aCodgr < 1, or aCodgr = 1 , the predi cate cl ass i s cal l ed non- encoded,
parti al l y encoded, or ful l y encoded respecti vel y.
Dejinition 19. Let Pl and P2 be t wo predi cate cl asses wi th the transcodi ng
TranscodT Pl, P2. Then the measure
i s cal l ed the code- adequati on or materi al adequati on of P1 to Pz wi th
regard to T.
The fol l owi ng notati ons shal l yet be i ntroduced:
DeJinition 20. Let agai n Pl and P2 be t wo predi cate cl asses wi th the
transcodi ng TranscodT P1,Pz. Then TranscodT P1,P2 is cal l ed a real i -
zati on of Pl, or an abstracti on f r omP2 respecti vel y, accordi ng as,
respecti vel y,
Obvi ousl y 0 < aCod < 1.
acodgrc, (Ud < or > a m p C , (U21,
where T = Tel , c2 (PI, UI; P2, U2).
TranscodT Pl, P2, i n parti cul ar, shal l be cal l eda
parti al real i zati on,
full real i zati on,
parti al abstracti on,
ful l abstracti on,
accordi ng as, inthe above sequence
Ul is non- encoded, and U2 parti al l y encoded,
U1 is non- encoded, and U2 ful l y encoded,
Models 163
Ul i s fully encoded, and U2 parti al l y encoded,
U, is fully encoded, and UZ non- encoded.
Definition 21. Let agai n PI and PZ be t wo predi cate cl asses wi th the
transcodi ng TranscodT Pl, P2. Let PI be a descri pti on of an obj ect 01,
and P2 one of an obj ect 02. Now, if to each predi cate of PI, coded by
l, a new code-si gn dz (#I) i s co-ordi nated i nPZ by T, then OZ i s cal l ed
an anal ogy model of 01. If, on the contrary, i nthe transcodi ng TranscodT
Pl, PZ all code-si gns of the predi cates of PI are preserved, 0 2 is cal l ed an
i sohyl i c model of O1.
Definition 22. A model of an ori gi nal that is at once i somorphi c and
i sohyl i c i s cal l edan equate model or a copy of this ori gi nal .
3.3 Furzher measures
To A. A. Mol es and A. Schi i tzenberger (1955) (cf. al so Mol es, 1958,
1960, 1962) we owe a structural measure of compl exi ty for systems,
keepi ng cl osel y to the entropy measure of Shannon. Wi th certai n
modi fi cati ons, this measure can be appl i ed to predi cate cl asses, especi al l y
to (predi cati ve) systems, and thereby to the obj ects (attri bute cl asses)
descri bed by t hem; based on it, compari sons of compl exi ty between
ori gi nal and model can be made and can be rel ated to the structural
measures of adequati on (3.1).
If, i ndoi ng so, predi cate cl asses are concei ved as bei ng messages inthe
sense of the i nf ormati on theory, then, pri nci pal l y, for these predi cate
cl asses there can be defi ned al so measures of i nf ormati on (content)
on a syntacti c, semanti c, and pragmati c l evel , whi ch are a modi fi cati on,
or mor e stri ctl y, a general i zati on, of the f ormul a of Shannon.
On the semanti c l evel Y. Bar-Hi l l el (1953) and R. Car nap have, for
very si mpl e obj ect-l anguages, whi ch contai n onl y a finite number of
i ndi vi dual s and one- pl ace predi cates, defi ned a measure of i nf ormati on
(content) of predi cati ve statements, whi ch is such a general i zati on. Any
devel opments of thi s approach, whi ch uses purel y extensi onal -l ogi cal
means, coul d al so be of i mportance for the general theory of model s. On
the l evel of pragmati cs P. Gng (1967) has, by means of consi derati ons
of the utility theory and the psychol ogy of moti ves, defi ned a measure of
the pragmati c i nf ormati on-content of messages for Kybi ak- organi sms,
164 Her ber t Stachowiak
whi ch can sti l l be mor e cl osel y rel ated to the general theory of model s.
Further measures, whi ch, i t is true, woul d meet wi th great di ffi cul ti es,
if treated ina general way, and coul d probabl y be defi ned onl y separatel y
for certai n ki nds of model s, ought to be referred tothe model -theoreti cal
i mportant concept of contrasti ng. By contrasti ng i s meant the property
of many model s speci al l y to expose, to over-stress certai n attri butes,
or cl asses of attri butes respecti vel y, of the ori gi nal by over-quanti fi cati on .
4. SIGNIFICANT TYPES OF MODELS
4.1 Dynamic models
By a dynami c model i s meant a non- semi oti c obj ect 0 2 , whi ch is a ti me-
dependent model of an ori gi nal 01, whi ch is likewise ti me- dependent, and
bel ongs to the semi oti c zero-l evel D (cf. p. 150). The attri bute cl asses
consti tuti ng 0 1 and 0 2 are attri buti ve systems consi sti ng onl y of zero-
order and fi rst-order attri butes. The predi cate cl asses descri bi ng O 1 and
O 2 are therefore fi rst-order predi cati ve systems, whi ch are symbol i zed by
Ol(l.ni) and 02(19%2) respecti vel y.
A (fi rst-order) attri buti ve systemi s, in thi s connecti on, cal l ed ti me-
dependent (or dynami c), if at l east one el ement out of the compl ete cl ass
of i ts i ndi vi dual s, properti es, and rel ati ons i s ti me- dependent, i.e. changes
i n time. For the descri bi ng systems 51(1~%) and &(l7%) thi s means that a
t i me parameter occurs i n at l east one predi cate p 1 of Sl(lln1) and i n at
l east one predi cate p 2 = F(P1) of S2(17n2), wi th I cop S1(13n1), S2(19n). z
By further deEni ti ons wi thi n the dynami c type of system, or of model
respecti vel y, are speci fi ed mai nl y the open systems and thei r dynami c
model s. Her e L. von Bertal an ys i nvesti gati ons can deal wi th the
cl ari fi cati on of concepts such as fi nal i ty, equi -fi nal i ty, adaptati on,
i ntegrati on and di fferenti ati on, central i zati on and decentral i zati on,
growt h, decay, etc. of systems.
x
* The limitation of the space available does not allow to explain the notion of sys-
tem, used here intuitively. An explication was to be given in point 1.3, which had,
however, to be cancelled. Of course, attribute systems are a special kind of attribute
classes.
Models 165
4.2 Cybernetic models
By a cyberneti c model (in a restri cti ve scnse) is meant a dynami c model
of a cyberneti c attri buti ve system. In thi s connecti on an attri buti ve
syst em01 i s cal l ed cyberneti c, if 1) at l east one (not empty) subset of the
set of i ndi vi dual s of O1 consi sts of i ndi vi dual s that are ti me-acti ve el e-
ments, so that a change i n the state of the i nput of an el ement i na certai n
reacti on t i me bri ngs about a change inthe output of thi s el ement, and if
2 01 i s a stabl e - ergodi c - system, i.e. a systemwhi ch, if i t is not yet in
equi l i bri um, wi th each sequence of states (begi nni ng wi thi n i ts domai n of
stability) tends to an equi l i bri um.
A necessary condi ti on for the stabi l i ty of 01 is the exi stence of at l east
one f eedback coupl i ng in 01, i.e. at l east one cl osed chai n of ti me-acti ve
el ements, so that outputs of one el ement of thi s chai n are i nputs of an
el ement precedi ng i t in the same chai n. In stabl e systems the f eedback
coupl i ng must , moreover, be compensatory agai nst di sturbances.
The reader will fi nd further i nformati on in Lange (1962; 1965)
to whomwe owe a mathemati cal theory, rather hi ghl y evol ved al ready,
of the dynami c and especi al l y of the cyberneti c systems.
4.3 Si mul ati on models
Si mul ati on ina mor e comprehensi ve sense is the i mi tati on, consi derabl y
shorteni ng, ingeneral , of a (dynami c) ori gi nal wi th an al terati on of i ts
materi al qual i ti es. Si mul ati on ina restri cti ve sense i s l i mi ted to cyberneti c
ori gi nal -systems (cf: 4.2). Accordi ngl y, by a si mul ati on model (in a
restri cti ve sense) i s meant a cyberneti c anal ogy model (p. 163)-i n general
reduci ng i ts ori gi nal consi derabl y - of a (cyberneti c) ori gi nal -system.
Wi t h the ori gi nal -descri bi ng, or model -descri bi ng respecti vel y, predi cate
systems, the anal ogi zi ng transference of meani ngs f r omthe ori gi nal
attri butes to the model attri butes has i ts f ormal expressi on in the
transcodi ng of all encoded predi cates of the pre- domai n of the corres-
pondi ng i costructural ori gi nal - model - mappi ng.
Normal l y, the producti on of si mul ati on model s and operati ons on
such model s are not i mmedi atel y perf ormed by the model l i ng Kybi ak-
organi sms, but by i nterposed automati c si mul ators. Anal ogous si mu-
l ators generate sequences of si mul ati on model s, whi ch map conti nuousl y
changi ng ori gi nal syst ems. Di gi tal si mul ators perf ormthe mappi ng
166 Herbert Staclzowiak
of thei r ori gi nal s (whi ch themsel ves can be, for i nstance, physi co-
techni cal cyberneti c functi on model s of bi o-cyberneti c systems) by
means of comput er programmes.
Another cl assi fi cati on of the si mul ators fol l ows the di fferent val ues of
the ti me-scal e mt concerni ng the si mul ated and the si mul ati ng system.
mt = 1 i s spoken of as real -ti me si mul ators, mt 5 1 can be spoken of as
hi gh- speed- moti on si mul ators or sl ow- mot i on si mul ators respecti vel y.
The present report partl y anti ci pates resul ts whi ch the author will
submi t i n hi s book General Model Theory, soon to be publ i shed by
Spri nger-Verl ag, Vi enna and NewYor k.
REFERENCES
BAR- HI LLEL, Y. and CARNAP, R., (1953) Semant i c i nformati on , Brit. Journ. Philos. of
FSGEN, P., (1959) Model l e und Model l regel n , Wehrtechn. Monatshefte, 56,241-251
GNG, P. (1967) Pragmati sche I nf ormati on , Grundlagenstud. Kyb. Geisteswiss., 8,
LANGE, O., (1962) Calo,? i rozwj w iwietk cybernetyki, War saw, Pal i stwowe Wydaw-
ni ctwo Nakowe; Wholes and Parts. A General Theory of System Behaviorw, Oxf ord-
London- Frankf urt , Per gamon, 1965 (inEngl i sh).
MARTI N, R. M. , (1959) Toward a Systematic Pragmatics, Amst er dam, Nor t h Hol l and.
Publ . Co.
MAXTI N, R. M. , (1964) Towar d a l ogi c of i ntenti ons. I n: J. R. Gr egg and F. T. C.
Harri s (eds.), Form and Strategy in Science, p. 146-167, Dordrecht , D. Rei del .
MOLES, A. A. and SCH~TZENBERGER, A. , (1955) Soci omtri e et crati vi t, Rev. Psych.
Appl., V, 3.
MOLES, A. A. , (1958) Thorie de linformation et perception esthtique, Pari s, Fl amma-
ri on.
MOLES, A. A. , (1960) ber konstrukti onel l e und i nstrumentel l e Kompl exi tt , Grund-
lagenstud. Kyb. Geisteswiss., 1,33-36.
MOLES, A. A., (1962) Produkte: i hre funkti onel l e und strukturel l e Kompl exi tt , Ulm
(J. Hochschul e fiir Gestal tung), 6,412.
STACHOWI AK, H. , (1965) Gedanken zu ei ner al l gemei nen Theori e der Model l e ,
Studium Generale, 18, 432463.
STACHOWI AK, H. (1969) Denken und Erkennen in kybernetischen Modellen, Vi enna-
New Yor k, Spri nger (second edition).
Science, 4, 147-157.
(first publ i cati on in 1939).
77-90.
9 Management in cybernetic terms
STAFFORD BEER
1. THE HISTORY AND NATURE OF CYBERNETICS
The sci ence of cyberneti cs has been devel oped under that name for a
quarter of a century. It was defi ned withuni que cl ari ty and authori ty by
the l ate Norbert Wi ener to mean the sci ence of control and communi ca-
ti on inthe ani mal and the machi ne.
Thi s statement is pregnant wi th meani ng: a meani ng l atent rather than
hi dden. Peopl e have been sl owtounderstand it, al though i t has been there
all the ti me. The poi nt i s thi s. There are general l aws whi ch govern control
processes, whatever the systemunder governance. These l aws appl y to
computers and servomechani sms, to the human nervous system, to
popul ati ons of ani mal s, to the economy, and to every other l arge, compl ex
probabi l i sti c system- such as the busi ness fi rm. We i ncl ude the adj ec-
ti ve probabi l i sti c inthi s l i st of words qual i fyi ng system , si nce even well-
speci fi ed systems that are trul y compl ex can inpracti ce be descri bed onl y
inthi s way.
A full hi stori cal revi ewof the subj ect woul d systemati cal l y reveal how,
qui te gradual l y, sci enti sts expert i nmany branches of knowl edge came to
understand the i nvari ant nature, at an abstract l evel , of the processes
whi ch control l arge systems. Wi th hi ndsi ght, we can be mor e bri ef. It turns
out that we recogni ze systems tobe such for f our mai n reasons:
- because they cohere wi thi n some f rame of experi ence;
- because they survi ve through timewi thi n some appropri ate defi ni ti on
of conti nued i denti ty;
- because, to achi eve these ends, they prescri be unt o t hemsel ves
certai n rul es of equi l i bri a1 acti vi ty whi ch are tol erabl e to thei r
conti nued exi stence;
- and because they assi mi l ate thei r unfol di ng experi ence i nto sel f-
regul ati ng processes of l earni ng, adaptati on and evol uti on.
168 Stnfloord Beer
I f these are the common characteri sti cs of l arge vi abl e systems, then
cyberneti cs sets i tsel f to determi ne how the rel evant mechani sms wor k.
Havi ng understood somethi ng about mechani sm, sci ence may then be
rel i ed upon to general i ze i ts understandi ng - across the vari ous types of
systemthat i t studi es. Thi s process l eads to concepts of l aw, whereby
sci ence makes a further general i zati on: we expect that any newcompl ex
vi abl e systemthat i s goi ng to survi ve will be f ound to utilizethe general -
i zed mechani sms of survi val al ready el i ci ted. But to make the concepts of
mechani smand of l awef f ecti ve i ndesi gni ng or redesi gni ng systems, we
shal l need concepts of behavi our, by whi ch to understand the practi cal
appl i cati on of the model s whi ch embody our l aws and our mechani sms.
All of thi s means that the scientific apparatus requi red to understand,
desi gn and regul ate l arge vi abl e systems is becomi ng avai l abl e. It is thi s
very apparatus, based on a corpus of knowl edge, of whi ch the manage-
ment communi t y stands most i nneed. For if cyberneti cs is the sci ence of
control , management i s the professi on of control . Every manager, whether
he runs the firmi tsel f or a maj or department of government, whether
he runs the country or an aspect of i nternati onal affai rs, faces an i denti cal
probl em. He faces, that is, the need tomai ntai n a vi abl e systemfar mor e
compl i cated than he personal l y can understand. And the begi nni ng of
wi sdomfor management at any l evel i s the real i zati on that vi abl e systems
are, inl arge measure, sel f-regul ati ng and even sel f-organi zi ng.
Cyberneti cs reveal s the nature of these natural phenomena. It must do
so, if i t is to hel p at all. For al though management must accept respon-
sibility for everythi ng that happens, i t cannot assume di rect autocrati c
control of everythi ng that happens. The systems concerned are j ust too bi g.
This is why cyberneti cs turned to the study of sel f-regul ati ng systems
(such as those f ound in the rel ati onshi p between organi sms and their
envi ronment) and to sel f-organi zi ng systems (such as the brai n) for i ts
i nsi ghts. There si mpl y i s no manager i denti fi abl e inan ecol ogi cal or a
neurol ogi cal system. Yet bot h of t hemwork. Management theory, say
cyberneti ci ans, has been l ed astray instudyi ng the nature of the exi stent
and accepted manageri al process itself. To make studi es of the manager
was the obvi ous thi ng to do; and yet the manager turns out to be the
embodi ment of somethi ng we mi ght call management pri nci pl e -
that set of characteri sti cs di scovered inany vi abl e systemwhi ch conduces
to vi abi l i ty. I n thi s sense, then, the peopl e who call themsel ves managers
are real l y catal ysts of a systemi c metabol i smwhi ch is al ready at work.
Management in cybernetic terms 169
But of coursethi sconsi dersmanagements as somethi ng whi ch happens to be
there inan on- goi ng system, and somethi ng whi ch is acknowl edged by the
system s component s as havi ng authori ty. The maj or probl ems ari se when
matters are not goi ng well - or at l east when i t is obvi ous that somethi ng
needs to be done . Managers, the men themsel ves, are then expected to
take some ki nd of acti on.
I thi nk i t i s a maj or cyberneti c concl usi on to dr awf r omthese remarks
that managers general l y approach thi s probl eminthe wr ong way. They
usual l y try to i ntervene inthe equi l i bri a1 processes of the sel f-regul ati ng
system- thereby, perhaps, maki ng i t f undamental l y unstabl e. The
sensi bl e course for the manager i s not to try to change the system s
i nternal behavi our whi ch typi cal l y resul ts i n mammot h osci l l ati on, but to
change i ts structure - so that i ts natural systemi c behavi our becomes
di fferent. All of thi s says that management is not so much part of the
systemmanaged as i t is the system s own desi gner.
What are the aspects of the systemwith whi ch management can
sensi bl y i nterf ere, whi ch i t can desi gn - or re-desi gn? They are the
mechanisms of that system: the structures, and the rul es governi ng the
behavi our of structures, whi ch are usual l y taken as gi ven. They are the
arrangements al ready accepted as i nsti tuti onal conventi ons. It is these
that need i nvesti gati on.
2. CONCEPTS OF MECHANISM
One of the maj or mechani sms, perhaps the most i mportant of all, whi ch
operates inthe sel f-regul ati on of systems is the mechani smof feedback.
This t er mhas become very popul ar, and is often wrongl y - because too
l oosel y - used. Expl i ci tl y: f eedback does not ref er to a response provoked
by a sti mul us. (Managers commonl y say: I have had some f eedback
about my suggesti on on so and SO.) Thi s usage perverts a cri ti cal l y
i mportant i dea.
The i dea is this. When a systemi s i nfull operati on i t produces an output.
Dependi ng on the viewwe take of thi s output, we shal l feed back a si gnal
to those el ements of the systemwhi ch govern the nature of the output -
a si gnal whi ch actual l y al ters them. When the systemat l arge - as i t
al ready exi sts - next operates on i ts i nputs, i t will procure a di fferent
output - one hopeful l y nearer to our desi re. The mai n reason why thi s
mechani smi s so i mportant is that i t does not requi re us ei ther to under-
170 Stufford Beer
stand or to i ntervene inthe maj or operati ons of the system. These are l eft
al one, and treated as sel f-control l i ng. What we are doi ng i s to take a
l ogi cal l y (and not necessari l y manageri al l y) superi or vi ew. I f we have a
perfectl y good systemthe outputs of whi ch are al ways exactl y hal f what
we requi re, then we alter the i nputs tothat system(inthe si mpl est case)
by a factor of t wo. Then the perfectl y good system, qui te unaffected by
thi s, produces the anwers we want. It has not been denatured, or t hrown
i nto conf usi on, inthe process.
The si mpl est f or mof f eedback to understand i s cal l ed error-actuated
negati ve f eedback, whi ch i s the one j ust menti oned. I n thi s si tuati on, we
have an expl i ci t statement of what the output i s i ntended to be. A profi t
of ten million dol l ars, a fi fteenpercent return on capi tal , are exampl es
of such a cri teri on inthe firm. I f the output of the systemis di fferent f rom
thi s, a devi ce i s needed whi ch measures the di fference between the actual
and the i ntended resul t. Thi s error is fed back to the start of the system,
where i t i s enabl ed to change certai n parameters of the system, so that the
resul t - next ti me r ound - is nearer to the target than before. So far, so
good. But there are other aspects of f eedback, of whi ch the noti on of
ampl i fi cati on i s the most i mportant. I nstead of a fi xed output cri teri on,
we may have si mpl y a statement of preferred trend. Let us i ncrease the
profi ts i s an exampl e. Then, if the systemis tendi ng to i ncrease the profi t,
the positive f eedback si gnal encourages the system s cri ti cal vari abl es to
enhance yet further the fi nal effect. Thi s i s cal l ed a reward system(its
negati ve is a puni shment system), the obj ecti ve of whi ch is to cause the
systemto learn f r omi ts past experi ence. These are unwor ked exampl es,
but they i l l ustrate the vital noti on.
The whol e poi nt about any f eedback mechani smis that, havi ng set i t
up, the manager may go away. After all, he cannot attend toeverythi ng
si mul taneousl y. When he has set up a useful f eedback mechani sm, he
shoul d be abl e to forget about the subj ect; because what hzppens will
henceforth regul ate itself.
Now al though we may thi nk of a l arge organi zati on, such as a firm, as
havi ng a si ngl e output (l abel l edprofi tabi l i ty for exampl e), and al though
we may i magi ne a l arge f eedback l oop whi ch woul d adj ust the i nputs to
the systemso as to keep the output at a desi red l evel , thi s is no mor e than
a conceptual i zati on of how management works. It is i nconcei vabl e that
the compl exi ti es of a company s operati ons coul d actual l y be stuffedi nto
a si ngl e mechani smof thi s ki nd. But we may start by t hi s met hod to
Management in cybernetic terms 171
account for the way the busi ness works. Then we ask questi ons about the
way the systembreaks up - i nto di vi si ons of the company and di vi si ons
of themarket, for i nstance, and model those. Thus we shal l gradual l y be
abl e to devi se a mor e compl i cated model , redol ent withf eedback l oops,
whi ch i s of practi cal val ue. I n doi ng all thi s we pass f r omthe noti on of a
strai ghtforward f eedback mechani smto the noti on of mul ti pl e l oop
systems. Outstandi ng i nthi s area of study i s the wor k of J. W. Forrester
(1961; 1969).
There are three basi c characteri sti cs of vi abl e systems whi ch can prof-
i tabl y be studi ed by cyberneti cs in t er ms of mul ti pl e f eedback l oops.
First, the si gni fi cant outputs must be under control : we must know how to
mani pul ate the systemso as to produce a desi red effect. Second, and thi s
is vi tal , the systemas a whol e must be stabl e. I nstabi l i ty is the maj or
sympt omof a badl y- managed organi zati on, and we detect thi s sympt om
i n t erms (for exampl e) of wi l d stock fl uctuati ons; of surges of work- i n-
progress; of recurri ng pani c inthe conduct of day- to- day busi ness; and
of stop- go pol i ci es inthe area of devel opment. But cyberneti cs has i ndeed
i ntensi vel y studi ed the mechani sms whi ch govern equi l i bri a1 behavi our -
and can i ncorporate themi n i ts management model s.
The maj or concept whi ch emerges f r omsuch studi es is that of homeo-
stasi s. Thi s means the capaci ty of a systemto hol d i ts cri ti cal vari abl es
wi thi n physi ol ogi cal l i mi ts. Not e that we do not say: wi thi n prescri bed
limits. The poi nt i s that there i s no sati sfactory way of speci fyi ng accept-
abl e limits of vari ati on, except through a study of the system s dynami c
behavi our. Thi s shoul d l ead to the di scovery of that degree of vari ati on
for critical parameters whi ch the organi smas a whol e can tol erate before
i t is denatured.
It turns out that we can desi gn a si mpl e system, and prescri be limits for
i t, because we understand i t compl etel y. We know how i t works, and we
know both the ki nd and the magni tude of the perturbati ons i t must
wi thstand. This desi gn-process yi el ds a stabl e system: thi nk of regul ati ng
a central heati ng i nstal l ati on for exampl e. When systems become
unthi nkabl y compl ex, however, we can be sure of none of these thi ngs;
we do not real l y understand the process itself, nor the perturbati ons
whi ch affect i t, nor the i nteracti on of these t wo. Theref ore we adopt the
pri nci pl e that we need not onl y sel f-regul ati on (the f eedback mechani sm),
but a capaci ty to generate i nternal el ements of control whi ch can adj ust
to everythi ng el se that i s goi ng on.
172 StafSord Beer
The best known exampl e of homeostasi s, whi ch i l l ustrates thi s, i s the
control of body temperature. Thi s is achi eved by myri ad i nteracti ons
wi thi n the body, taki ng account of the total set of ci rcumstances f r om
moment to moment , and not by some ki nd of thermostat worki ng to
fi xed temperature l i mi ts. That is why we say that homeostati c control l ers
use sel f-organi zi ng pri nci pl es to hol d critical vari abl es wi thi n physi ol og-
ical l i mi ts. And so we have i denti fi ed a sel f' -organi zi ng characteri sti c (the
mechani smof whi ch i s homeostasi s). The wor k of W. Ross Ashby (1954)
provi des the management cyberneti ci an wi th f undamental knowl edge on
thi s score: he knows now how to go to work.
We sai d earl i er that a rel ati vel y si mpl e mul ti pl e l oop f eedback system
coul d be desi gned - infull knowl edge - whi ch coul d be stabl e. In desi gn-
i ng very compl i cated systems (wi thout full knowl edge) to be sel f-
organi zi ng by homeostasi s, we emerge witha newcri teri on: ul trastabi l i ty.
Thi s cri teri on rel ates to the capaci ty of a systemto wi thstand pertur-
bati ons whi ch have not been foreseen by the desi gner.
This capabi l i ty at fi rst sounds i mpossi bl e to real i ze. That i s because we
take the wr ong model : a model dr awn by our trai ni ng f r omenergy-
systems (with whi ch we are fami l i ar) rather than f romi nf ormati on
systems (wi th whi ch we are not). All we are real l y doi ng i s to construct
control devi ces whi ch recogni ze unknown threats to the system- through
thei r prel i mi nary effects on the smoothness of operati ons. Gyroscopes
and engi ne governors, for i nstance, are exampl es of homeostati c
control l ers. They are i ntended to keep shi ps and ai rcraft, or engi nes,
runni ng ina regul ated way - regardl ess of the reasons why the shi p is off
course, or the engi ne raci ng. They achi eve thi s, not by anal ysi ng the
causal i ty of the si tuati on and comi ng to t er ms wi th the root probl em, but
by detecti ng wi thi n t hemsel ves pathol ogi cal sympt oms whi ch are di rectl y
used to take regul atory acti on. I have cal l edthi s pri nci pl e ' i ntri nsi c control ' .
Si mi l arl y inthe firm, we requi re manageri al control devi ces to damp
down osci l l ati ons (such as those recorded above as sympt oms of
instability), regardl ess of thei r cause. Thi s i dea is very i mportant in
management , because i t may often take too l ong to i denti fy the cause of
troubl e and to correct i t at source. The common bel i ef that this i s the onl y
sci enti fi c way of proceedi ng is mi staken, and deri ves f roma very ol d-
f ashi oned vi ewof sci ence itself. Keep the engi ne governor inmi nd.
So far we have concentrated on cyberneti c mechani sms i ntended to hol d
thi ngs steady, to keep outputs and critical vari abl es under control , to
Management in cybernetic terms 173
make systems at l arge both stabl e and ul trastabl e. All these thi ngs are
necessary to managers if they are not to l i ve in a state of conti nuous
crisis. But there i s another si de to the pi cture. The ul ti matel y stabl e state
is death. If we are not careful , we shal l have devi sed a paradi gmof the
firmwhi ch i s a reci pe for peaceful demi se. All vi abl e systems, cyberneti cs
poi nts out, are geared not onl y to tol erate a degree of tensi on, but to make
constructi ve use of i t.
Muscl e tone makes a good exampl e of this tensi on-by-anti thesi s inthe
human body. Wi thout i t we shoul d j ust fall down. Theref ore we look for
mechani sms in management systems whi ch provi de anti theti c i nternal
stresses, mai ntai ni ng tonal i ty short of the cri si s l evel . Fr omthese
mechani sms spri ng others whi ch al one can l ead the firmf orward. It i s a
cyberneti c resul t that systems cannot l earn unl ess they make errors;
they cannot gr owwi thout mutati on; they will not adapt unl ess they
experi ment; they never evol ve unl ess l earni ng, growi ng and adapti ng.
ALI thi s, and the cyberneti c understandi ng of i t, means that i t i s now
possi bl e to desi gn i nto a management systema proper machi nery for
generati ng i nnovati on. That is the r awmateri al of evol uti onary devel op-
ment, and of the company s growth.
3. CONCEPTS OF L AW
When sci ence has acqui red an understandi ng of mechani sms whi ch .
appear to be general to i ts area of study, i t seeks out the common
pri nci pl es whi ch must underl i e thei r operati on. For exampl e, a tour of i n-
dustry woul d reveal that (whereas many obj ects ar emovedabout byapower
source) in al most every factory there are exampl es where obj ects are
del i beratel y and successful l y moved i n the absence of a power source.
Fr omthe observati on of these i nstances we shoul d (if we wer e sci enti sts
f romMar s, whol l y unfami l i ar withterrestri al mechani cs) be abl e to i nf er
the general pri nci pl e of gravi ty feed. And f r omhis we shoul d inturn
suspect the exi stence of the l awof gravi ty itself. So i t is wi th cyberneti c
mechani sms.
The stuff of control is not any ki nd of obj ect; i t i s somethi ng cal l ed
variety. Vari ety is defi ned as the number of possi bl e states of a system.
I n what sense can this mathemati cal construct be cal l ed the very fabri c
of system? Consi der the el ectri c l i ght on your desk. Its vari ety i s t wo,
174 Stafsord Beer.
because i t is ei ther on or off. The control of this si tuati on i s a si mpl e
matter, because we have a swi tch mar ked on and off. Consi der a
l abour force of ten men, operati ng under a rul e that anyone who di d not
i mmedi atel y do what he was tol d woul d be shot. Thi s systemhas a
vari ety of ten and i s readi l y control l ed by anyone wi th a gun and ten
bul l ets. Now rel ax the constrai nt that di sobedi ence means death. At once
the vari ety of this t en- man t eamri ses to somethi ng approachi ng i nfi ni ty.
These men may now do anythi ng at all: obey orders, di sobey orders, vary
orders, i gnore orders. Or they may go away and pl ay cards.
I n short, the cyberneti ci an calls any ordi nary si tuati on, inwhi ch an
enor mous number of possi bl e states mi ght be real i zed, a vari ety generator.
The management probl emis preci sel y a probl eminhandl i ng vari ety: if
we exami ne any manageri al acti on, we shal l fi nd that i t is a vari ety
reducer. I n the l i mi t, when any parti cul ar goal has been achi eved (for
exampl e, the order has been accepted by the customer) there i s no vari ety
left. The systemhas onl y one state: i t i s success, and that i s a termi nal
state.
Now the ki nds of acti on open to managers appear to be many and
vari ed. There are hi gh-l evel pol i cy deci si ons; there are l ow-l evel short-
t ermi nstructi ons; there are deci si ons about organi zi ng the firm; there are
verdi cts whi ch choose between al ternati ve courses of acti on. It i s i nterest-
i ng that so many di fferent manageri al acti vi ti es, categori zed as they
normal l y are by functi on and by rank, di scussed as they normal l y are i n
t er ms of di fferent cri teri a appl i ed by wi del y di fferi ngtechni ques, shoul d
share a common measure of success. Thi s i s preci sel y the measure of
vari ety reducti on. It will al so qui ckl y be noted that the hi gher-l evel
deci si ons el i mi nate mor e potenti al vari ety than the l ower-l evel deci si ons.
These observati ons i ndi cate that i t mi ght be possi bl e to desi gn an
organi zati on, and withi t a systemof management control , based on the
measure of vari ety. And so i t proves tobe. Now l et us l ook for the natural
l aw, exceedi ngl y si mpl e but exceedi ngl y potent, whi ch actual l y governs
vari ety regul ati on. We have cl ues to i ts f or mal ready inthe two-state
el ectri c l i ght wi th i ts two-state swi tch, and i n the t en- man teammade
obedi ent by ten bul l ets. The pri nci pl e i s thi s: onl y vari ety can absorb
vari ety.
Once thi s is sai d i t appears very obvi ous. But there is no difficulty in
thi nki ng of i nnumerabl e everyday si tuati ons in whi ch peopl e behave as
if i t were not the case. As to management , the appl i cati on i s cl ear: the
Management in cybernetic terms 175
manageri al system, taken as a whol e, must be capabl e of generati ng as
i nuch control vari ety as the si tuati on i s capabl e of prol i ferati ng un-
control l ed vari ety. When these condi ti ons are not met, and often they are
not met, unt oward events occur whi ch the management cannot bri ng
back i nto l i ne.
Exampl es of wel l - known attempts to control prol i ferati ng si tuati ons
wi th vari ety-starved control l ers are l egi on. For exampl e, we fi nd govern-
ments attempti ng to control a vari ety-ri ch economy by changes i nbank
rate; we fi nd managi ng di rectors ( who shoul d know better) setti ng out
si mpl e admi ni strati ve rul es about the way the company shoul d run -
rul es whi ch are i mmedi atel y and readi l y ci rcumvented by peopl e l ower
down, because they operate in a vari ety-ri ch envi ronment. We fi nd
pol i cement attempti ng to regul ate traffic at i ndi vi dual cross-roads, when
thei r i nf ormati on does not extend around corners - never mi nd to the
next pol i ceman at another cross-roads. And so on. None of these
managers can possi bl y be incontrol ; they are attempti ng to di sobey a
natural l aw. I repeat: onl y vari ety can absorb vari ety. Thi s i s known as
Ashby s Law of Requi si te Vari ety (1954).
If a manager must needs depl oy as much vari ety as the si tuati on he
seeks to bri ng under control , how is he to do t? The si mpl est met hod i s
by produci ng a preci se mat ch between the vari ety of control and of the
control l ed - as we sawin our prel i mi nary exampl es. Most games are
organi zed on thi s basi s - we have the sai ne number of pl ayers on each
si de, havi ng the same resources. If we move toa mor e seri ous f or mof the
game , such as a war, we can see how the same rul e i s appl i ed. As ci vi l i an
lifebecomes i ncreasi ngl y compl ex, whi ch means that every i ndi vi dual has
hi gh vari ety ( compare feudal ti mes), i t becomes i ncreasi ngl y necessary to
mat ch the total vari ety of hal f the citizens wi th a control vari ety consi sti ng
of the other hal f. Today we have j ust about reached the poi nt where the
pol i ci ng of the state - int er ms of taxati on, wel fare, educati on and so
forth, as well as cri mi nal i ty - consumes hal f the total effort of soci ety.
All this may seemto be necessary, if Ashby s Law i s to be obeyed. But
that is ilot the case. There is another way to depl oy requi si te vari ety, and
that i s togenerate matchi ng vari ety through a vari ety ampl i fi er. If we say
that all traffic will keep to one si de of the road, we have set forth a l ow-
vari ety rul e. Thi s is mat ched, i n any i ndi vi dual case, by a l ow-vari ety
acqui escence: I will do it. But if everyone is trul y acqui escent, then the
control vari ety has been ampl i fi ed by the number of i ndi vi dual s inthe
176 Stayflbrd Beer.
system. Shoul d the popul ati on of dri vers at whi ch this rul e is ai med not
acqui esce, then (natural l y enough) the l ow-vari ety rul e will not wor k. We
shoul d need a hi gh-vari ety rul e such as a l awenf orcement officer dri vi ng
inevery car.
Thus all management s are faced witha triple choi ce: they may absorb
prol i ferati ng vari ety by the ampl i fi cati on of control vari ety, or by one- t o-
one matchi ng, or by reducti on of uncontrol l ed vari ety. The fi rst met hod
has the demeri t that a hi gh-vari ety si tuati on overthrows the control
systemcompl etel y - in the si mpl e act of di sobedi ence. The second
met hod appears to have the demeri t that i t ipsofacto uses up hal f one s
total resources. The thi rd met hod works very well; but i t reduces
opportuni ty and i s i ni mi cal to maj or i nnovati on.
But once agai n there is an al ternati ve sol uti on. Remember the game
si tuati on inwhi ch one t eamabsorbs the vari ety of the other t eam. This
control devi ce does not consume manageri al vari ety at all. It uses the
rul e of j udo: one s opponent is defeated by his own strength, not yours.
Most management si tuati ons can be defi ned i n t er ms of t wo anti theti c sets
of acti vi ty. For exampl e, the company and i ts market, each generati ng
very hi gh vari ety, can be model l ed by a homeostat inwhi ch each set is
del i beratel y organi zed as a vari ety sponge vis--vis the other set.
Then we are presented wi th a further probl em. We may certai nl y
concei ve of t wo aspects of the si tuati on for whi ch, as managers, we are
responsi bl e, as absorbi ng each others vari ety. And we may take organi -
zati onal steps to ensure the accompl i shment of t hi s pl an - whether we use
vari ety generators, vari ety absorbers, or vari ety reducers. But recal l what
was sai d inthe last secti on about the homeostat as a machi ne for runni ng
to a standsti l . The homeost at is itself a control l er of vari ety. But unl ess
there is a control l er of the homeostat, we shal l not be abl e to i ntervene in
the si tuati on as managers at all.
Not that managers do not attempt thi s feat. A seni or manager often has
the noti on that he may i ntervene in the homeostati c systems whi ch
operate under his aegi s. He has the authori ty to do so, of course. But the
mi nute he di rectl y engages ina hi ghl y compl ex si tuati on, on l evel terms
as i t were wi th those whose i nteracti ons are perf ormi ng the bal anci ng
acti vi ty of the homeostat i tsel f, the seni or manager abandons his
ol yi i i pi an rol e. His own personal vari ety is that of a human bei ng,
however el evated his status. No: the rol e of the seni or manager i s to
remai n above t he homeostati c fray, and to consi der what i s happeni ng in
Munugemerat in cybernetic ternis 177
t er ms of hi s hi gher l evel understandi ng. Because he is outsi de the system,
i nfact, and because he partakes inanother systemwhi ch is no concern of
hi s subordi nates, hi s met hod of control i s expl i ci tl y to alter the cri teri a
accordi ng to whi ch the l ower l evel systemi s operati ng.
Let us go back to the i l l ustrati on f romgames. Suppose that as a hi gher
manager you have the responsi bi l i ty to ensure that t eamA wi ns in a
game whi ch i s al ready bei ng pl ayed between t eamA and t eamE, where
the scori ng is al ready even. You coul d dress yoursel f i nthe appropri ate
regal i a and charge onto the field of pl ay. The pl ayers woul d recogni ze
you. Your own si de mi ght defer to your tacti cs (but perhaps you are not a
very good tacti ti on?), whi l e the other si de woul d do thei r l evel best to p~it
you out of acti on. Thi s is not the way to behave at all. I f you real l y had
authori ty over this si tuati on, the cl ever acti on woul d be to change the
rul es of the game so that your si de must wi n it. You bel ong to a hi gher
order systemthan the game system; your i nf ormati on is better; you
command the facilities for vari ety generati on. Then do not act as if none
of thi s were true.
Thi s i l l ustrati on seeks todefi ne the noti on of metasystem. A metasystem
i s a systemover- and- above the systemitself. I ts maj or characteri sti c i s
that i t tal ks a metal anguage; and thi s is a ri cher, better i nf ormed, way of
tal ki ng than i s avai l abl e tothe systeml ower down. It shoul d be noted that
the raison dtre of the metasystemi s gi ven inl ogi c; i t i s not necessari l y
anythi ng to do wi th the hi erarchy of status.
4. CONCEPTS OF MODELLING
Havi ng tal ked about mechani sms and the l aws whi ch underl i e t hem, we
may be abl e to see how cyberneti cs provi des a basi s for model l i ng a
management system. Now a model is nei ther a l i terary devi ce, as is a
si mi l e, nor a l ogi cal devi ce, as i s an anal ogy. A model is a f ormal account
of a systemwhi ch i denti fi es how i t actual l y wor ks (see Beer, 1966).
I nteresti ngl y enough, the rul es of model l i ng turn out to dupl i cate the
general l aws of cyberneti cs whi ch we have al ready been di scussi ng.
Consi der: a model seeks to mat ch the vari ety of the si tuati on model l ed -
and i nso far as i t fails i t i s a l ess effecti ve model . Thus the i deal model of
anythi ng is i tsel f; whi l e a model al most as good as thi s i denti ty model is
one whi ch matches every el ement of the real systemwi th an el ement inthe
178 Stafford Beer
model . Obvi ousl y such an arrangement provi des requi si te vari ety inthe
model l i ng process. Techni cal l y, such a model is cal l ed isomorphic wi th
real i ty, because every el ement i s matched. A paste copy of a pi ece of
j ewel ry i s an i somorphi sm.
If we cannot have an i somorphi c model , and thi s i s i ndeed unusual , we
may have i nstead a model whereby many el ements i nthe real systemare
represented by one el ement in the model . For exampl e, if we make a
model of Shakespeare s bi rthpl ace whi ch i s about six i nches l ong, we may
fai rl y readi l y produce an i t emwhi ch i s i denti fi abl e as the bui l di ng in
Stratf ord- upon- Avon. But every bri ck i n our model stands for many
bri cks inStratford. It fol l ows, then, that we are total l y unabl e to produce
every feature of the real bui l di ng. Thi s representati on of the many by the
one, techni cal l y cal l ed a many- one mappi ng, is not an i somorphi smbut a
homomor phi smof the ori gi nal . We have l ost vari ety. But the i mportant
poi nt i s that - if we are suffi ci entl y cl ever - we will not have l ost vari ety
that matters.
Everythi ng here depends on the purpose for whi ch the model is con-
structed. Suppose we wi sh to make peopl e thi nk about Stratford. We
bui l d our si x-i nch model , and we produce a photograph of i t. Thi s
photograph, at a casual gl ance, i mmedi atel y says: Stratf ord- upon- Avon
to anyone who has seen Shakespeare s bi rthpl ace. So there may be no l oss
of i nf ormati on at all. A homomor ph hel ps to regenerate the vari ety of the
i somorph, when i t is correctl y used. But if we wi shed to know how l ong
i t will be before the Stratford bui l di ng falls down, nothi ng at all will be
gai ned by studyi ng degenerati on inthe model .
I n a management system, we know (or ought to know) the features of
the si tuati on that matter. Suppose we offer toprovi de the manager wi th a
systemwhi ch correctl y predi cts whether the man who will be operati ng the
l athe inthe thi rd shop on the left next Tuesday morni ng will be weari ng
br own boots. Thi s woul d be a hi gh-vari ety model i ndeed; but one woul d
certai nl y expect the manager tosay that he was not i nterested i nthe l east.
So i t i s cl ear that we may el i mi nate vari ety i n the process of model l i ng
provi ded that we do not el i mi nate i nf ormati on whi ch the manager most
wants i ncl uded.
I n order to achi eve these ends, we shoul d begi n wi th the macrostructure
of the systemwe are studyi ng. This systemhas maj or outl i nes. There are
mani festl y several i denti fi abl e processes, markets, trade uni ons, etc. Then
our model l i ng process begi ns by speci fyi ng the maj or i t ems and the
Management incyberneti c terms 179
connecti ons between t hem. Al ready we have a model , whi ch may be
consi dered compl ete, al though i t has no detai l ed i nfrastructure. The
model is homomor phi c, and thus of very l owvari ety compar ed wi th the
ori gi nal . Neverthel ess, the manager may well - even at thi s poi nt - be
abl e to recogni ze the si tuati on whi ch he must control .
It is easy to i magi ne that the f ormof t hi s model , writtenon a pi ece of
paper, consi sts of boxes connected by l i nes. Each box refers to a maj or
acti vi ty, and is a very hi gh vari ety box. Nonethel ess: there i t i s, wi th a
l abel on it. We know what i t i s, but nothi ng at all about what i s i nsi de i t.
Such boxes are cal l ed bl ack boxes , for the very si mpl e reason that they
are not transparent. We cannot see what i s goi ng on i nsi de. Never mi nd,
for the moment . These boxes are connected by l i nes, whi ch may be used
to i ndi cate all manner of thi ngs. Above all, however, they i ndi cate the
way inwhi ch the boxes are connected interms of communi cat i on between
them. Communi cat i on, here, may mean anythi ng f r oma producti on fl ow
l i ne (by whi ch one box certai nl y communi cat es wi th another) to a message
(about the size of the stocks for exampl e); or i t may be a mor e subtl e
f or mof communi cat i on, havi ng mor e potency, whi ch carri es f eedback
i nformati on.
What can we say about these bl ack boxes at this moment inti me? Very
little. But i t may well be that we have a measure of thei r vari ety. There-
fore, if one box i s supposed to handl e the output of another, we al ready
know that i t will fail i n thi s attempt unl ess i ts vari ety is as hi gh as the
second. That is an i mmedi ate i nference, by the Law of Requi si te Vari ety.
So al though the boxes are bl ack, they have a di sti nct persona wi thi n the
cyberneti c system.
Let us then turn to the lines by whi ch boxes are connected, whi ch we
have al ready sai d represent communi cati on. Now we may treat the prob-
lemof communi cat i on as a vari ety probl emal so. The l east effecti ve
means by whi ch t wo boxes can be rel ated int er ms of communi cat i on is
by noi se whi ch passes between them. Noi se is defi ned as a si gnal havi ng
no meani ng. But we must note in passi ng that even noi se i denti fi es a
channel - the path through whi ch the noi se has passed. When we manage
to i denti fy a pattern i nthe noi se, we decl are that somethi ng meani ngf ul
i s bei ng communi cat ed. Whatever i s passi ng through the channel mi ght
now be cal l ed data. Dat a i s pattern become meani ngf ul . But data of
themsel ves have no i nfl uence upon anythi ng - after all, they may well
be i gnored; they may be stored for further reference (whi ch never hap-
180 Stafford Beer.
pens); they may engul f the reci pi ent, as anyone who has bought a news-
paper or a book whi ch he has never read well knows. Noi se become data
becomes i nf ormati on at a preci sel y i denti fi abl e poi nt. I nf ormati on is
what changes us. We never change wi thout an i nf ormati on i nput ( why
shoul d we?) If we real l y have recei ved i nf ormati on, we are bound to
change - because our vari ety is thereby i ncreased.
J ust as we can i nfer a great deal about our si tuati on, usi ng a model in
whi ch all the boxes are bl ack, so we can i nfer a great deal about our
capaci ty to react by l ooki ng at the communi cat i on channel s whereby the
boxes are connected. Do they contai n merel y noi se? Are there receptors
at the other end whi ch are capabl e of transf ormi ng noi se to data? Do
thi nki ng el ements lie behi nd these i nput transducers whi ch are capabl e
of transf ormi ng data to i nf ormati on? All these are hi ghl y rel evant
questi ons. But there i s mor e to say. Consi der t wo bl ack boxes, A and B,
whi ch are mutual l y i n a state of homeostati c equi l i bri um. What can we
say about the communi cat i on channel s whi ch connect t hem? Wel l , we
know that each box must have roughl y the same vari ety as the other. And
thi s equati on woul d be usel ess if the communi cat i on channel s wer e not
adecpate to transmi t that vari ety - inboth di recti ons.
By these means we are abl e to i denti fy the structure of communi cat i on
channel s whi ch will compl ete the necessary stabi l i zi ng l oops, and be
abl e to speci fy thei r measurabl e capaci ty. There i s a compl ete mat hemat -
i cal theory of communi cat i on, due especi al l y to Cl aude Shannon (1949),
whi ch provi des us wi th many f ormal t heorems for the el uci dati on
of communi cat i on probl ems. It sounds trivial to say that the chan-
nel capaci ty must be abl e to handl e the vari ety of the bl ack boxes. But
this rul e is often di sobeyed inpracti se - wi th di re consequences. I f we
consi der the managi ng di rector of a firmand one of i ts most humbl e
empl oyees, what coul d we possi bl y say about the communi cat i on
bet ween these t wo men? The answer is l i kel y to be that they cannot
(not do not, but cannot) communi cat e at all - because the channel
capaci ty is si mpl y not there. Shannon s t heorems el uci date much of
what the cyberneti ci an needs to know in hi s model l i ng process. Fr om
his poi nt of vi ew, at l east, one of the most i mportant resul ts of the
mathemati cal theory of communi cat i on is Shannon s tenth theorem.
Thi s says, in effect, that i t i s not enough to provi de channel s wi th a
capaci ty to transmi t the vari ety of the bl ack boxes. We need extra-
channel capaci ty to el uci date ambi gui ti es inthe message.
Management in cybernetic terms 18 1
Fr omall of this i t fol l ows that if one i s model l i ng a management
systemeven inthe most crude and macrocosmi c f ashi on, a great deal can
al ready cyberneti cal l y be sai d about i ts structure, and even about the
way i t wor ks and fails to wor k. Bl ack boxes j uxtaposed inhomeostati c
equi l i bri ummust have requi si te vari ety; channel s connecti ng themmust
obey the l aws of communi cat i on; and we shal l al ready be l ooki ng for
emergent metasystems. All of this can be sai d before we have even ap-
proached the model l i ng of detai l ed operati ons wi thi n the fi rm. Perhaps this
will never be necessary: we may settl e for a l ow-vari ety homomor phi sm,
dependi ng on the obj ect of the exerci se.
Mor e i mportant than the detai l s is the structure. We shal l soon fi nd that
we are bui l di ng a hi erarchy of systems, based on the pri nci pl es al ready
uncovered. For every systemdemands a metasystem; and therefore a
second metasystemwill be i denti fi ed beyond that. There i s no l ogi cal end
to the chai n, because at each l evel of l anguage there are proposi ti ons
whi ch l ogi ci ans call undeci dabl e except i nt er ms of a metal anguage.
Then when shal l the process cease? How do we know when to stop
model l i ng? The answer to this l i es inthe concept of the bl ack box once
agai n. For, when we have reached the limits of the systemwe are studyi ng,
we are compel l ed to say that i t subsi sts wi thi n the f ramework of a
hi gher order system- whi ch we have no bri ef to penetrate. Thi s does not
mean to say that i t can be i gnored. What we must do is treat this extra-
systemi c systemas an encompassi ng bl ack box - and l ook for i ts effects
upon the total i ty whi ch l i es under our j uri sdi cti on that cannot be deci ded.
Then we have to accept the resol vi ng i nput of the superi or bl ack box as
gi ven, as i nsuscepti bl e to anal ysi s.
Now thi s statement is true for every systemgoverned by a metasystem.
In so far as we are desi gni ng a total management structure, we oursel ves
understand everythi ng. But we must real i ze that each metasystemexi sts
preci sel y because there is no l ogi cal sense i nwhi ch the systemto whi ch i t
is met a can cope wi thout this unanal ysabl e hel p. Al though, i nl ogi c, thi s
i s a matter of some theoreti cal sophi sti cati on, i t seems l i kel y that real
managers usual l y understand the poi nt very wel l . When one expl ai ns
thi ngs to a chi l d, one perforce speaks the l anguage of the chi l d; and one
may have to do vi ol ence to ones own i nsi ght in so si mpl i fyi ng the
expl anati on gi ven that i t is al most i ncrorrect. Si mi l arl y, the head of a
firmmay fi nd i t total l y i rapossi bl e to expl ai n hi s acti ons to the wor k
force, in t er ms whi ch they understand, wi thout vi rtual l y fal si fyi ng the
182 StufSord Beer
true basi s of his pol i cy. Thi s is not i nthe l east because ei ther the chi l d or
the l abourer i s stupi d; the poi nt hol ds even if they can each be proved
mor e i ntel l i gent than the father and the boss. They are j ust not part of the
metasystem, and they do not speak the metal anguage. And when one
comes to government , i t i s a necessary duty - on occasi on - to decl are
that the reasons why a mi ni ster took certai n acti on i s not inthe publ i c
i nterest to di vul ge. But the real poi nt behi nd many mi ni steri al fai l ures of
communi cat i on is not a matter of securi ty at all, but a matter of
metal i ngui sti cs.
These are a fewexampl es of the operati on of another f undamental
pri nci pl e of appl i ed cyberneti cs. It i s named the Pri nci pl e of External
Compl ementari ty (Beer, 1967). That is because, at some poi nt, all
hi erarchi c systems, however sophi sti cated, rely for thei r l ogi cal com-
pl eti on on a bl ack box l yi ng outsi de.
5. CONCEPTS OF BEHAVIOUR
The i denti fi cati on of mechani sms l ed to concepts of l aw; and we have j ust
seen how the l aws of cyberneti cs govern our vi ew of model l i ng. But
model s are usel ess unl ess they are appl i ed; we go i nto acti on i na mana-
geri al si tuati on ar med with a model whi ch - hopeful l y - embodi es the
l aws and appl i es the mechani sms.
I n doi ng thi s, we fi nd oursel ves conf ronted witha behavi oural si tuati on.
Thi s is a real worl d: a worl d inwhi ch model s have no status, cyberneti c
l aws are unrecogni zed, and the very mechani sms themsel ves are not
understood. For thi s is the regi on of manageri al acti on; i t is the worl d the
manager hi msel f understands, i t is the arena where he demonstrates his
own competence as a l eader and maker of deci si ons. And today, ina way
unexampl ed by any previ ous epoch of hi story, thi s worl d i s a scene of
change. Change has al ways happened; and i t is a clich to observe that
change i s mor e f undament al and happens much mor e qui ckl y than ever
before - thanks to an exponenti al advance i n technol ogy. We are ob-
servi ng, i ndeed, a change i nthe rate of change.
It i s for this very reason that management , whi ch has hi therto devel oped
i ts own compet ence to handl e i ncreasi ngl y el aborate si tuati ons fairly
successful l y, now appears to be fai l i ng. Throughout the worl d, and at
every l evel of operati ons, there i s a managemcnt cri si s of one ki nd or
Manugement in cybernetic terms 183
another. Now, al though we sai d that tensi on was necessary to uphol d the
tonal i ty of a system, we expl i ci tl y noted that the degree of tensi on wi thi n
a vi abl e systemmust remai n short of persi stent crisis. Today, on the
contrary, cri si s is persi stent - and al most uni versal . We face envi ron-
mental cri si s int er ms of pol l uti on; by pesti ci des, by carbon monoxi de, by
noi se, by chemi cal effuent, and so on. We face a technol ogi cal crisis al so,
one whi ch began wi th the threat of thermo- nucl ear war - a threat to
whi ch al most unaccountabl y we have become i nured. The fact is that the
ri sk of thermo- nucl ear devastati on is mor e seri ous today than i t ever has
been before, thanks to the extensi on of nucl ear capabi l i ty, and the ever-
i ncreasi ng el aborati on of the control systems requi red to restrai n
prol i ferati ng vari ety.
But beyond these exi sti ng probl ems there is the threat of the computer,
whi ch is hardl y i magi ned yet inthe mi nds of most authori ti es, nor even of
those who will sooner or later abuse i ts i mmense power. There is the
devel opment of the l aser, whi ch will produce a revol uti on qui te as
i mpressi ve as that al ready attai ned by computers themsel ves. And
thi rdl y, there i s the threat of a new soci al cri si s of di sorder and vi ol ence,
begotten l argel y by the other two, ai ded and abetted by the col l apse of
soci al metasystems whi ch proved so stabi l i zi ng to the soci al homeostasi s
of the past. Thi s i s a reference to the contemporary l oss of respect for
organi zed rel i gi on, for l aw, and for i nheri ted mores of every ki nd.
The probl emmay be summed up as a need for the management of
change, whi ch outcl asses any previ ous experi ence whi ch manki nd has
had of this requi rement. It was sai d much earl i er that most of our
management techni que was devoted to the smoot h runni ng of affai rs, and
that speci al measures had to be taken to ensure that the capabi l i ty to
i nnovate was not l ost. Wi thout thi s, we sai d, there coul d be no growth,
no l earni ng, no adapti on, no evol uti on. All this is suffi ci entl y difficult to
underwri te when the worl d is standi ng vi rtual l y sti l l - and i ndeed the
capaci ti es of manki nd to undertake any ki nd of effort (whether inthe
speed of travel , the capabi l i ty to comput e, the capaci ty to lift and
proj ect, or anythi ng el se) remai ned, wi th hi ndsi ght, vi rtual l y stati c for
fi ve thousand years. I t i s i nthe last hundred years that we have seen the
change in the rate of change whi ch has left our management capaci ty
gaspi ng for breath.
Then we must organi ze; we must desi gn control structures competent
to cope wi th an unprecedented task of adapti on. I n general , the met hod
184 StaSford Beer
whi ch peopl e adopt appears to be this. They l ook at exi sti ng trends and
extrapol ate t hem, hopi ng thereby to create for themsel ves a scenari o
whi ch expresses what the future will be l i ke. They then ask themsel ves
whether they can create pol i ci es and undertake deci si ons whi ch will
enabl e t he systemfor whi ch they are responsi bl e to survi ve inthese ci r-
cumstances. But this approach will not do.
Thanks to the exponenti al characteri sti cs of contemporary change,
these extrapol ati ons reach poi nts of si ngul ari ty well wi thi n the compass of
the peri od under revi ew. That is to say: the forecast expansi on becomes
i nfi ni te at some finitepoi nt int i me. That makes nonsense of the accepted
approach to adaptati on. The i ssue i s nowhere mor e poi gnant than in
consi derati on of the worl d popul ati on, whi ch - at the present rate of
expansi on - is predi cted (by these j ej une methods) to become i nfi ni te
wi thi n the l i feti me of peopl e al ready al i ve. Si nce thi s is i mpossi bl e, we must
l ook to the mechani sms whi ch will i nevi tabl y fl atten the expl odi ng curves.
We fi nd those that are natural to ecol ogy all too readi l y: they are the
threats to the exti ncti on of manki nd whi ch have al ready been l i sted. Not
onl y does i t take vari ety to absorb vari ety; vari ety will absorb vari ety.
Thus we face the ul ti mate task incontrol . Casti ng must be done; but i t is
not a questi on any l onger of contempl ati ng how to adapt to the predi cted
change. Qui te cl earl y, the predi cted change i tsel f must be del i beratel y
modi f i ed. As Gabor has put i t, we must i nvent the future. If we do not,
the future will happen to us. We shal l not l i ke it. Above all, inthe terms
of thi s paper, we have to l ay hol d on the expl odi ng outputs of our
si tuati on, and drag down these outputs to a l evel whi ch soci ety can con-
tai n. Then regul atory mechani sms whi ch can hol d those outputs steady at
the desi red l evel must needs be i nsti tuted.
The sci ence of cyberneti cs has, j ust in ti me, come to understand
somethi ng about the f undamental processes whi ch are causi ng l oss of
control , and al so about those processes whi ch are needed to contai n
expl osi ve di sorder. I f management i s to make use of these di scoveri es in
timei t will need to wor k at the metasystemi c l evel . We can no l onger
af f ord (the timeto tinker wi th the i nternal mechani sms of establ i shed
i nsti tuti ons.
These i ssues, whi ch are inmy opi ni on cri ti cal to the very survi val of
manki nd, wer e extensi vel y di scussed at an OECD conference hel d at
Bel l agi o, I tal y, in1968 (see J antsch, 1969).
Management in cybernetic terms 185
REFERENCES
ASHBY, W. R., (1954) Design for a Brain, New Yor k, John Wi l ey (revi sed edi ti on in
BEER, S., (1959) Cybernetics and Management, Engl i sh Uni v. Press (revi sed edi ti on in
BEER, S., (1966) Decision and Control, New Yor k, John Wi l ey.
CHURCI I MAN, C. W. , (1968) The Systems Approach, Del acorte Press.
FEI GENBAUM, E. and FELDMAN, J., (eds.) (1963) Computers and nought, New York,
FORRESTER, J. W. , (1961) Industrial Dynamics, MIT Press and John Wi l ey.
F O R R E S T E R , J. W., (1969) Urban Dynamics, MIT Press.
J ANTSCH, E. (ed.) (1969) Perspectives of Planning, OECD.
&I R, G. and VALACH, M., (1967) Cybernetic Modelling, IlifFe.
SHANNON, C. and WEAVER, W. , (1949) The Mathematical Theory of Communication,
WI ENER, N. , (1948) Cybernetics, New Yor k, J ohn Wi l ey.
1960).
1967).
McGr aw- Hi l l .
Ur bana (Ill.), Uni v. of Il l i noi s Press.
10 Concerning the notions of operation
and optimation
ARNOLD KAUFMANN
As l ong as man had rel ati vel y l i ttl e control over his envi ronment, he
regarded hi s ways of changi ng i t ei ther ingeneral t er ms or inrel ati on to
certai n mi cr o- phenomena that were consi dered as havi ng l i ttl e connecti on
between t hem. Now however, he i s transf ormi ng nature by his i ndustri al
acti vi ti es: his envi ronment is a technosphere in whi ch he exerci ses
i ncreasi ngl y effecti ve control . These i ndustri al acti vi ti es i ncreasi ngl y
requi re energy and i nf ormati on - the most expensi ve component s, l i mi ted
i ntimeand inspace, component s whi ch must be used to best advantage
and cl earl y demarcated. Thi s is why operati ons anal ysi s and opti mati on
have acqui red such i mportance inmanagement thi nki ng at all l evel s in
moder n soci ety. The manager needs hi s operati ons map and processes for
choosi ng between the sol uti ons i t i ndi cates. Hence the growi ng use of what
Kotarbi nski calls praxeol ogy (deci si on theory) and the operati ons maps
that mi ght consequentl y be cal l ed praxeograms or praxeographs , wi th
the associ ated processes for prepari ng and sel ecti ng deci si ons that con-
sti tute the essence of operati onal research.
The demarcati on of operati ons i s the fi rst concern; inthe succeedi ng
stage, the rel ati ons between themare i nvesti gated. The most sui tabl e
mathemati cal l anguage for these purposes at present i s the Kni g and
Berge graphs theory, as the operati ons are consi dered as member s of
countabl e sets and the most frequentl y encountered rel ati ons (numeri cal
or otherwi se) permi t confi gurati ons wi th whi ch mathemati ci ans are now
mor e at ease.
The anal ysi s of operati ons shoul d be non-prescri pti ve, its obj ect bei ng
to descri be, as correctl y as possi bl e, that whi ch exi sts or is postul ated.
The difficulty of so descri bi ng the structures of producti on or di stri buti on
can be seen f r oma gl ance at the combi nati ons i nvol ved. They can run to
thousands of separate or conti nuous vari abl es; to be usabl e, graphs to
represent these probl ems have to be f ragmented and the parts taken
188 Arnold Kaufmann
separatel y. Thi s compl exi ty i nfl uences not onl y the computi ng techni ques,
but forecasti ng, management and economi c acti vi ti es general l y. The
l arge but l i mi ted possi bi l i ti es of the computers have made i t possi bl e to
overcome of the maj or di ffi cul ti es, but for newki nds whi ch have ari sen i t
i s sti l l i mpotent.
Bef ore parti cul ari zi ng, i t may be useful to gi ve a fewexampl es.
Bui l di ng a hospi tal necessi tates thousands of di verse and not very
repeti ti ve operati ons. So does the runni ng of a commerci al port. Ur ban
traffic sets engi neers probl ems of di sturbi ng compl exi ty. Admi ni strati on
ini ndustry and the publ i c servi ces al i ke is begi nni ng to dri ve speci al i sts
to despai r, despi te all thei r data processi ng resources. Research may
i nvol ve endl ess parameters. Educati on inand out of school i s an i mmen-
sel y compl i cated combi nati on of probl ems.
How can operati ons be anal ysed, parti cul ari zed, have the rel ati ons
between themdetermi ned? Her e enters the model , representi ng somethi ng
real or hypotheti cal . As all the parameters cannot be i ncorporated, a
model is necessari l y si mpl i f i ed and arbi trary, but i t does gi ve the mi nd
somethi ng i t can grasp.
These model s can be very di fferent:
1. Deterministic: For each set of causes there is one set and one set
onl y of resul ts (these model s al so cal l ed programmes , but this wor d is
used for many, wi del y-di fferi ng concepts).
2. Stochastic (Probabi l i sti c): For each set of causes there is mor e
than one set of resul ts, but the coeffi ci ent of probabi l i ty of all the
possi bl e states of the systemcan be eval uated. Most often sel ected for
descri bi ng phenomena whi ch alter wi th ti me.
3. Uncertain but structured: By hypothesi s, all the possi bl e states are
known but at l east one is not and i ts coeffi ci ent of probabi l i ty cannot be
eval uated. Appl i es to cases where nature or other determi nants
i ntervene.
4. Uncertain but imperfectly structured: Not only are the probabi l i ti es
of the states i nadequatel y known, but al so the actual states themsel ves.
In thi s l ast case, constructi ng the model of a systemmeans worki ng
back by modi fi cati ons and si mpl i fi cati ons to case (3). Her e, however, the
i gnorance of probabi l i ti es for each of the possi bl e states inthe model is so
great a di ffi cul ty that the onl y thi ng to do is attempt to measure the
chances as accuratel y as possi bl e. So, further changes and si mpl i fi cati ons
are i ntroduced. However , deci si ons can rarel y be taken on the basi s of
Concerning the notions of operation und optimation 189
stati sti cal i nf ormati on or probabi l i ty l aws; resort is had to mor e preci se
measures (averages, vari ances, medi ans, quarti l es and other coeffi ci ents)
to bri ng the hypotheses down to earth.
Determi ni sti c model s are much used inprepari ng deci si ons, e.g.:
- Short- termforecasti ng (Graphs): Each operati on or acti vi ty is
si tuated intimein rel ati on to others. Exampl es: PERT ( Pr ogr amme
Eval uati on and Revi ewTechni que) graphs, Roy s potenti al graphs.
- Sequenci ng and combi ni ng. Preparati on and ti mi ng of tasks.
Of ten hard to transl ate i nto mathemati cal l anguage and usual l y hi ghl y
compl i cated, hence difficult to mani pul ate.
- Locati on and al l ocati on. Exampl e: vari ous acti vi ti es have to be
assi gned inrel ati on to a gi ven set of posi ti ons, tasks, executants, means
of producti on, l ocal i ti es.
- Economi c transfers (cf. Leonti ev matri ces).
- Economi c pr ogr ammes i nvol vi ng l i near or non-l i near constrai nts,
- Modul ar representati ons for anal ysi ng f l ows of all ki nds ( model
- Transport, ci rcui t and di stri buti on model s.
i nteger or di screte vari abl es.
consti tuted by the resul ti ng graphs and the i nter-fl owrelations).
Stochasti c (probabi l i sti c) model s are al so much used. The best known
and si mpl est is the Mar kov chai n, a sequence of r andomevents inwhi ch
the probabi l i ty of each event depends on the out come of i mmedi atel y
previ ous trials. For operati onal anal ysi s, their use requi res a del i cate
mani pul ati on of probabi l i ty theory; if the model s are quanti fi abl e, graph
theory can hel p to si mpl i fy and cl ari fy. Some maj or uses: i nventory,
queui ng, rel i abi l i ty, and equi pment repl acement probl ems.
In practi ce, stochasti c model s are sel domwhol l y or even adequatel y
expl oi tabl e. The statistics are general l y i l l - known and the probabi l i ti es
deri ved f r omt hemdubi ous. The typi cal model is mor e l i kel y to be
uncertai n but structured (3) than stochasti c (2). It mi ght be thought that
knowi ng the events possi bl e but havi ng little i dea of thei r respecti ve
probabi l i ti es woul d rul e out usabl e concl usi ons. However , these model s
are of uni versal l y acknowl edged val ue for risk assessment. As the prob-
abi l i ti es are poorl y known, an attempt i s made to eval uate the conse-
quences of certai n si tuati ons. But there are so many possi bl e combi nati ons
that onl y a l i mi ted number can be studi ed, despi te the ever- growi ng
possi bi l i ti es of si mul ati on methods.
Descri pti on, i.e. the non- normati ve anal ysi s of the operati ons, is of
190 Arizolcl Kaufnzaiin
course necessary, but is not concerned withthe human mi nd s demand
for effi ci ency or, mor e preci sel y, for the best. What do we mean by the
best?
To a mathemati ci an, the noti on i s cl ear. Because of certai n con-
strai nts, there is a subset of possi bl e sol uti ons; of these, one (or more)
whi ch will gi ve a maxi mumor mi ni mumresul t in rel ati on to a chosen
numeri cal cri teri on, e.g. a producti on pr ogr amme whose executi on takes
l east t i me, a wor ks organi zati on whi ch yi el ds maxi mumprofi t. Mat hemat -
i cal programmi ng rests on the assumpti on that there i s a si ngl e cri teri on
for acti on; inreal life, thi s i s rarel y the case (cf. Pareto).
Thc real ai ms of a busi ness executi ve, pol i ti ci an or other deci si on-
maker may be extremel y compl ex, and depend on a mul ti tude of cri teri a,
as often as not contradi ctory. Maki ng a deci si on or deci si ons i s a very
del i cate matter, wi th no uni que, excl usi ve rul es. Consci ousl y, and even
unconsci ousl y, f ree choi ce enters i n, but is rarel y deci si ve inmaki ng the
fi rst order of acti vi ti es. A comparati ve revi ewof the possi bi l i ti es will
someti mes yi el d this fi rst order, but this is sti l l not enough to i ndi cate a
maxi mum(or mi ni mum) i n the set theory sense. Other el i mi nati ons or
addi ti ons must f ol l owand even then, the resul t will be i nadequate i n
certai n cases. I deal l y, the i t ems to be compared will fall i nto a total ,
strict order, but thi s is very rare.
Hence the maj or i mportance of cri teri a sel ecti on i n all opti mi zati on
probl ems. Operati onal research ori gi nal l y tended tobe somewhat acade-
mi c, i t bei ng taken for granted that i t had a si ngl e, uni que obj ecti ve. The
trend i s now away f r omthis vi ew(cf. Roy, 1969).
There is a growi ng tendency nowadays to break down the ul ti mate ai m
i nto i ts maj or component s, convert these inturn i nto obj ecti ves, and then
seek the best means of achi evi ng them. Wor k done for the Uni ted States
Depart ment of Def ense f rom1961 onwar ds under Secretary of State Mc-
Namar a cl earl y affected research inbusi ness corporati ons l and publ i c servi -
ces. P P B s (Pl anni ng, Programmi ng, Budgeti ng System) anal ysi s i nvol ves
consi derabl e changes in the way of payi ng regard at hi ghest l evel to
i ntenti ons and possi bi l i ti es. The termi nol ogy is i ntended to hel p i n
determi ni ng the ul ti mate ai mas a functi on of i ntermedi ate tasks,
mi ssi ons and obj ecti ves, and vi ce-versa. Di scussi ons sti l l conti nue
regardi ng termi nol ogy and programmi ng methods. It will be i nteresti ng
to see the cyberneti c f ol l ow- up, the i ndi cati ons provi ded by el ectroni c
si mul ati ons, and effecti ve resul ts.
Concerning the notions of operation and optiniation
191
I n this new approach inthe choi ce of the sol uti ons there arc refl ecti ons
of the contri buti ons of von Neumann and Wal d to val ue theory vi z.,
seeki ng todetermi ne the preference of one sol uti on toothers by wei ghti ng,
and maki ng a sequenti al readj ustment accordi ngl y. However , numeri cal
adj ustment met hods have seri ous drawbacks (non-addi ti ve, non- homoge-
neous) and research tends, i nstead, to l ook to the pattern of possi bl e
preferences (where, i n any case, numeri cal eval uati ons can be used). It
is difficult to convert a combi nati on of i ntenti ons i nto cri teri a to be
used in maki ng a sel ecti on whi ch must i tsel f be consi stent wi th the
ori gi nal combi nati on of i ntenti ons. As the obj ecti ves are not homoge-
neous, the theory of probabi l i ti es affords l i ttl ehel p indeci di ng subj ecti ve
wei ghti ngs, whi l e measurement theory is not (as was ori gi nal l y hoped)
sui tabl e. However , di stri buti ve lattices seemto offer i ntercsti ng possi bi l -
ities.
When one cri teri on stands out but the others neverthel ess cannot be
negl ected, opti mi zati on by changi ng parani eters or by I c-opti mal i ty, can
be attempted. Al teri ng one or more of the parameters may gi ve opti mal
sol uti on(s) much better adj usted if the combi nati on of preferences i s
compl ex. k-opti mal i ty can be taken i nseveral ways. I f the sets are fi ni te,
cl ass 1- opti mal is the subset of opti mal sol uti ons, cl ass 2- opti mal the
subset of opti mal sol uti ons of the whol e less the cl ass l - opti mal sol uti ons,
and so on. A certai n arbi trary order of sol uti ons that mi ght be of i nterest
for other cri teri a can be f ound by taki ng in cl asses made up of sol uti ons
whi ch vary f romthe opti mal sol uti on or sol uti ons by a di fference equal
to k (k-opti mal i ty).
What do we mean by opti mal i ty ina system? I n nature - inbi ol ogy for
i nstance - i t is not easy to see what the noti on of opti mal i ty can corres-
pond to. On the other hand, the i dea of equi l i bri umvery frequentl y recurs.
All the same, i t sti l l remai ns necessary to gi ve a better defi ni ti on of
exactl y what is understood by equi l i bri um. I n economi cs and soci ol ogy
the noti on of opti mal i ty i s probabl y artificial and mi ght not be abl e to
stand up tocertai n contradi cti ons, but the habi t is now i ngrai ned and the
term' opti mal ' is i ncreasi ngl y used. It will not di sappear unti l we fi nd a
better. Let us then recal l what opti mi zati on consi sts in or, to use the
termi nol ogy accepted by a number of authors inFrance, l et us study the
opti mi zati on of systems.
Take a systemcompri si ng men or/ and machi nes or/ and products. The
constrai nts inthe model of this systemi nvol ve t wo sorts of vari abl es:
192 Arnold Kauf mann
1) those the deci si on- maker can control , and 2) those he cannot. The
model is assi gned a si ngl e-val ued functi on. The ai mis then to fi nd out
whi ch val ues of the control l abl e vari abl es will gi ve an opti mal val ue
for the functi on assi gned to the model ( maxi mal or mi ni mal as appro-
priate). The val ues assi gned to the non-control l ed vari abl es f ol l owa
seri es of hypotheses.
A fewexampl es will cl ari fy. A factory has m machi nes wi th whi ch i t
proposes to make n products. An operati on Oz3 consi sts i n getti ng
product i off machi nej . There are vari ous constrai nts: order in whi ch
certai n operati ons must take pl ace, operati ng t i mes havi ng tobe fi ttedi n
at speci fi ed i nterval s, and so on. The val ue functi on to be opti mi zed
(mi ni mi zed) is total machi ne i dl e time.
Such probl ems are f requent in i ndustry. They are usual l y hi ghl y
compl ex, and when the quanti ti es mand n are greater than ten, the time
needed to process t hem, even with hi gh-capaci ty computers, may be
much too l ong. The deci si on- maker accordi ngl y settl es for a reasonabl e
compromi se. Met hods whi ch provi de a passabl e (even t hough non-
opti mal ) sol uti on are often termed heuri sti c (f romthe Greek verb to
fi nd out); they take much l ess ti me.
Depot s i n central poi nts suppl y di stri buti on centres wi th vari abl e
quanti ti es of a product. Lorri es of known capaci ty are avai l abl e. The
di stances between the depots and the centres and between the centres are
known. The probl emi s to reduce to a mi ni mumthe total di stance the
l orri es must cover i na gi ven peri od, e.g. a week. This ki nd of probl em
occurs inthe di stri buti on of oil and petrol , mi l k col l ecti ng, operati on of
school buses, and so on. Her e agai n, heuri sti c met hods can often gi ve
adequate sol uti ons.
Locati ng a centre or centres of producti on. One or mor e products will
have to be del i vered to other l ocal i ti es. Transport and l abour costs inthe
possi bl e si te areas are known. Other factors i ncl ude accessi bi l i ty,
avai l abi l i ty of housi ng, l ocal charges, cl i mate. Sel dominsuch cases is
there one cri teri on onl y, and several val ue functi ons coul d be proposed
for opti mi zati on. Usual l y, one is sel ected and heuri sti c met hods are used
to fi nd opt i mumval ues for the others.
The al gori thmi c, and some of the heuri sti c met hods used for the above
three types of probl emare based on a very general procedure. The set of
sol uti ons i s broken down by an n-parti ti on (very often y1 = 2) and each of
the subsets thus created is gi ven an upper (l ower) bound. The process is
Concernihg the notions of operution and optinution 193
repeated for a subset havi ng the greatest l ower bound, and each new
subset i s agai n bounded. It is then conti nued wi th a subset havi ng the
l east upper bound, sel ected f r omthe subsets whi ch have not yet been
subj ected to breakdown, and so on unti l a subset contai ni ng onl y one
sol uti on, a mi ni mal sol uti on, is obtai ned. Thi s el egant eval uati on
process is known as the branch and bound met hod inthe Uni ted State
and as the mthode de sparation et vahation progressive inFrance. It is
of very general appl i cati on, provi ded that the boundi ng and branchi ng
can be properl y done, whi ch is not al ways easy. However , the rooted-
tree curve f ol l owed acqui res mor e and mor e peaks as the eval uati on
conti nues, and i t may take much too much ti me, even wi th hi gh capaci ty
computers. I n such cases, heuri sti c met hods can be used at a certai n poi nt.
Branch and bound opti mi zati on can be used to fi nd ei ther a maxi mumor
a mi ni mumafter adj ustment of the al gori thm, and i t works inprobl ems
i nvol vi ng conti nuous or mi xed vari abl es.
It can al so be used inconj uncti on withPol ya s screeni ng met hods. The
human mi nd commonl y tri es to fi ndthe opti mal sol uti on by stages - by
successi ve screeni ngs or si fti ngs.
Li near programmes are among the compl ex- systemmodel s in wi dest
use today. All the constrai nts of the model are l i near equati ons or
i nequal i ti es and the sol uti ons are si tuated in a convex domai n. These
equati ons or i nequal i ti es will take the f ormof a set of rel ati ons such as:
j = N
2
j= 1
aij xj bt O i = 1, 2, ..., M
The symbol 0 i ndi cates that the equal i ty or i nequal i ty may be in ei ther
di recti on. The vari abl es xj are non-negati ve, whi l st the coeffi ci ent a$$
and bi may be any real numbers. A val ue f uncti on i s taken whi ch is al so
a l i near f uncti on:
j=N
Z
j= 1
f(x1, x2, ..., x,) =
cj xj where the q are real numbers
The obj ect then i s to fi nd a sol uti on or sol uti ons gi vi ng an opti mstl
( maxi mal or mi ni mal ) val ue ofx
Much used withthis type of model is the Si mpl ex al gori thm, devel oped
some twenty years ago by G. Dantzi g inthe Uni ted States and Kant orov-
itchin the us s R (al though the actual i dea was moot ed much earlier).
194 Arnold Kaufmann
Thc fol l owi ng i s an exampl e of a l i near pr ogr amme model of a
producti on system. A firmhas toturn out 14di fferent types of products.
There are three assembl y l i nes on whi ch each of the products can be
processed. The order i nwhi ch the products come off the l i ne i s uni mpor-
tant. The capaci ty of each l i ne, the producti on cost per ton of each
product on each l i ne and the hourl y output of each l i ne for each product
are known. The demand for each product is also known. The model
accordi ngl y takes the fol l owi ng f orm:
j =14
< c.I
z - i = 1, 2, 3.
j = 1, 2, . . . , 14.
xz5
j =1 4
xu + Xzj + x3j = dj
where xtj i s the quanti ty of the product j to be produced on the line i ;
&j is the output of product j on linei ;
ci is the capaci ty of linei ;
dj i s the demand for product j .
ti on) can be wri tten:
The val ue functi on whose mi ni mumis sought (the total cost of produc-
Comput ers can now process pr ogr ammes contai ni ng thousands of
vari abl es and thousands of constrai nts, as maj or economi c programmes
requi re. Vari ants of the Si mpl ex met hod, devi sed to make the processi ng
more rapi d, are often used inconj uncti on withapproach processes whi ch
are ni ore rapi d incertai n cases (gradi ent met hod, Lagrange s met hod of
mul ti pl i ers). A consi derabl e l i terature exi sts on thi s subj ect.
Constrai nts are agai n presented inthe f ormof equati ons or i nequal i ti es
i n model s known as i nteger-val ued l i near pr ogr ammes or mi xed-
vari abl e l i near programmes . I n these model s, the control l abl e vari abl es
can take onl y non- negati ve i nteger val ues or el se Bool ean val ues (Oor 1)
or, agai n, some vari abl es can take onl y i nteger val ues whi l e the others can
have conti nuous val ues. These probl ems are much mor e difficult tosol ve,
and a great deal of wor k has been done on t hemi nrecent years. Many
f orms of al gori thms have been suggested, nearl y all based on branch and
bound processes. Economi c and other probl ems of this type, wi th
Concerning the notions of operution und optimution 195
hundreds of vari abl es, can now be processed, but requi re much mor e com-
pt er timethan i n cases where all the vari abl es have conti nuous val ues.
In some model s, the val ue functi on orl and the constrai nts are not
l i near, but may, e.g. be quadrati c or mi xed. Met hods have been suggested
by Wol f e, Beal e and Rosen, to menti on onl y recent wor k.
Before consi deri ng model s whose subj ect matter i s randomi zabl e, we
must look at dynami c programmi ng, i.e. model s i nwhi ch the changes in
the states of the systemoccur sequenti al l y, At moment s di stri buted
di sconti nuousl y or conti nuousl y in time, the systemchanges. At these
moment s the vector of states may change. For our present purpose, we
shal l assume that this change takes pl ace ina predetermi ned manner (the
case of the stochasti c processes will be consi dered late). I n these
determi ni sti c processes, an opti mal traj ectory will be sought by sequenti al
processes. The pri nci pl e for the resol uti on of sequenti al pr ogr ammes of
thi s nature has been stated by R. Bel l man i nthe Uni ted States, and may
be resumed as: An opti mal traj ectory can onl y be f ormed by opti mal
subtraj ectori es. To appl y thi s pri nci pl e, i t i s necessary to prove certai n
theorems deri vi ng f r omi t whi ch, as a rul e, are very easy to prove and i n
certai n cases are si mpl y trui sms. If an opti mal traj ectory coul d be f ormed
of non- opti mal subtraj ectori es, then the opti mal traj ectory woul d have to
pass through other subtraj ectori es and coul d no l onger be opti mal ,
whi ch contradi cts the hypothesi s. But the determi ni sti c sequenti al
processes treated i n dynami c pr ogr ammi ng must be careful l y exami ned
before bei ng subj ected to opti mati on processi ng; i nparti cul ar, they must
be associ ati ve.
Certai n combi nati on probl ems can be processed by dynami c program-
mi ng, needi ng onl y to have a sui tabl e structure that i s ei ther sequenti al o
can be rendered such. But i t no l onger wor ks - even wi th computers -
when the vari abl es of state are numerous. I ts possi bi l i ti es are thus
l i mi ted, but val uabl e inundertaki ng the constructi on of sequenti al model s.
One i nteresti ng poi nt is the boundary wi thi n whi ch dynami c program-
mi ng operates. For each probl em, the model must have a sui tabl e
boundary, upon whi ch the resul ts usual l y very l argel y depend. Accordi ng
as the val ue functi on used i s short-termor l ong, the opti mal traj ectori es
will be wi del y di fferent.
I f di screte vari abl es are i nvol ved, dynami c pr ogr ammi ng may often
need k-opti mal i ty processes (see above), especi al l y if there can be mor e
than one si ngl e sel ecti on cri t- , non.
196 Amold Kaufmann
Sequenti al model s are used i n i nventory, sequenci ng, i nvestment and
equi pment control probl ems; hence the practi cal val ue of dynami c
programmi ng. A theory of opti mal processes will be di scussed bel ow.
Determi ni sti c sequenti al processes i mmedi atel y evoke stochasti c
processes. I f randomi zati on i s feasi bl e i nconstructi ng a model , the model
i tsel f will mor e often than not be stochasti c. Mar kov chai ns, whi ch pl ay
so i mportant a part i n heredi ty phenomena inbi ol ogy, are si mi l arl y
i mportant i nthe human sci ences.
Assi gni ng a val ue f uncti on to states in Mar kov chai ns can provi de
good and representati ve model s of vari ous i nventory and equi pment
repl acement probl ems. R. Howar d fi rst dr ewattenti on to the advantage
of consi deri ng the normati ve aspect of Mar kov processes, and hi ghl y
i nteresti ng ergodi c properti es have been brought to l i ght insubsequent
studi es; for these di screte processes, the use of demonstrati ons empl oyi ng
graphs represents another appreci abl e advance. Al though stati onary
processes are rare inreal phenomena, the hypothesi s of thei r exi stence i n
the sel ected model makes i t possi bl e to bri ng out cl earl y the essenti al
condi ti ons for equi l i bri um. I n the case of processes where the vari abl es
take their val ues i nthe cont i nuum, the ti me el ement bei ng di screti zed
or not, rather tri cky probl ems of i ntegrati on and di fferenti ati on are posed
and thei r exhi bi ti on is di ffi cul t or sti l l remai ns to be worked out.
Deci si on- hazard processes represent an i nteresti ng extensi on of
Mar kov chai ns. A sequenti al deci si on-hazard process is a process in
whi ch nature and the deci si on- maker i ntervene al ternatel y. It is assumed
that suffi ci ent stati sti cal observati ons of nature have been made to
assi gn probabi l i ti es to all the rel evant changes of state; that each change
of state has an associ ated val ue; and that the changes of state consequent
upon deci si ons can be eval uated. Thus, to any strategy, a total val ue can
be assi gned whi ch i s the sum of the val ues of all the changes of state and
those that deri ve f r omthe deci si ons. I n thi s ki nd of model , the process
need no l onger be stati onary, and i t can approxi mate fairly cl osel y to a
di versi ty of real -l i fe si tuati ons. I n these processes, the i nterval of anti ci pa-
ti on or economi c hori zon i nvol ved pl ays a very i mportant part. However ,
in cases whi ch are not stati onary, or stati onary for a smal l number of
deci si on-hazard phases onl y, ergodi c equi l i bri umi s rul ed out.
These noti ons have a theoreti cal i mportance for l earni ng and adapti ve
processes whi ch goes well beyond that of thei r practi cal appl i cati ons.
The l i near rel ati ons whi ch enter i nto Mar kov processes permi t the use
Concerning the notions of operation und optimation 197
of common functi onal transformati ons: the Lapl ace transf ormfor con-
ti nuous processes, or the z transformati on for processes di sconti nuous
inti me. These transformati ons hel p to demonstrate convergences whi ch
are parti cul arl y useful in studyi ng and compari ng vari ous l ong- term
pol i ci es.
Stochasti c processes al so have a very wi de appl i cati on inprobl ems of
economi cs, soci ol ogy and even psychol ogy (in queui ng, i nventory,
i nvestment, and other phenomena previ ousl y menti oned). Queui ng occurs
in many aspects of soci al and economi c life. Servi ce poi nts provi de
servi ces for cusers (machi nes and products as wel l as peopl e). Servi ce
poi nts and users together f or ma queui ng system . It is assumed that
probabi l i ty di stri buti ons can be assi gned to the states of this system, and
that i ntegral or di fferenti al equati ons will yi el d worki ng esti mates.
Pol i ci es can then be worked out to avoi d bottl enecks. Di fferent f or ms of
i mpati ence can be studi ed. Pri ori ti es are establ i shed to faci l i tate the
traffi c and sati sfy l egi ti mate demands. The probl emis often to mi ni mi ze
the total cost of queui ng for servi ce poi nts and users al i ke; or to reduce
ri sks. The soci ol ogi st can study and quanti fy i ndi vi dual and gr oup
behavi our. Queui ng theory can be appl i ed to road, air and sea trafi c, to
producti on - if executi on t i mes are vari abl e and can be measured
statistically - and to many physi cal phenomena. Phenomena whi ch are
too compl i cated for compl ete anal ysi s by such model s can be studi ed by
si mul ati on. Si nce Erl ang s initial wor k, thousands of publ i cati ons have
di scussed cases of the most di versi fi ed ki nds.
I nventory probl ems are cruci al i n many systems. Stati onary r andom
processes can be used i ncertai n rather rare cases, for whi ch equi l i bri um
condi ti ons and permanent pol i ci es can then be wor ked out. Vari ous ki nds
of costs are i nvol ved: management and hol di ng costs, costs deri vi ng f r om
shortages or extent of shortage, cost of re-stocki ng. I f equi l i bri umis
possi bl e, i t is worth whi l e determi ni ng the opti mal pol i cy or pol i ci es.
Thi s may be rel ati vel y easy, but if the i nventory i nvol ves several stages
of di stri buti on and management and transport di ffi cul ti es, no acceptabl e
anal yti cal study may be feasi bl e and si mul ati on will have to be used.
General l y speaki ng, i nventory probl ems change too rapi dl y to have any
stati onary aspects. Study has to depend on short- termchanges and f ore-
casts whi ch often wor k out qui te well, i nfactori es, bi g stores, warehouses,
and so on. These forecasts rel y on si mul ati on processes that are al so
l earni ng and adapti ve processes.
198 Arnold Kuufmunn
The i ncreasi ng extent and compl exi ty of moder n technol ogy demand an
i ncreasi ngl y cl ose check on i ts consti tuent parts (cf. inthe producti on of
energy, transport, transfer of processi ng of i nf ormati on) and, for
mai ntenance purposes, i ncreasi ngl y refi ned studi es of rel i abi l i ty. Econom-
ic opti mi zati on has to pay attenti on to mai ntenance and repai r costs,
masses (e.g. for space rockets), vol umes, and so on. Several criteria may
have to be sati sfi ed, and parametri c transf ormati on (see above) i s often
necessary.
Mar kov chai ns are often used inrel i abi l i ty and equi pment repl acement
studi es. The equati ons are the same for the life and death of equi pment
as for the life of cells i n bi ol ogy, except that, in nature, man can onl y
very rarel y i mpose his control systems, whereas he can i ncorporate t hem
inhis technol ogi es. But the anal ogi es between technol ogi cal and bi ol og-
i cal life shoul d not be stretched too far. However , bi ol ogi sts can get
i deas f r ommodel s constructed for economi c and soci al studi es, and
economi sts and soci ol ogi sts have much to l earn f r omexami ni ng the
met hods of bi ol ogi sts. A machi ne is a general i zed aut omat on that
resembl es any l i vi ng bei ng, except that i ts use can be control l ed inendl ess
ways.
Normal l y, this article shoul d consi der the model s commonl y used
to study systems that are uncertai n but structured (see 3, page ISS), i.e.
where all possi bl e states are known but the coeffi ci ent of probabi l i ty of
at l east one cannot be eval uated. The cases are in fact covered under
games strategy and deal t wi th by another contri butor to this book.
However , toti e inwhat has j ust been sai d about opti mi zati on mor e cl osel y
wi th what is to f ol l ow, games strategy can be consi dered f roma speci al
angl e, i.e. games agai nst one or mor e opponents or/ and agai nst nature.
Her e we take games agai nst nature onl y, i n whi ch the possi bl e states
must be defi ned and a val ue assi gned to each. Dependi ng on the purpose
of the author (e.g. von Neumann, Lapl ace, Hurwi tz, Savage), the op-
ti mi zati on may be ei ther mi ni max or maxi mi n, a si mpl e wei ghti ng,
mi ni mal regret, and so on. Ther e coul d be endl ess argument about the
rel ati ve meri ts of vari ous criteria - thi s i s obvi ousl y a subj ecti ve matter.
But a critical exami nati on of these cri teri a gi ves an excel l ent i nsi ght i nto
the contemporary psychol ogy of deci si on- makers; in parti cul ar, the
procedures suggested by certai n domi nati ng i deas make i t possi bl e (q.
page 191 above) to bri ng a combi nati on of preferences or i ntenti ons i nto
l i ne wi th cri teri a for acti on. Another advantage of these rather academi c
Concerning the notions of operation and optimation 199
studi es i s that they gi ve a better i dea of what is meant by a strategy agai nst
nature and, mor e especi al l y, what is meant by a mi xed or wei ghted
strategy. For the sake of accuracy, a di sti ncti on must be made between
pol i cy and strategy . A pol i cy appl i es ina case whi ch i s determi ni sti c
(see 1. on page 188) and i nvol ves scal ar or vectori al deci si ons; a strategy
appl i es in cases whi ch are not determi ni sti c but ei ther probabi l i sti c, or
uncertai n but structured (see 2. and 3. on page 188).
I n games agai nst nature, sequenti al games are extremel y i mportant.
The sequenti al f or mof the deci si on-hazard process (see3. and 4. on page
188) recurs, but thi s ti me, nature i ntervenes, and the coeffi ci ent of prob-
abi l i ty cannot be eval uated. As the combi nati on el ement enters fairly
extensi vel y i nto these games, i t is difficult to appl y t hemin practi cal
cases. The comput er doesn t seemtobe abl e to do much to hel p (machi nes
are needed wi thci rcui ts inparal l el to hi gh degree; sequenti al machi nes
are too sl ow, even if they can work at the rate of a nano- second for each
bit). The i nterest of these games i s that they al l owa sci enti fi c approach to
deci si on preparati on.
Mor e el aborate control processes are descri bed as l earni ng and
adaptati ve . I n the real , technol ogi cal worl d inwhi ch men take deci si ons,
such deci si ons are sel domfi nal , or even regul ar and systemati c. When too
l arge a change occurs inthe obj ect, the systemor the envi ronment, the
deci si on- maker i ntervenes and adapts the systemtobri ng i t i nto linewi th
hi s cri teri a. Starti ng f r oma rel ati vel y modest knowl edge of the systemand
the envi ronment, he acqui res - over a peri od of timewhi ch vari es in
l ength - a better knowl edge of both and of all the condi ti ons i nvol ved. He
can thus wor k out a control process that takes account of the wi dest
range of observed human habi ts and behavi our.
Thi s pi l oti ng makes opti mi zati on much mor e real i sti c. The deci si on-
maker goes about i t as fol l ows. He l ocates a model of the systemhe
wi shes to control in a future whi ch may or may not be fi ni te, but in
pri nci pl e al ways will be. To deal wi th the combi nati on of hypotheses he
has adopted, he seeks the strategy that gi ves the best resul ts inrel ati on to
his cri teri on, or the best compromi se if his cri teri a are pl ural . He feeds
inthe deci si ons for the present and the future that the strategy di ctates.
The strategy proves correct up to a poi nt onl y because nature al so
i ntervenes. Hestudi es the newsi tuati on. He can conti nue wi th the same
model or choose another mor e realistic. He agai n works out an opti mal
or compromi se strategy, for e. g. a shorter or l onger future, or for di fferent
200 Arnold Kaufmann
val ues of the vari abl es over whi ch he has control . Another new si tuati on
ari ses . . . and so on.
Thi s is the approach most frequentl y used i ncyberneti cs. It is used in
i ts most advanced and real i sti c f ormineconomi cs, for exampl e, to wor k
out f undament al patterns, but can equal l y well be used in gui di ng a
spacecraft, steeri ng a shi p or other machi nery, ingovernment, inmanage-
ment , or i n studyi ng the way in whi ch an i ndi vi dual behaves in his
envi ronment under the i mpul se of his reason and his feel i ngs.
Pi l oti ng i s of parti cul ar practi cal i mportance i n macro- economi cs.
Uxt i l qui te recentl y, pl anners drewup devel opment pl ans for a speci fi c
peri od (e.g. fi ve years). The authori ti es then tri edby every possi bl e means
to achi eve the obj ecti ves, even if, meanwhi l e, condi ti ons and opi ni ons
changed. In practi ce, no medi umor l ong t er mpl an was carri ed t hrough to
achi eve the obj ecti ves as set. Pl anners now prefer a sl i di ng scal e of
obj ecti ves or even a sl i di ng and vari abl e scal e. Yearl y, or sooner if
necessary, the pl an is exami ned to see how i t i s worki ng out, and i t i s
sui tabl y modi fi ed. It i s surpri si ng that economi sts inadvanced countri es
are onl y now di scoveri ng thi s i dea whi ch common sense woul d seemto
suggest, probabl y because of i ngrai ned habi ts and all ki nds of brakes on
the necessary but not al ways foreseeabl e changes i nsoci ety.
In faci l i tati ngthis mor e fl exi bl e approach inpl anned or semi -l i beral
economi es, computers will have an i mportant part; for, after each
l earni ng and adaptati ve phase (see page 199), i t will be necessary to
si mul ate the new si tuati ons that correspond to new hypotheses. As these
si mul ati ons become mor e compl ex and del i cate, the mathemati cal
resources of computers become i ndi spensabl e to the anal yst. The ap-
proach of economi c pl anner and deci si on- maker here is al most exactl y that
of McNamar a and Kantorovi tch (see page 190 above); the mai n di fference
i s in the manner and order inwhi ch the PPBS eval uati ons and cl assi -
fi cati ons are made.
Ci vi l i ans now fairly often fol l owthe mi l i tary practi ce of appl yi ng the
cost-effecti veness cri teri on to deci si on probl ems. I n many economi c
probl ems, cost and effecti veness are the mai n cri teri a, and i t i s difficult to
say whi ch is the mor e i mportant. A graph pl otti ng vari ati ons in
effecti veness agai nst cost (or vi ce-versa) i ndi cates the best sol uti ons in
t er ms of i ndi vi dual preferences or those of supposedl y coherent groups.
The cyberneti c flow chart on page 201, showi ng the phases in thei r
correct posi ti ons, summari zes the i deas about economi c pi l oti ng set
Concerning the notions of operution and optmution 201
I
I
Evaluation
of conditions
of
environment
Analysis Study Choice
system motivations objectives
+ of the of --c of
-
t
Adaptation
ofthe
model
- - - -
Construction pLL:ead: of a
Measuring
of results
effectiveness
t
Selection of
means of
I
Black box
(nature) Implementation Decision
J
out above. The anal ysi s of the systemi s preceded by an eval uati on of the
envi ronment and fol l owed by a study of moti vati ons, after whi ch the
obj ecti ves are defi ned. A model is constructed or adapted. Once total cost
has been worked out and ef f ecti veness speci fi ed (both can have vectori al
forms), the possi bl e means of i mpl ementati on and the possi bl e resources
Resour ces
(budget)
202 Arnold Kuufmunn
are eval uated. The deci si ons are taken and thei r i mpl ementati on i s
supervi sed. Nat ure i s the ul ti mate deci di ng factor at thi s stage. The resul ts
are tested and a newoperati on begi ns (see pages 199-200). The l i nks will
be noted between the means of i mpl ementati on and the adaptati on of the
model , and between the eval uati on of resources and the choi ce of
obj ecti ves. (Thi s f l owchart i s of course gi ven onl y as an i l l ustrati on, and
woul d be adapted as parti cul ar probl ems may requi re.)
These cyberneti c f orms can be expressed in strictly mathemati cal
t erms - parti cul arl y necessary in technol ogi cal procedures, where the
deci si on has to be pr ogr ammed in advance and i ntervenes at a speed
consi derabl y hi gher than that of the reacti ons of the human brai n.
A theory of opti mal control i n sequenti al processes put f orward by
Pontryagi n and his Sovi et research group takes up and devel ops vari ati on
cal cul us pri nci pl es and i nvol ves pr ogr ammed control . The cases for
whi ch i t i s val i d may be determi ni sti c or probabi l i sti c (see page 188). I n
bri ef and obvi ousl y i nadequate outl i ne, i t can be summari zed as fol l ows.
The fol l owi ng di fferenti al equati ons appl y:
dxf
- = ft(x1, , . . , x, u1, ...) ur), i = 1, 2, . . . , n.
dt
where xz represents the state vari abl es, and d the control vari abl es.
What val ues can be assi gned to the tif vari abl es inan i nterval between t wo
i nstants to and tl, assumi ng that the u3 vari abl es are functi ons of ti me, and
that t he initial val ues at t = to of the xf vari abl es are known, vi z. xb?
Take a functi onal i ntegral
tl
j
t = to
J = fO (XI, ..., x, uI, ..., ur) dt
wheref o i s a gi ven functi on. For every control defi ned by the uf vari abl es
i n the i nterval to to tl, the i ntegral takes a finite val ue. I f a control
exi sts whi ch converts the systemf r omstate vector xi to state vector xi,
the i ntegral J becomes mi ni mal .
Pontryagi n s met hod to some extent combi nes Bel l man s opti mal i ty
pri nci pl e with the usual vari ati on cal cul us versi on, al l owi ng a strict
control of the processes i n many i mportant cases. Thi s was because
rel i ance coul d no l onger be pl aced on mor e or l ess i ntui ti ve mathemati cal
methods. Pontryagi n s met hod has proved al most i deal for opti mal i ty
Concerning the notions of operation and optimation 203
anal ysi s and control and, withthe many vari ants si nce wor ked out, has
been of vital i mportance inspace navi gati on, i n the real t i me control of
conti nuous manuf acturi ng processes, and in the study of many still
unsol ved theoreti cal opti mati on probl ems.
REFERENCES
AUMALE, G. D, (1968) Programmation des dcisions, Pari s, Presses Uni versi tai res de
BXKENBACH, E. I., (1964) Applied Combinatorial Mathematics, New York, John
BELLMAN, R., (1960) Dynamic Programming, Pri nceton Uni v. Press.
BERGE, C., (1962) L a Thorie des graphes et ses applications, Pari s, Dunod.
BOOT, J. C. G. , (1964) Quadratic Programming. Algorithms. Anomalies. Applications,
CAUDE, R. and MOLES, A. , (1964) Mthodologie. Vers une science de laction, Entrepri ses
DANTZI G, G. (1963) Linear Programming and Extensions, Pri nceton Univ. Press.
HAMMER, H. P. and RUDEANU, S. (1968) Boolean Methods in Operations Research,
HOWARD, R. (1960) Dynamic Programming and Markov Processes, New Yor k, J ohn
KAUFMANN, A. , (Vol . I, 1970; Vol . II, 1968) Mthodes et modles de la recherche opra-
KAUFMANN, A. (1966) LHomme daction et la science, Pari s, Hachette.
KAUFMANN, A. (1968) Introduction la combinatorique, Pari s, Dunod.
KNIG, D. (1936) Theorie der endlichen und unendlichen Graphen, Lei pzi g, Akad.
K~~NzI , H. P. and KRELLE, W., (1962) Nichtlineare Programmierung, Spri nger- Verl ag.
LUCE, R. D. and RAI FFA, H. (1957) Games and Decisions, New Yor k, John Wi l ey.
NEUMANN, J. VON and MORGENSTERN, O. , (1944) Theory of Games and Economic
PONTRYAGI N, L. S. et al. (1962) The Mathematical Theory of Optimal Processes, New
ROY, B., (1969) Algbre moderne et thorie des graphes, Pari s, Dunod.
THRALL, R. M. et al. (1957) Decision Processes, New Yor k, J ohn Wi l ey.
WALD, A. (1950) Statistical Decision Functions, New Yor k, J ohn Wi l ey.
France.
Wi l ey.
Amst er dam, Nor t h Hol l and Publ . Co.
moder n= d di ti on.
Spri nger- Verl ag .
Wi l ey.
tionnelle, 2nd edi ti on, 2 vols. Pari s, Dunod.
Verl ag.
Behavior, Pri nceton Univ. Press.
Yor k, John Wi l ey.
11 Game theory
NICOLA1 N. VOROBYEV
1. FUNDAMENTAL CONCEPTS OF THE THEORY OF GAMES
1. 1 Game theory is the theory of mathemati cal model s for deci si on-
maki ng inconfl i ct si tuati ons (e.g. incontests).
The i mpl i cati ons of thi s very wi de defi ni ti on need detai l ed anal ysi s.
The theory deal s wi th model s, i.e. not wi th actual opti mal deci si ons but
wi th i deal i zed schemata of t hem. All the general statements that may be
made about arbi trary model s appl y i n game theory. These model s are
not j ust any ki nd of model , but f ormal , symbol i c, mathemati cal model s.
Thi s means that all the game-theoreti cal statements must be suffi ci entl y
preci se tol end themsel ves to compl ete formal i zati on.
I n parti cul ar, the very obj ect of the game theory, the opti mal deci si on-
maki ng under confl i ct must be f ormal i zed, and thi s inturn requi res the
formal i zati on of the concepts of deci si on-maki ng , opti mal i ty and of the
confl i ct inquesti on.
The formal i zati ons for each of the three concepts are probl ems of
enti rel y di fferent compl exi ty.
1.2Deci si on- maki ng is understood i nthe si mpl est sense as the choi ce by
the deci si on- maker of an arbi trary deci si on f romwi thi n a set of possi bl e
deci si ons gi ven i n advance. Game theory is not necessari l y concerned
withthe physi cal or soci al nature of the deci si on, the deci si on-maker s
reasons for or the means of sel ecti ng hi s deci si on, much l ess his manner
of i mpl ementi ng i t - these are partl y matters for cyberneti cs and opera-
ti ons research. Neverthel ess, i n deal i ng wi th speci fi c cases, game theory
may al so have to bear these aspects inmi nd. An el ementary but very
i nstructi ve exposi ti on of game theory contai ns i nthe book of M. D. Davi s
(1970).
206 Nicoiai N. Vovobyev
1.3 The formal i zati on of the i dea of a confl i ct i s appreci abl y mor e
compl ex.
A confl i ct is a phenomenon inwhi ch we know who is i nvol ved and how;
what are the possi bl e out comes; and who is concerned by these out comes
and how. Thi s descri pti on can be enl arged upon.
First, we must l i st all theacti ngfactors. Thesecanbecol l ecti veandpossess
some i nternal structure. Theref ore they may be cal l ed acti ng coal i ti ons. The
total of all such coal i ti ons consti tutes a set, whi ch we can desi gnate Ra.
SecondZy, all permi ssi bl e deci si ons for each operati onal coal i ti on must
be i ndi cated. Usual l y, i n the theory of games no di sti ncti on is made
between the deci si on and i ts i mpl ementati on (a ki nd of i denti fi cati on
frequentl y used i n mathemati cs). The acti ons, behavi ours and deci si ons
of each acti ng coal i ti on are cal l ed i ts strategi es. The set of all strategi es of
acti ng coal i ti on K can be desi gned SK.
Thirdly, we have to i ndi cate all the possi bl e out comes, whi ch are
basi cal l y the resul ts of the choi ces of strategi es by all the coal i ti ons. These
out comes are cal l ed the si tuati ons. The set of all si tuati ons we desi gnate S.
The coal i ti ons havi ng chosen and carri ed out thei r strategi es (1.2
above), the out come is assumed to be compl etel y predetermi ned, i.e. the
si tuati on is uni vocal . Thi s assumpti on is l ess restri cti ve than i t mi ght
seem. Suppose the choi ce of strategy by the coal i ti ons does not l ead
uni vocal l y to a parti cul ar physi cal out come but l eaves open the possi bi l i ty
of any of the out comes of a subset of the set of all outcomes. The
uncertai nty here coul d be of t wo ki nds. 1) The out come i s r andomand the
pmbabi l i ti es of i ts real i zati ons depend on the strategi es chosen by the
acti ng coal i ti ons. We can consi der i t as a si ngl e, compl ex outcome, i.e. a
si ngl e si tuati on; for exampl e, the confl i ct i s a chess mat ch i nwhi ch the
wi nner gets a l ottery ticket as a pri ze. 2) The strategi es chosen by the
acti ng coal i ti ons can have di fferent outcomes to whi ch no probabi l i ti es
can be assi gned. Her e, at l east one other factor not i ncl uded i nconstruc-
ti ng the model is pl ayi ng a rol e. Hence, the model does not descri be the
confl i ct adequatel y, and must be revi sed.
Fourthly, we must i ncl ude all the parti es i nterested in any way in
parti cul ar out comes of the confl i ct. Li ke the acti ng factors, they too may
be col l ecti ve and structured and can be cal l edi nterest coal i ti ons. The set
of all i nterest coal i ti ons is sii.
In real confl i cts, the operati onal and i nterest coal i ti ons usual l y coi nci de,
but this is not a ri gi drul e. A footbal l f an watchi ng a mat ch on tel evi si on
Game theory 207
has an i nterest inthe out come but cannot i nfl uence the resul t. On the other
hand, the ref eree has no i nterest inthe resul t, (or shoul d have none. . .), but
exerci ses consi derabl e i nfl uence on the course of the mat ch and i ts resul t.
Fgthly, we must i ndi cate i nwhat the i nterests of the i nterest coal i ti ons
consi st. Thi s is done by showi ng, for each, pai rs of comparabl e si tuati ons
and i ndi cati ng, for each comparabl e pai r, whi ch of the t wo i s preferabl e. It
is not essenti al , however, that any t wo si tuati ons for each i nterest
coal i ti on be comparabl e. I ndeed, i t may happen that a parti cul ar si tuati on
will be absol utel y non- comparabl e (exactl y as inethi cs no compari son is
al l owabl e between the val ue of money and the val ue of human life). The
fact that si tuati on s for i nterest coal i ti on K i s preferabl e to si tuati on t
will be desi gnated by s t-k t.
Very often, the preferenti al rel ati onshi p for i nterest coal i ti ons is des-
cri bed as fol l ows. For each coal i ti on of i nterests K for all si tuati ons S is
defi ned anumeri c functi on Hk and s +fi t i s accepted where H k (s) > H k (t).
The functi on Hk is cal l ed the payoff functi on of the coal i ti on.
It fol l ows that any confl i ct may be f ormal l y descri bed as a qui ntupl e
i n whi ch the meani ng of each l etter is as expl ai ned above; the subscri pt
near each of the braces i ndi cates that the content of the braces is vari abl e
and depends on the i ndex that accompani es i t and that the i ndex itself
vari es wi thi n the bounds of the set, as fol l ows f r omthe symbol s adj acent
to the braces.
The f ormal model of the confl i ct, represented by the system("), i s
cal l ed the game. Basi cal l y, the mathemati cal theory of games deal s with
the study of confl i cts treated as games, inthe sense of thi s defi ni ti on.
Thi s defi ni ti on of games may appear too compl i cated and abstract.
However , i t reproduces i n a f ormal l anguage all the features whi ch are
i nevi tabl y i nherent inany confl i ct, wi thout at the same ti me contai ni ng
any of the i ndi ces proper onl y to speci al i sol ated cl asses of confl i cts.
1.4The formal i zati on of the concept of confl i ct l ed us to the game
concept. The formal i zati on of the noti on of opti mal i ty, on the contrary,
is not a prel i mi nary step for worki ng out game theory, but a f undament al
probl emof the theory itself. Moreover, i t i s ingame theory that opti mal -
i ty, its nature and i ts vari ants, becomes a maj or sci enti fi c probl emand,
above all, a mathemati cal one (see Secti on 4).
208 NiColui N. Vorobyev
2. EXAMPLES OF GAMES
The fol l owi ng exampl es will hel p to expl ai n.
2.1 Morra : Thr ee pl ayers, I , I I and I I I , each si mul taneousl y show 1 or 2
fi ngers. If all t hree show the same number of fi ngers, the payof f for each is
nil. If one of the pl ayers shows a number of fi ngers di fferent f r omthe
number shown by the other t wo, he wi ns as many uni ts as the number
of fi ngers he has shown. The wi n is pai d f r omthe bank, whi ch is i ndepen-
dent of the gai ne. What are the pri nci pal component s of thi s game?
Cl earl y the pl ayers are i nvol ved i nbot h acti ng and i nterest coal i ti ons.
Theref ore we can wri te:
sa = st = ( I , I I , Irr>
Each pl ayer has t wo strategi es, 1 and 2. Fori nal l y, thi s is represented:
SI = SII = sIII = (I, 2)
Each triplet consi sti ng of a strategy I , a strategy I I and a strategy I I I
f orms a si tuati on. The game cl earl y i nvol ves 2 x 2 x 2 = 8 si tuati ons.
The preferences are determi ned by the payoff functi ons. The payoff
val ues may be represented as fol l ows:
SI TUATI ONS
111 112 121 122 211 212 221 222
Payof f of Pl ayer I
Payof f of Pl ayer I I
Payof f of Pl ayer I I I o 2
2.2 Two bandi ts ( Pri soner s di l emma): Two bandi ts (Pl ayers I and 11)
gui l ty of a seri ous cri me are inpri son on remand. Each can confess or
deny. I f both confess they will both be i mpri soned, with al l owance for
the mi ti gati ng ci rcumstances of the confessi on (mandatory sentence 8
years). I f both persi st i n denyi ng the cri me, then, gi ven the l ack of di rect
evi dence, they will be convi cted onl y for the illegal use of arms
(mandatory sentence 1 year). I f one confesses and the other does not, the
one who has confessed goes f ree (such are the l aws of the State inwhi ch
Game tlzeory 209
the acti on takes pl ace) and his accompl i ce recei ves all the puni shment
(a 10-year sentence). I n this game each pl ayer has t wo strategi es (Con-
fessi on or Deni al ), so that the total number of si tuati ons inthe game is
2 x 2 = 4. The payoffs can be summari zed i nt wo matri ces:
I s payoff
I I s strategy
I I s payof f
I I s strategy
C D C D
Is strategy D c I -8 -10 -; 1 I s strategy D cn
The payoffs shown i n the matri ces are negati ve, because they represent
l osses tothe pl ayers.
2.3 Stone, paper, sci ssors. Two pl ayers (I and II) si mul taneousl y make
the si gn for one of the three (stone, paper, sci ssors) (a cl enched fi st means
stone, the cupped hand paper and the V- si gn scissors).
I f the t wo pl ayers si gnal the same obj ects, the payof f for each i s nil. I n
other cases, stone wi ns 1 agai nst sci ssors (stone breaks sci ssors) and
l oses 1 agai nst paper (paper covers stone), and in turn paper l oses 1
agai nst sci ssors (sci ssors cut paper).
As inthe previ ous exampl e, the pl ayers here are i nvol ved i r, both acti ng
and i nterest coal i ti ons.
Thus,
sa = As= {I, 11)
Each pl ayer has three strategi es: stone, paper, sci ssors. Formal i zed,
this l ooks l i ke:
SI = SII = (S, P, SC>
Each pai r consi sti ng of a strategy 1 and a strategy I I f orms a si tuati on.
Cl earl y the game presents 3 x 3 = 9 si tuati ons. The preferences are
determi ned by the payoff functi ons. The val ues of the payoff functi on
f or I can be descri bed as fol l ows:
210 Nicolai N. Vorobyev
IIs strategy
S P sc
P
I s strategy
The payof f of Pl ayer II is equal to I s payoff but wi th the opposi te si gn,
that i s, to hi s l oss.
2.4 The duel : The adversari es move towards each other f romstarti ng
poi nt a to a fi nal poi nt b; each may fi re one shot at hi s opponent. The
duel ends when at l east one of the duel l i sts i s hit. The duel l i st who has
hit hi s adversary and remai ns untouched hi msel f scores 1 (his opponent
at the same timel oses 1). If by shooti ng si mul taneousl y, both score a hit
or both miss, each scores O. The preci se moment i n whi ch he will fire
i s pl anned by each pl ayer in advance. The probabi l i ty of hi tti ng the
opponent f r omdi stance x is p(x).
Here the pl ayers strategi es are the di stances at whi ch they pl an to
fi re. By a sui tabl e sel ecti on of the ori gi n and of the scal e of the di stances,
i t can be posed that a = O and b = 1. Thus the strategi es will be represen-
ted by number s between O and 1, and the si tuati ons by the pai rs of those
numbers. El ementary cal cul ati ons of probabi l i ti es show that the val ue of
the payof f functi on for Pl ayer I i n si tuati on (x,y) i s defi ned by the
equati ons:
P(4 - P(Y>+ P(X> P(Y>, if x < Y
Hl(x7 y) = o , i f x=y
{ P(4 - P(Y>- P(4 P(Yh i f x > Y
The payof f functi on of Pl ayer I I differs f r omthe above onl y i ni ts si gn.
2. 5 Chess: A game of chess i s supposed to be pl ayed accordi ng to or-
di nary rul es. There are t wo pl ayers, Bl ack and Whi te. The strategy of
each is a systemof choi ces of admi ssi bl e moves in every posi ti on in
whi ch he pl ays. Mor e f ormal l y, the strategy of each pl ayer is every
functi on defi ned for all the posi ti ons he pl ays, and the val ue of the
functi on for t he posi ti on is an admi ssi bl e move inthat posi ti on.
Game theory 211
The si tuati on as a pai r of t wo strategi es compl etel y determi nes the
course and out come of the pl ay. The payoff for each pl ayer i s 1 ineach
si tuati on in whi ch he wi ns, - 1 in each si tuati on inwhi ch he l oses, and
O ina draw.
2.6 Di vi di ng the cake: A mot her (sol e acti ng coal i ti on) has made a cake
(whi ch we can take as a uni t) for her t hree chi l dren. She i s abl e to di vi de
i t arbi trari l y i nto three parts: x1 for the ol dest (tobe cal l ed l), x2 for the
second (cal l ed2) and x3 for the youngest (cal l ed 3). Thi s means that the
set of strategi es for the mot her is the set of all possi bl e tri pl ets of the
f or m(XI, x2, x3) where XI, x2, x3 2 O and XI + x2 + x3 = 1.
Geometri cal l y, for each of these tri pl ets i t is possi bl e to fi nd a corres-
pondi ng poi nt of a tri angl e for whi ch the number s XI , x2 and x3 will be
the barycentri c co-ordi nates (readers unfami l i ar wi th this geometry need
not bother about i ts exact meani ng and may be sati sfi ed merel y wi th the
possi bi l i ty of such correspondence).
It is pl ai n that here the mother s strategi es are the si tuati ons of the
game. The chi l dren persuade one another of the fai rness (and hence
necessi ty) of one di vi si on or another of the cake, by usi ng any argument s
avai l abl e (and probabl y al so thei r fists). It soon becomes cl ear that none
of t hem(not even the el dest) acti ng al one can get hi msel f even the
smal l est cr umb of the cake. Together the t wo ol der ones (1 and 2) can
ensure a di vi si on that al l ows the t wo of t hem0.7of the cake. If they
recei ve mor e, they can consi der i t a wi ndfal l , to whi ch they have no
reasonabl e cl ai m. Si mi l arl y, 1 and 3 mi ght together obtai n 0.6 of the
cake, and 2 and 3 together 0. 5.
Figure 1
212 Nicolai N. Vorobyev
Thus the i nterest coal i ti on i s represented here not by the three i ncl i vi d-
ual chi l dren but by thei r possi bl e pai ri ngs, three innumber.
To descri be the preferenti al rel ati ons for each of the three coal i ti ons
of i nterests, we can consi der t wo si tuati ons (two parti ti ons that are
frequentl y cal l ed i mputati ons) x = (xl, xz, x3) and y = (YI, yz, y3). The
(1,2) coal i ti on prefers parti ti on x to parti ti on y (the f ormal expressi on is
x e-1,~ y) if, fi rst, x1 + x2 < 0.7, and second, if the share of x of each
member of the coal i ti on i s l arger than his share of y, that i s, if XI > y1
and x2 > y2. Further, i t i s not any t wo parti ti ons whi ch l end t hemsel ves to
compari son, but onl y those where the sumof the fi rst t wo component s
does not exceed 0.7 (inFi g. I this is the tri angl e M3N).
The preferenti al rel ati ons for the other t wo coal i ti ons of i nterests are des-
cri bed si mi l arl y. Thus the game takes the f or mi ndi cated at (*) above.
3. CLASSIFICATION OF GAMES
3.1 The defi ni ti on of games as systems (*) f ormul ated i n Secti on 1
provi des a conveni ent basi s for the cl assi fi cati on of games. An el aborated,
t hough inno way exhausti ve, cl assi fi cati on i s gi ven inVorobyev (1970).
It may be noted that, f r omthe poi nt of viewof f ormal devel opment,
i t ems of the f or m(*), in whi ch the i nterest coal i ti ons set is empt y or
compri ses onl y one coal i ti on of i nterests, are al so games, but the fi rst
bel ongs i n the study of the purel y descri pti ve theori es whi ch descri be
processes occurri ng innature or soci ety, wi thout havi ng any obj ecti ve.
The study of second type items i s concerned wi th probl ems whose
mathemati cal content consi sts in the defi ni ti on of extremes of gi ven
functi ons under gi ven constrai nts. Proper games begi n f r omthe poi nt
where genui ne confl i cts ari se, i.e. where at l east t wo parti es are i nvol ved.
Accordi ngl y, we shal l deal f romnow on wi th games i nvol vi ng t wo or
mor e coal i ti ons of i nterests.
The fi rst i ndi cator by whi ch to cl assi fy games proper i s the uni ci ty or
mul ti pl i ci ty of the acti ng coal i ti ons. Wi t h several coal i ti ons the theory
of course becomes mor e compl i cated, but i mmedi atel y bri ngs to l i ght
questi ons of much i nterest and practi cal i mportance; inthe case of games
wi th a si ngl e coal i ti on, extremel y deep probl ems have to be tackl ed
ri ght away. We shal l begi n by exami ni ng an i mportant cl ass wi th many
acti ng coal i ti ons, non- co- operati ve games.
Game theory 213
3.2 I n a non- co- operati ve game each acti ng coal i ti on is at the same time
an i nterest coal i ti on, and vi ce-versa. The set of all the pl ayers is desi g-
nated by I, the pl ayers themsel ves by the number s 1, 2 ... n. So that i n
non- co- operati ve games we have 52, = 53s = I = (1,2 ... n]. The
strategy of pl ayer i consti tutes a set SZ and each n-tupl e s = ($1, s2, ..., sn)
where SI E $6 i s a si tuati on. Si nce the set of si tuati ons is well defi ned by the
set of strategi es, i t wi l j not necessari l y be speci fi cal l y menti oned i n the
i denti fi cati on of the game.
Let the payoff of pl ayer i in si tuati on s equal H&). The fi xi ng of the
payoff functi ons of Ht compl etes the i denti fi cati on of the non- Co-
operati ve game, whi ch takes the f orm:
<J, iS*>Ze J { H 4 ] Z ~ J >
Mor r a and Two bandi ts are exampl es of this general type of non-
co-operati ve game.
3.3 Among the vari eti es of non- co- operati ve games a speci al pl ace i s
reserved for those inwhi ch the number of pl ayers i s t wo: It = 2 (tradi -
ti onal l y cal l ed Pl ayer I and Pl ayer 11) and the payoffs of Pl ayer I I in
any si tuati on are of equal val ue but opposi te si gn to those of Pl ayer I:
H-(s) = - HI&+ (**I
Such games are cal l ed zero- sumt wo- person games (because the equati on
(**) may be rewri tten HI($) + HI&) = O) or synonymous, antagoni sti c
games.
The set of pl ayers inthe two- person game bei ng al ways the same, i t is
not necessary to i ndi cate i t every t i me; al so, i t fol l ows f r om(**) that the
val ue of the payof f functi on for Pl ayer I i t is al ways possi bl e to fi nd the
val ue of the payof f functi on for Pl ayer I I . Thus, i ni denti fyi ng the t wo-
person game one states onl y the payof f functi on for the fi rst pl ayer, and
the number of the pl ayer as the subscri pt of the f uncti on i s omi tted. I n
thi s case, we somet i mes speak onl y of the payoff f uncti on of the game,
meani ng thereby the gai ns of Pl ayer I i n the vari ous si tuati ons or, what
amount s to the same thi ng, the l osses of Pl ayer I I .
3. 4 The second i ndi cator that provi des a natural basi s for cl assi fi cati on
of two- person games l i es inthe nature of the pl ayers set of strategi es. I n
parti cul ar i f the sets are finite, the game is al so cal l ed fi ni te. I n the op-
214 Nicolai N. Vorobyev
posi te case, the game i s cal l ed i nfi ni te. The game of Stone, paper, sci ssors
i s a t wo- person finitegame. Taki ng this game as an exampl e, we can see
that the val ues of the payoff functi on ina t wo- person finitegame can be
arranged as a matri x. Accordi ngl y, two- person finite games are often
sai d to be matri x games.
As an exampl e of a two- person i nfi ni te game we can take the duel . I n
thi s game the si tuati on, bei ng a pai r of number s between O and 1, may be
shown as a poi nt of a uni tary square on the pl ane of the co-ordi nates.
All t wo- person games for whi ch such presentati on is possi bl e are often
cal l ed games on a square.
3.5 W e have so far assi gned no speci fi c properti es to the pl ayers
strategi es, consi deri ng t hemsi mpl y as el ements of abstract sets. However ,
i nreal i ty, the choi ces made by the pl ayers often prove to be, not mo-
mentary deci si ons but sequence of parti al deci si ons or even processes
constantl y unfol di ng inti me. Such games are cal l ed games in extensi ve
f or m(or posi ti onal or dynami c games). Chess is an exampl e. Other
exampl es are most outdoor games and card games, and confl i cts wi th
pursui t or capture el ement s of any ki nd.
3.6W e now come to the games with onl y one acti ng coal i ti on. The
behavi our of this si ngl e coal i ti on obvi ousl y compl etel y determi nes the
out come of the confl i ct. Accordi ngl y, the coal i ti ons strategi es can be
i denti fi ed wi th the si tuati ons and need be menti oned no further inwhat
fol l ows. That is why we someti mes call such games non-strategi c games.
As a rul e, i n non-strategi c games, a set of i nterest coal i ti ons is regar-
ded as a parti cul ar fami l y of subsets of the set of pl ayers I. Usual l y, the
payoffs (inthi s case i t woul d be appropri ate to say the payof f vectors)
of all the pl ayers ineach si tuati on are gi ven and the preferred choi ce f r om
the comparabl e si tuati ons for each i nterest coal i ti on i s that gi vi ng the
best payof f for each of the pl ayers i n the coal i ti on.
A very i mportant cl ass of non-strategi c games consi sts of those in
whi ch the si tuati ons are vari ous parti ti ons ( i mputati ons ) of some
unl i mi ted di vi si bl e obj ect; for each coal i ti on the maxi mumshare of the
obj ect that that coal i ti on can ensure for itself is i ndi cated. Such games
are cal l ed co-operati ve games. Di vi di ng the cake is an exampl e.
Garne theory 215
4. FUNDAMENTAL PROBLEMS IN GAME THEORY
4.1 The content of game theory i s opti mal behavi our in confl i cts
rather than the study of the confl i cts themsel ves. That is why bot h the
descri pti ve exami nati on and exposi ti on of the game as a f ormal con-
cepti on, and the constructi ve questi ons concerni ng the process of i denti -
fyi ng the game, take a subordi nate pl ace in game theory, at l east at
present. The mai n aspect of game theory is normati ve; i t speci fi es the
goal s of i nterested parti es and reveal s the expedi ent acti vi ti es of acti ng
parti es.
Opt i mal si tuati ons and strategi es i na game are usual l y sai d to be i ts
sol uti ons. Game theory i s occupi ed wi th mathemati cal representati on
of the rel ati onshi ps between the condi ti ons of the game and i ts sol uti ons.
Theref ore for every game, as well as every cl ass of games the three
f undament al questi ons are: What do we mean by sol uti on? (what is the
very noti on of opti mal i ty i nthis game?); Do opti mal sol uti ons exi st for
thi s parti cul ar game?; What are they and how can they be f ound? These
are the three mai n questi ons of game theory.
It was J. von Neumann who posed and resol ved inpart these questi ons
inhi s arti cl e of 1928. They are al so the essenti al t heme of the monogr aph
by von Neumann and Morgenstern, publ i shed i n 1944, whi ch l ai d the
f oundati ons of game theory as a mathemati cal sci ence.
4. 2 Let us begi n wi th the fi rst questi on. So far, no one has f ound
opti mal i ty pri nci pl es suffi ci entl y preci se to appl y to all games, but they
have been f ormul ated for certai n cl asses of games.
Many behavi our pri nci pl es i ntui ti vel y felt to be opti mal i ty pri nci pl es are
infact nothi ng of the ki nd - f ormal anal ysi s shows t hemtobe unachi ev-
abl e, e.g. the nai ve i dea of each i nterest coal i ti on achi evi ng i ts opti mal
si tuati on and the maxi mumval ue of i ts payof f functi on is practi cabl e i n
onl y a very f ewcases of very l i ttl e i nterest.
As a basi s for many pri nci pl es of opti mal i ty we can take certai n
features of tradi ti onal non-confl i ct opti mal i ty, e.g. the si tuati on i s
opti mal if no devi ati on can i mprove it. Let us extend thi s to non-
co-operati ve games. The opti mal si tuati on will now be a si tuati on in
whi ch no pl ayer who devi ates f romi t (and all other pl ayers keep to the
strategi es al ready chosen) can i ncrease hi s wi n by so doi ng. Such
si tuati ons are cal l edequi l i bri umsi tuati ons.
216 Nicolai N. Vorobyev
4. 3 In a two- person game, the equi l i bri umof a si tuati on means that any
devi ati on f romi t by Pl ayer I can onl y reduce his wi nni ngs, and devi ati on
by Pl ayer I I woul d reduce what he inturn stands to wi n, i.e. reduce the
l osses of Pl ayer I , whi ch amount s to i ncreasi ng payoff I. Thi s l ends itself
to an i nteresti ng geometri cal i nterpretati on. Let a t wo- person game
consi st inthe choi ce of a poi nt on a terrai n, Pl ayer I choosi ng i ts l ati tude,
Pl ayer I I i ts l ongi tude, and the al ti tude of the poi nt representi ng the
payof f functi on. Suppose there i s a chai n of mountai ns ri si ng above the
terrai n, runni ng north- south; a l owsaddl e inthe chai n corresponds to the
equi l i bri umsi tuati on inthe game. Because of the possi bi l i ty of such an
i nterpretati on, si tuati ons of equi l i bri umintwo- person games are often
cal l ed saddl e-poi nts.
Agai o, i t so happens that for two- person games (as can be demon-
strated mathemati cal l y wi thout much troubl e) the pri nci pl e of opti mal i ty
requi ri ng the ai mi ng at (stri vi ng for) saddl e-poi nts i s equi val ent to the
pri nci pl e requi ri ng ai mi ng at (stri vi ng for) maxi mi zati on of the l east
possi bl e, i.e. the absol utel y guaranteed, gai n; thi s is known as the maxi mi n
pri nci pl e. The strategi es whereby Pl ayer I maxi mi zes his mi ni mumgai n
(and Pl ayer I I mi ni mi zes his maxi mumloss) are cal l edopti mal strategi es.
4. 4 Equi l i bri umsi tuati ons, and they onl y, can secure an agreement be-
tween the pl ayers whi ch i t will be i nno party s i nterest to vi ol ate uni l zter-
al l y. If any attempt is made toi ncorporate non- equi l i bri um, i t will be in the
i nterest of at l east one pl ayer f romthe very start tovi ol ate the agreement.
Another pri nci pl e of opti mal i ty innon- co- operati ve games i s Pareto s
opt i mum: a si tuati on i s Pareto- opti mal i f there i s no other si tuati on whi ch
woul d be at l east as f avourabl e as for all pl ayers and even mor e favourabl e
for at l east one of them.
Pareto s opti mal i ty i s ina way at the opposi te pol e to Nash s equi l i b-
ri um. I n a Nash equi l i bri umsi tuati on, i ndi vi dual i mprovement s cannot
be secured by i ndi vi dual devi ati ons, whi l e Pareto opti mal i ty si gni fi es that
general i mprovement i s not possi bl e under general devi ati ons. Pareto s
opti mal i ty may be inf ormal contradi cti on withNash s equi l i bri um. For
exampl e, i n Two bandi ts the confessi on of both bandi ts resul ts i n an
equi l i bri umsi tuati on, while the si tuati on when the bandi ts refuse i s
opti mal inPareto s sense.
In two- persons games, Pareto s opti mal i ty has no rel evance si nce, there,
any si tuati on is opti mal inhis sense.
Game theory 217
4.. 5 As non-strategi c games have been mor e deepl y s udi ed than non- co-
operati ve games, a l arger number of rel ati vel y refi ned opti mal i ty pri n-
ci pl es have been worked out for t hem.
One cl assi cal defi ni ti on of the opt i mumi s the val ue of a functi on such
that no other val ue is preferabl e to it. Appl yi ng this to non-strategi c
games, the opti mal si tuati on is one to whi ch no other is preferabl e for
any coal i ti on of i nterests. The set of all si tuati ons wi th this property
is cal l edthe core of the game (whi ch may compri se many Situations). I n
Di vi di ng the cake the core consi sts of all the si tuati ons bel ongi ng to the
tri angl e ABC.
I f such opti mi zati on has several sol uti ons, i.e. if the obj ecti ve functi on
has several opti ma, we may defi ne opti mal i ty i nthe fol l owi ng manner.
The set R of val ues of the functi on consi sts of opti ma, if
1) none of these val ues is preferabl e to another;
2) for any val ue taken outsi de the set R, a preferabl e val ue can be
f ound wi thi nit.
Now l et us appl y thi s noti on of opti mal i ty to non-strategi c games. The
out come is that the set of si tuati ons R i s composed of opt i ma if
1) for nei ther of t wo si tuati ons r1, r2 E R and K E Ai is i t possi bl e that
r1 -K r2;
2) whatever the si tuati on s d R there will be a si tuati on r E R and a
coal i ti on of i nterests K E Sa such that r t~ s.
Any set of si tuati ons possessi ng these t wo properti es is cal l ed the sol u-
ti on of the game inthe von Neumann- Mor genst er n sense or the stabl e set
(1944); the properti es themsel ves are referred to as i nternal and external
stabi l i ty.
Such a sol uti on can be appl i ed to a parti cul ar systemof l egal and
ethi cal nor ms in a soci ety, that is, appl yi ng to a maj ori ty of pl ayers I.
I nternal stabi l i ty then means that inthe soci ety i nI there are no forces
(i nterest coal i ti ons K) i nterested i npassage f romone nor mto another;
external stabi l i ty means that inthe event of any devi ati on f r omthe nor m
(i.e. any goi ng outsi de of the limits of the sol uti on) some forces i n the
soci ety will want to re-establ i sh it.
By si mpl e l ogi cal anal ysi s i t is easy to conf i rmthat, if a game has a core
and a sol uti on (cf. von Neumann- Mor genst er n, 1944), the core must lie
in the sol uti on. For exampl e, in Di vi di ng the cake the sol uti on is
represented by the tri angl e ABC pl us the curvi l i near segments AD, BE
and CF. These segments are not uni vocal l y determi ned and may be
218 Nicolai N. Vorobyev
traced under some restri cti ons very arbi trari l y; inparti cul ar they may be
recti l i near segments perpendi cul ar tothe correspondi ng si des of the great
tri angl e.
4.6 As pri nci pl es of opti mal i ty one can consi der sati sfyi ng, si tuati ons
shoul d contai n certai n condi ti ons of fai rness. I f a pl ayer i ncreases the
gai n of any col l ecti ve by exactl y the amount he secures for hi msel f acti ng
i ndi vi dual l y, ina fair si tuati on he must recei ve as much as he woul d have
recei ved wi thout j oi ni ng any col l ecti ve. Agai n, if t wo pl ayers i ncrease the
gai ns of any col l ecti ve by equal amounts, ina fair si tuati on thei r shares
must be equal . Devel opi ng thi s approach, Shapl ey constructed the defi ni -
ti on of a fai r si tuati on. For exampl e, i n a fair di vi si on i n Di vi di ng the
cake , the el dest chi l d woul d recei ve Q + of the cake, the mi ddl e chi l d
-5, and the youngest Js - &.
4.7 The second of the mai n probl ems of the game theory is that of the
real i zabi l i ty of parti cul ar pri nci pl es of opti mal i ty. Otherwi se, the whol e
exerci se is poi ntl ess. As i ndi cated earl i er, an i ntui ti ve attracti ve pri nci pl e
al l owi ng the most f avourabl e si tuati ons for all i nterested parti es is very
rarel y workabl e inpracti ce and is thus vi rtual l y usel ess.
Other pri nci pl es of opti mal i ty menti oned above appear mor e promi s-
i ng, but the actual demonstrati on remai ns difficult.
Zermel o has proved the exi stence of opti mal strategi es in a broad
cl ass of dynami c games, i ncl udi ng chess. Yet, si mpl e games like Morra ,
Stone, paper, sci ssors and Duel have no equi l i bri umsi tuati ons. Thi s
can be conf i rmed very easi l y.
To escape f r omthi s i mpasse one can appl y an approach that is typi cal
inpure mathemati cs.
Al ongsi de the pl ayers ori gi nal strategi es, new general i zed strategi es are
i ntroduced; f r omthese, si tuati ons are constructed to whi ch the defi ni ti on
of the payof f functi on is extended. Equi l i bri umsi tuati ons are sought
among these newsi tuati ons. The new strategi es, si tuati ons and expanded
payoff functi ons must of course have a natural and pl ausi bl e i nter-
pretati on interms of the ori gi nal strategi es, si tuati ons and payoff f unc-
ti ons.
The fol l owi ng extensi on of non- co- operati ve games is very frui tful .
Suppose each pl ayer can not onl y choose any parti cul ar strategy, but
sel ect strategi es at r andom, wi th arbi trary, but wel l -defi ned probabi l i ti es.
Game theory 219
These courses of acti on are cal l ed mi xed strategi es. The i ndependent
choi ce of mi xed strategi es by the pl ayers gi ves a mi xed strategy si tuati on,
inwhi ch the gai n of each pl ayer i s a r andomvari abl e whose mathemati cal
expectati on i s consi dered to be i ts payoff f uncti on val ue.
On the basi s of each non- co- operati ve game a new game is now con-
structed whi ch i s cal l ed the mi xed extensi on of the fi rst game. Nash has
shown that each fi ni te non- co- operati ve game has an equi l i bri umsi tuati on
in mi xed strategi es, i.e. in i ts mi xed extensi on. Thi s was demonstrated
earl i er by von Neumann for finitet wo- person zero- sum(matri x) games.
The equi l i bri umsi tuati on in Morra appears to be when each pl ayer
shows one fi nger wi th the probabi l i ty of 1/(1 + 42) (and therefore t wo
fi ngers withthe probabi l i ty of 42/(1 + 2/2)).
I n Stone, paper, sci ssorsy, each pl ayer s mi xed opti mal strategi es
consi st i n choosi ng between the t hree strategi es S, P, and SC wi th a 4
probabi l i ty each.
4. 8 For the non-strategi c games - even for the speci al cl ass of co-
operati ve games - the si tuati on seems to be not as sati sfactory as for the
non- co- operati ve games.
Games withno core (or mor e accuratel y, where the core is empty) are
easy to fi nd. I n Di vi di ng the cake for exampl e, if the shares that the pai rs
of chi l dren can secure i ncrease to 0. 8, 0.7, and 0.6respecti vel y, the
resul ti ng game has an empt y core.
Have i ndi vi dual co-operati ve games stabl e sets (i.e. sol uti ons inthe von
Neumann- Mor genst er n sense) ? The questi on i s very compl i cated. For
many (but very narrow) cl asses of games, the exi stence of a sol uti on is
proved; for arbi trary 5- person games, the questi on i s sti l l open. Qui te
recentl y (1967), an exampl e of a 10- person game wi th no sol uti on has been
constructed by Lucas.
Further, if the game has a stabl e set, firstly, i t may be non- uni que and,
secondl y, may consi st of several si tuati ons. Consequentl y, a compl emen-
tary probl emhas to be deal t wi th for sol vabl e games: sel ecti ng, f r omall of
the sol uti ons, and then f r omall the si tuati ons composi ng the sol uti on
sel ected, the sol uti on and si tuati on wi th, so to speak, the hi ghest
opti mal i ty.
It is at l east some rel i ef to r emember that a Shapl ey vector exi sts for
every co-operati ve game, that i t i s uni que, and consi sts of a si ngl e
si tuati on.
220 Nicolai N. Vorobyev
4.9 The thi rd f undament al questi on (see p. 215) i s, what are the opti mal
sol uti ons (if such exi st) and how can they be f ound? Thi s is an auto-
nomous and very i mportant probl emwhi ch is sti l l open, despi te the
amount of game theory l i terature devoted to i t. Hi therto, sol uti ons
have been f ound for fai rl y narrowcl asses of games onl y. Even then, the
fi nal answer usual l y appears not in the f or mof anal yti c f ormul ae, but
as an al gori thm. Hence, further devel opment is cl osel y ti ed inwiththe
mathemati cal devel opment and the use of comput er techni ques.
5. GAMES PLAYED AGAINST NATURE
As a rul e, incondi ti ons of confl i ct, no party i ntends togi ve his adversary
i nf ormati on about hi s pl an and i mpendi ng acti ons. Consequentl y,
opti mal deci si on- maki ng in confl i ct condi ti ons means deci si on- maki ng
under uncertai nty.
I n condi ti ons of uncertai nty, inturn, there i s a set of vari ants of the
condi ti ons inwhi ch the deci si on is to be taken; and the party taki ng the
deci si on has no preci se knowl edge ei ther of these condi ti ons or even of
the a priori probabi l i ty di stri buti on of the vari ants. The di fferent vari ants
can be treated as the strategi es of a pl ayer l abel l ed nature , so that
deci si on- maki ng under uncertai nty i s equi val ent to deci si on- maki ng
under confl i ct (i.e. confl i ct withnature).
Uncertai nty and confl i ct bei ng thus cl osel y rel ated, we can defi ne game
theory as a theory of mathemati cal model s for the maki ng of opti mal
deci si ons under confl i ct or uncertai nty. Let us now transl ate thi s i nto the
l anguage of game theory.
To take the si mpl est case: a party who i s maki ng the deci si on and
nature, his opponent, are the t wo pl ayers inthe non- co- operati ve game
we are constructi ng. His strategi es are obvi ousl y the deci si ons avai l abl e
to hi mand nature s strategi es are, as al ready noted, the possi bl e com-
bi nati ons of ci rcumstances whi ch cannot ei ther be predi cted or forecasted
exactl y.
The val ue of the pl ayers payoff functi on is the extent to whi ch he
attai ns his purpose, this depends on his deci si on and on nature s actual
strategy. What i s nature s payoff f uncti on? By the way in whi ch the
defi ni ti on i s made, the anthropomorphi zati on of nature can be avoi ded
by attri buti ng to the pl ayer some psychol ogi cal atti tude towards his
Game theory 221
uncertai n gai n (or, what amount s to the same, t owards his uncertai n
losses).
Assume he i s concerned to maxi mi ze his guaranteed ( mi ni mum) gai n.
Thi s means he must act as if nature i s hi s consci ous enemy, ani mated by
the desi re to inflict maxi mumdamage on hi m. Thi s makes the game
pl ayed agai nst nature, i.e. deci si on- maki ng under uncertai nty, a t wo-
person game; regardi ng i t as such is a l ogi cal consequence of the assump-
ti on that the deci si on- maker assumes the max- mi n pri nci pl e as hi s
pri nci pl e of opti mal i ty (see 4.3). Thus, to establ i sh the case for the t wo-
person game agai nst nature, all that i s requi red i s to establ i sh the
soundness of the max- mi n pri nci pl e.
I n deci si on- maki ng probl ems i t is extremel y di ffi cul t to veri fy the
pri nci pl es of opti mal i ty froi n a consi derati on of thei r consequences. The
al ternati ve is to deduce the pri nci pl e we want f r oma fewaxi oms that
seemto be reasonabl e or that can i mmedi atel y be experi mental l y veri fi ed.
A set of axi oms that necessari l y bri ngs i nthe max- mi n pri nci pl e has
been proposed by Mi l nor (1964); Vi l kas has constructed another based
upon the i dea of a fair gai n under uncertai nty.
Natural l y, max- mi n does not sati sfy all the usual requi rements for
opti mal i ty. Mi l nor hi msel f has shown that if, i nhis set of axi oms, some
axi oms are repl aced by others, somet i mes no l ess l i kel y, another pri nci pl e
of opti mal i ty is obtai ned i nstead of the max- mi n pri nci pl e. Nothi ng that
has been sai d shoul d be taken as di scredi ti ng the max- mi n pri nci pl e, but
merel y as evi dence that deci si on- maki ng under uncertai nty bristles wi th
psychol ogi cal nuances wi th sti l l el ude formal i zati on. We may hope in
timeto l earn how to ci rcumscri be exactl y and f ormal l y the cl asses of
probl ems of opti mal deci si on- maki ng under uncertai nty in whi ch the
max- mi n pri nci pl e coul d be used wi thout reservati ons, as well as the
cl asses of probl ems where i t coul d not be so used.
6. GAME THEORY AND STATISTICAL DECISIONS
6.1Certai n authors (e.g. Chernof f and Moses, 1959) defi ne mathemati cal
statistics as a theory of deci si on- maki ng under uncertai nty. I n the l i ght of
what has been sai d above, that means that statistics is cl osel y rel ated to
game theory. In poi nt of fact, stati sti cs is qui te cl earl y goi ng to be
consi dered a consti tuent part of game theory, di sti ngui shed by i ts
222 Nicolui N. Vorobyev
speci fi c probl emrange and i ts termi nol ogy. Thi s of course in no way
di mi ni shes the range and si gni fi cance of mathemati cal statistics.
6.2 Wal d (1950) has systemati cal l y studi ed the approach to stati sti cal
probl ems f r omthe vi ewpoi nt of games theory. His book ci tes several
poi nts of correspondence between f undamental concepts i nstatistics and
t wo- person game theory.
Two-person game
Pl ayer I
Pl ayer I I
Set of strategi es of Pl ayer I
Set of strategi es of Pl ayer I I
Payof f of Pl ayer I
Mi xed strategy of Pl ayer I
Opti mal strategy of I
Opti mal strategy of I I
Problem of statistical
decision-making
Nature
Stati sti ci an
Cl ass of di stri buti on functi ons
(states of nature set)
The stati sti ci ans Deci si on
functi on
The stati sti ci ans ri sk
A priori di stri buti on over states
of nature set
Least f avourabl e a priori
di stri buti on over states of
nature set
Deci si on on the mi ni - max
pri nci pl e.
These correspondences can be taken further. The further devel opment of
the correspondences between concepts and resul ts in statistics and in
game theory (especi al l y inthe theory of games inextensi ve f orm) is at
present under di scussi on.
7. APPLICATIONS OF GAME THEORY
7.1When we speak of appl yi ng the theory of games, a di sti ncti on must
be made between: 1) the descri pti on, ingame theory terms, of phenomena
Game theory 223
f ound i n nature or in soci ety, and 2) real deci si ons arri ved at by games
theory met hods indeal i ng withspeci fi c, practi cal questi ons.
The second are still sporadi c and i sol ated. It i s not proposed to go i nto
detai l , but si mpl y toi ndi cate the difficulties.
First, to sol ve a practi cal probl emby game theory methods, i t is f i rst
necessary to f ormul ate i t wi th preci si on, i.e. gi ve an exact descri pti on of
the rel evant game. All the pl ayers and their strategi es must be determi ned,
and the preferenti al rel ati onshi ps over the set of si tuati ons shown. Thi s
inturn requi res accurate i nf ormati on about the phenomenon i nvol ved.
I f the preferenti al rel ati ons are determi ned by the payoff functi ons, the
i nf ormati on must be quanti tati ve. But thi s i mpl i es exact measurement s,
and the met hods for obtai ni ng t hemhave not yet been sati sfactori l y
worked out.
Secondly, the game theory sol uti on does not al ways satisfy the executor
who carri es out the deci si on. For exampl e, a mi xed strategy r ecommenda-
ti on rai ses the questi on of responsi bi l i ty in the event of unf avourabl e
consequences. Moreover, the mul ti pl i ci ty of the game s sol uti ons
(al ready di scussed) makes i t i mpossi bl e to obtai n a speci fi c r ecommenda-
tion on whi ch to rel y.
We therefore di scuss onl y the possi bl e rather than actual appl i cati ons
of game theory, i ncases inwhi ch the phenomenon i nvol ved l ends itself to
descri pti on i ngame-theoreti cal l anguage.
As i ts defi ni ti on i ndi cates, game theory i s i n pri nci pl e usabl e inconfl i cts
of any ki nd, and these can be f ound innearl y all di sci pl i nes. Accordi ngl y,
many phenomena innature and soci ety l end themsel ves to treatment in
game theory terms.
7.2 I ts ori gi nators sawgame theory mai nl y as a mathemati cal tool to
hel p inmaki ng opti mal economi c deci si ons. The ti tl e of the fi rst basi c
monogr aph by J. von Neumann and O. Morgenstern, Theory of Games
und Economic Behavior (1944), is fairly expl i ci t. In a competi ti ve economy,
game theory coul d be used instudyi ng competi ti on for markets, adver-
ti si ng, pri ce questi ons, stock market operati ons, and so on.
The theory has proved to have much broader possi bi l i ti es. It can
equal l y serve i n the condi ti ons of a pl anned economy, i n studyi ng for
exampl e the respecti ve meri ts of central i zi ng or decentral i zi ng i ndustri al
management , opti mal pl anni ng interni s of several i ndi ces, pl anni ng under
uncertai nty ari si ng, for i nstance, f romtechnol ogi cal progress, and so on.
224 Nicolai N. Vorobyev
7. 3 War i s of course the most obvi ous of all cases of confl i ct. Mi l i tary
appl i cati ons of game theory are deal t with in Drescher s monogr aph
(1961) and a mul ti tude of articles. I n 1964, NATO hel d a conference
devoted mai nl y to game theory; i ts proceedi ngs consti tute a bul ky
vol ume (1966). However , i t must be emphasi zed that games theory al so
produces mathemati cal model s for di sarmament studi es. Saaty s book
(1968), despi te some debatabl e proposi ti ons, i ndi cates some very
promi si ng devel opments,
7. 4 The l awi s another scene of evi dent confl i cts - fi rst between those who
make up a soci ety. But the maki ng of the l awis i tsel f based on a confl i ct;
for the l egi sl ator i s at gri ps wi th the factors, not al ways very cl ear or
speci fi c but neverthel ess very real , whi ch gi ve ri se to cr i me or breaches
of civil l aw. Legi sl ati on to anti ci pate mass devel opments that have
j uri di cal consequences and accordi ngl y may need to be i ncorporated i n
l awal so falls wi thi n the purvi ewof game theory. Court trials are com-
peti ti ve, i nvol vi ng opti mal behavi our and nor ms of procedure. Verdi cts
are ordi nari l y made in condi ti ons of uncertai nty; and l aw- makers and
courts al i ke can be consi dered as pl ayers maki ng deci si ons.
7.5 Ethi cal and l egal nor ms havi ng much in common, appl i cati ons of
game theory to moral questi ons are very si mi l ar to those inj uri sprudence.
The fact that, incontrast to l aw, ethi cs is not codi fi ed onl y i ncreases the
uncertai nty i nthe choi ce of opti mal ethi cal behavi our.
7. 6 Soci ol ogy has a vari ety of openi ngs for game theory - soci al and cl ass
confl i cts, the uti l i zati on of resources (parti cul arl y those whi ch cannot be
descri bed inpurel y economi c terms), rel ati ons of the i ndi vi dual withthe
col l ecti vi ty and wi th other i ndi vi dual s, and so on,
7. 7 Appl i cati ons in bi ol ogy are rather speci al . The ani mal and, even more,
the vegetabl e ki ngdoms can hardl y be sai d to take del i berate deci si ons
wi th a preci se end inview, or tobe consci ous of i nterests that confl i ct or
do not confl i ct. Neverthel ess, the i nterpretati on ingame theory t er ms of
such phenomena is both possi bl e and useful . For exampl e, the struggl e
for exi stence between t wo speci es can be treated as a t wo- person game i n
whi ch the admi ssi bl e f orms of behavi our of ei ther speci es fi gure as thei r
strategi es and the number of i ndi vi dual s i n each as thei r obj ecti ve
Gume theory 225
functi ons. It is parti cul arl y worth whi l e to treat questi ons of symbi osi s
and parasi ti smingame theory t erms.
Another natural subj ect for game theory treatment is the study of
evol uti onary processes, whi ch may be consi dered as confl i cts between
the speci es compl i cated by r andomexternal condi ti ons, wi th the l i ne of
evol uti on of a speci es as i ts strategy and the pri nci pal evol uti onary trends
in speci es as thei r obj ecti ve functi ons. Si nce optimal strategi es are or-
di nari l y mi xed, i t i s natural to assume that i t is the mutati ons that
represent the mi xed strategi es in thi s case. The descri pti on of these
phenomena in t er ms of game theory may l ead to new di scoveri es i n
heredi ty and inthe shapi ng of the geneti c i nheri tance.
REFERENCES
BERGE, C., (1957) Thoriegnrale des jeux ir npersonnes, Pari s, Gauthi er- Vi l l ars.
CHERNOFF, H. and MOSES, L. E. (1959) Elementary Decision Theory, New Yor k, J ohn
DAVIS, M. D., (1970) Game Theory, New Yor k, London, Basi c Books.
DRESCHER, M., (3961) Games of Strategy. Theory and Applications, Engl ewood Cliffs
KLAUS, G. , (1968) Spieltheorie in philosophischer Sicht, Berl i n, Deut scher Verl ag der
LUCE, R. D. and RAIFFA, H. , (1957) Games and Decisions. Introduction and Critical
MENSCH, A. (ed.) (1966) Theory of Games. Techniques and Applications. A Nato Confe-
MILNOR, (1964) in G a m e Theory and Related Approaches to Social Behavior, M. Shubi k
NEUMANN, J. VON, (1928) Zur Theori e der Gesel l schaftsspi el e , Math. Annalen, 100,
NEUMANN, J. VON and MORGENSTERN, O., (1944) Theory of Games and Economic
SAATY, T. L., (1968) Mathematical Models of Arm Control and Disarmament,
VOROBYEV, N. N,, (1970) Moder n State of Game Theory (in Russi an), Uspehi Mat.
WALD, A. (1950) Statistical Decision Functions, New Yor k, J ohn Wiley.
Wi l ey.
(N. J.), Prenti ce-Hal l .
Wi ssenschaf ten.
Survey, New Yor k, J ohn Wi l ey.
rence, New York, Amer i can El sevi er Publ i shi ng Company.
(ed.), New Yor k, J ohn Wi l ey.
295-320.
Behavior, Pri nceton (N.J.), Pri nceton Uni v. Press.
New Yor k, J ohn Wi l ey.
Nauk, Vol. XXV, 2.
12 Metatheory
MARIO BUNGE
INTRODUCTION
Stri ctl y speaki ng, a metatheory i s a theory about some theory or cl ass of
theori es. And a theory or deducti ve systemi s, inturn, a set of statements
[e.g. equati ons) ordered by the rel ati on of deduci bi l i ty. Jn thi s way no
statement in the theory remai ns i sol ated: every statement i s ei ther an
assumpti on or a concl usi on. Mor e preci sel y, a theory i s a hypotheti co-
deducti ve system, as i t can be f ormul ated insuch a way that every f ormul a
i ni t is ei ther an initial premi se [hypothesi s, axi omor datum) or a l ogi cal
consequence of a set of i ni ti al assumpti ons. Stri ctl y speaki ng, for a body
of theory to qual i fy as a metatheory i t has got to be a theory i tsel f: a
mor e or l ess l oose set of remarks on a theory does not consti tute a
metatheory proper.
Metatheoreti cal research started in mathemati cs at the turn of the
century wi th the central ai mof strengtheni ng the ri gour of proof s and
ensuri ng consi stency. Met amat hemat i cs was accordi ngl y often cal l ed
proof theory . It soon became apparent, t hough, that metamathemati cal
i nvesti gati on is i nteresti ng ini tsel f and, moreover, that i t can be useful
i n other ways. I ndeed, i t can al so l ead to the di scovery of structural
si mi l ari ti es among otherwi se di fferent mathemati cal theori es and i t can
establ i sh rel ati ons (or, mor e preci sel y, mappi ngs) among whol e sets of
theori es, thus throwi ng newl i ghts on the archi tecture of mathemati cs.
The same can be sai d of l ogi c, parti cul arl y when vi ewed as a branch of
mathemati cs and handl ed wi th the hel p of mathemati cal tool s. Metal ogi c
and metamathemati cs are inany event fl ouri shi ng. They can safel y be
i gnored by the appl i ed l ogi ci an and mathemati ci an, but they are no
l onger negl ected by the i nventors of new basi c theori es and by those
who are i nterested i nthe f oundati ons and inthe structure of l ogi cal and
mathemati cal theori es.
228 Mario Bunge
Outsi de l ogi c and mathemati cs, metatheoreti cal progress has been
sl ow- probabl y for t wo reasons. The fi rst is the bel i ef that a factual
(empi ri cal ) theory i s, after all, an i nducti ve synthesi s of empi ri cal data,
hence somethi ng based on (rather t han tested by) experi ence, i tsel f a
non-theoreti cal i tem; on this vi ew, the f oundati ons of theoreti cal sci ence
shoul d be sought inempi ri cal procedures, not i nhypotheses, l et al one in
f undamental hypotheses, i.e. postul ates. Secondl y, there is the pragmati c
atti tude of most contemporary sci enti sts, who are i na hurry to do some
hard and publ i shabl e wor k even at the pri ce of conceptual obscuri ty.
I t is hardl y necessary to poi nt out that thi s pol i cy of negl ect of, or even
cont empt for, the conceptual f oundati ons of factual sci ence i s bound
to l ead to superfi ci al i ty and even to i nconsi stency - as i t di d l ead in
mathemati cs not too l ong ago.
Be that as i t may, there are hardl y any metatheori es proper concerni ng
factual theori es: fi el d, though enormous, is sti l l l argel y unexpl ored. If
we were to sti ck to the stri ct meani ng of the t erm metatheory we mi ght
i gnore i t al together in the present survey. I nstead, we f ound i t mor e
i nteresti ng and useful to use a l ax defi ni ti on, one that will enabl e us to
expl ore i na prel i mi nary fashi on the vari ous aspects - f ormal , semanti c,
epi stemol ogi cal , pragmati c, and phi l osophi cal - of the metatheory of
sci enti fi c theori es. I n general , l ess attenti on will be pai d to the mor e
cl assi cal and therefore mor e tal ked about metamathemati cal achi evement
(likeHi l berts, Gdel s and Tarski s) than to some promi si ng new l i nes of
research i nthe metatheory of mathemati cal and sci enti fi c theori es. The
succi nct bi bl i ography at the end shoul d ai d the reader anxi ous for
further i nf ormati on.
1. LANGUAGE AND THEORY
Every theory js expressed or f ormul ated in some artificial or symbol i c
l anguage, whi ch i s often a combi nati on (or, mor e preci sel y, the uni on)
of several l anguages. A l anguage is here taken to be a set of si gns that can
be grouped i nto expressi ons whi ch can inturn be transf ormed i nto other
expressi ons. Admi ssi bl e expressi ons in a gi ven l anguage are cal l ed i ts
wel l - f ormed f ormul ae, whi l e admi ssi bl e transformati ons of expressi ons
are cal l ed the val i d transformati ons. I n a formal i zed l anguage both the
rul es of f ormati on of admi ssi bl e expressi ons and the rul es of val i dtrans-
f ormati ons must be stated expl i ci tl y. Whet her the expressi ons say
Metatheory 229
anythi ng about real i ty, and whether the transf ormed expressi ons say
anythi ng new, are questi ons that go beyond the limits of a f ormal
characteri zati on of l anguages. For our purposes, what matters i s that an
artificial l anguage of some sort or other i s i ndi spensabl e in order to
f ormul ate a theory proper: the natural l anguages are too poor and
i nexact for thi s purpose - so much so that even one of the si mpl est
mathemati cal theori es, the theory of semi - groups, cannot be f ormul ated
wi thout the assi stance of el ementary l ogi c and of a handf ul of extral ogi cal
techni cal noti ons (set, bi nary operati on on a set, and associ ati vi ty).
Theori es are often referred to as l anguages, but thi s is mi staken on
several counts. Fi rst, whi l e some theori es are true, l anguages cannot be
cal l ed fal se even t hough they may be exact or i nexact, ri ch or poor,
abstract or concrete, and so on. Second, a si ngl e l anguage may be used
by a number of theori es and, conversel y, some theori es can be expressed
or f ormul ated in al ternati ve l anguages. Thi rd, every l anguage contai ns
wel l - f ormed expressi ons that are i nadmi ssi bl e as f ormul ae of some theory
- e.g. 1 # 1 can certai nl y be sai di nthe l anguage of ari thmeti c but i t is
not a val i d f ormul a of thi s theory. Consequentl y we must di sti ngui sh
between a theory and i ts l anguage. Whi l e a mathemati cal (bi ol ogi cal )
theory concerns mathematical(biologica1) obj ects such as sets (organi sms),
i ts l anguage concerns nothi ng at all: i t i s non-referenti al , i t is j ust a
f ramework for bui l di ng statements concerni ng somethi ng.
There are uni versal l anguages and regi onal ones. The f ormer are the
l anguages of the l ogi cal theori es: all others are non- uni versal , as they
contai n speci fi c t er ms such as set or atom . I ndeed, every f ormul a,
whether intheoreti cal chemi stry or i n mathemati cal soci ol ogy, fi ts some
pre-exi sti ng f ormul a of l ogi c: mor e preci sel y, i t has the f or mof some
f ormul a of the predi cate cal cul us wi th i denti ty, and this qui te apart
f r omi ts meani ng and of i ts truth val ue. That i s, the l anguage of l ogi c i s a
ready- made, exact and uni versal l anguage. Set theory, too, has a ready-
made, exact and al most uni versal l anguage - but not qui te so uni versal ,
Every other theory has a l anguage of i ts own that can often be used by
several other theori es as well. Thus, the l anguage of physi cs i s the uni on
of the l anguages of predi cate l ogi c, set theory, and many others. Thi s
does not mean that physi ci sts expl i ci tl y empl oy many t heorems of l ogi c
or of set theory: i t does mean that they use, in some way or other, the
concepts (though not al ways the correspondi ng terms) of statement and of
deducti on, of cl ass membershi p and of set i ntersecti on, and so on.
230 Mario Bunge
A non-l ogi cal theory is not one that is i l l ogi cal but one that goes beyond
l ogi c i n that i t contai ns at l east one extra-l ogi cal assumpti on, i.e. a
statement concerni ng some extra-l ogi cal concept, such as the concept of
f uncti on or the concept of soci ety. Accordi ngl y, the basi c assumpti ons
(axi oms) of a theory are the assumpti ons of i ts underl yi ng l ogi c - i.e. of
the l ogi cal theory i t presupposes - pl us speci fi c initial assumpti ons (that
depend on the subj ect matter). Every other statement i na wel l -organi zed
theory will fol l owf r omthose basi c assumpti ons in accordance wi th
the rul es of i nference admi tted inthe underl yi ng l ogi c.
2. METALANGUAGE A N D METATHEORY
It came as a shock when Gdel proved that any domai n of some com-
pl exi ty, e.g. ordi nary ari thmeti c, contai ns true statements that cannot be
proved wi thi n i ts own f ormal i zed theory, i.e. by resorti ng excl usi vel y to
the axi oms of the theory and the rul es of i nference of the underl yi ng l ogi c.
Ei ther such statements must be justifiedby other means, or el se the theory
must be enri ched wi th further assumpti ons that make the deri vati on
possi bl e - but even wi th the extensi on, other true statements will sti l l be
excl uded. Gdel s i ncompl eteness theoremi s a standard exampl e of a
met at heoremin metamathemati cs, i.e. of a provabl e statement that
concerns (is about, refers to) mathemati cal statements rather than
mathemati cal obj ects such as sets, maps or spaces. I n general : a
metatheorem i s a provabl e metastatement and, inturn, a metastatement
is a statement about another statement or statements.
[ The fol l owi ng, based on the noti on of predi cate, is a mor e preci se
el uci dati on of the noti ons of statement and metastatement. A predi cate P
may be regarded as a f uncti on that maps n-tupl es of obj ects of any ki nd
i nto statements, i.e. P : 0% --f S, where 0% desi gnates the nth cartesi an
power of a set O of obj ects and S desi gnates a set of statements. Li kewi se
a metapredi cate MP, l i ke provabl e or compl ete , may be anal yzed as a
mappi ng of statements i nto metastatements, i.e. MP : S --f MS, where
MS is i ncl uded in S. These parti al el uci dati ons bri ng out the reference
cl ass and they are i ndependent of the concept of truth - whi ch is an
advantage i nthe case of factual sci ence, where total truth is so hard to
attai n.]
Every theory contai ns metatheorems. The earl i est exampl es of met a-
Metatheory 23 1
theorems wer e the dual i ty pri nci pl es i nproj ecti ve geometry, ensuri ng the
i nterchangeabi l i ty of poi nt and l i ne, or of l i ne and pl ane , in the
statement of the theorems. Si mi l ar dual i ty pri nci pl es wer e l ater di scovered
i nl ogi c. Moreover, the i nference rul es may be regarded as metatheorems
rather than as pragmati c i t ems or rul es of procedure. In general , every
provabl e statement of the f or m A entai l s B, where A and B are state-
ments, is a metatheorem. Li kewi se every provabl e statement of the f or m
Thas the property E, where Tis a set of statements (e.g. a theory), is a
metatheorem.
Even factual sci ence contai ns metatheorems, and not onl y those in
the f ormal (l ogi cal and mathemati cal ) theori es i t presupposes. Any
statement to the effect that a gi ven f ormul a is (or fails to be) co- vari ant
under a certai n group of transformati ons i s a metatheorem. The f amous
CPT t heorem(charge-pari ty-ti me) i n physi cs i s a metatheoremof thi s
ki nd, as i t asserts that a certai n set of f ormul ae (vi z., those of all rel ati vi sti c
fieldtheori es) retai n thei r f or munder combi ned charge, pari ty and t i me
reversal s. Far f romrepresenti ng a l awof nature, thi s is a typi cal illus-
trati on of the concept of metanomological statement. The poi nt i s i m-
portant not onl y f roma l ogi cal but al so f roma practi cal poi nt of vi ew: for,
if the CPT theoremi s i ndeed a metatheorem, then i t cannot be expected
that i t will be conf i rmed or refuted inthe l aboratory. Only the theory i t
concerns can be so tested. If conf i rmed, the CPT metatheoremstands;
if refuted, i t becomes poi ntl ess.
The col l ecti on of metatheorems of a gi ven theory need not i tsel f
consti tute a theory, i.e. a systemof f ormul ae cl osed under deducti on. Thus
the CPT theoremi s not part of a comprehensi ve metatheory of physi cs,
but a stray metatheoreti cal statement. Whet her i t mi ght be possi bl e to
bui l d a metatheory for a whol e cl ass of physi cal theori es is another
matter - one, i nci dental l y, that has not been expl ored. As poi nted out in
the i ntroducti on, i t is onl y wi th reference to l ogi cal and mathemati cal
theori es that we fi nd fully devel oped metatheori es.
Whet her stray, or systemati c (bel ongi ng to a deducti ve system), a
metatheoremis al ways expressed ina metalanguage, i.e. a l anguage used
total k about another l anguage (al anguage enabl i ng us to tal k about the
obj ects referred to by a theory is cal l ed an object language of the theory).
A metal anguage may or may not be formal l y as well devel oped as the
pri mary or obj ect l anguage; i ts l evel of devel opment will depend
pri mari l y on our needs. To state a stray met at heoremi t i s accordi ngl y
232 Mario Bunge
often suffi ci ent to empl oy ordi nary l anguage, suppl emented with some
semanti cal t er ms such as desi gnates and some metal ogi cal t erms such
as proved . Ri cher metal anguages are needed onl y for ful l y devel oped
metatheori es. I n present-day metamathemati cs, a number of such sophi s-
ti cated l anguages are infact empl oyed. Thus the al gebrai c characteri za-
ti on of a deducti ve systemas a fi l ter requi res a metal anguage contai ni ng,
at the very l east, the expressi ons precedes , parti al l y ordered set and
filter. Met amat hemat i cs at present uses as i ts l anguage the uni on of the
l anguages of all the basi c theori es of mathemati cs - al gebra, category
theory, set theory, topol ogy - i naddi ti on to a number of expressi ons of
i ts own ( theorem , proof , model , deci dabl e , categori cal ). Al though
there shoul d be no conceptual obstacl e, nothi ng of the sort exi sts i n
connecti on wi thfactual theori es.
A metatheory, whether fully devel oped (stri ct sense) or not (lax sense),
i.e. whether i t is inturn a theory or merel y a set of mor e or l ess ri gorous
and l ogi cal l y connected remarks, can bear on any of the fol l owi ng
aspects of the obj ect theory (or cl ass of theori es) i t concerns: the syntax
or structure, the semanti cs or content, the pragmati cs or use, or, fi nal l y,
the phi l osophy of the theory (or set of theori es) inquesti on. Thus a met a-
theory of physi cal theory coul d deal , several l y or j oi ntl y, wi th the f ol l ow-
i ng aspects of physi cal theory: the l ogi cal structure, the mathemati aal
structure or f ormal i sm, the i nterpretati on, the empi ri cal val i dati on and
the technol ogi cal appl i cati on, and the underl yi ng metaphysi cal and
epi stemol ogi cal assumpti ons. Bef ore briefly consi deri ng these vari ous
aspects of metatheory, we must fi rst exami ne the concept of theory
formul ati on.
3. THEORY FORMULATION
All theori es i nl ogi c, mathemati cs and sci ence are mor e or l ess formal i zed.
It i s not onl y that ordi nary l anguage may l ack the techni cal terms - after
all, these can be added to any vocabul ary. But these terms usual l y are
i nterrel ated i n such a way as to consti tute techni cal expressi ons that go
beyond the resources of ordi nary l anguage. Formal i zati on provi des a
means of attai ni ng conceptual ri chness and preci si on, and of renderi ng
deducti ve i nference as ri gorous and mechani cal as possi bl e.
However, there are degrees of formal i zati on. We may di sti ngui sh the
Metatheory 233
fol l owi ng stages of devel opment of a theory i n this regard: 1) natural or
semi - f ormal i zed, 2) axiomatic, and 3) fullj) formalized. A natural or nai ve
f ormul ati on of a theory i s one in whi ch the basi c i ngredi ents are not
cl earl y si ngl ed out, and one inwhi ch the sci enti st is al l owed to i ntroduce
forei gn el ements at any stage if they are necessary to prove somethi ng. At
this stage rel evance (to the subj ect matter) and consi stency (non- contra-
di cti on) are the onl y mandatory requi rements. Most of mathemati cs
and theoreti cal sci ence fol l owthi s semi - f ormal pattern, whi ch is often
modestl y cal l ed i nf ormal or non- f ormal i zed .
The next stage, axiomatics, i s attai ned when: 1) the basi c (pri mi ti ve
or defi ni ng) concepts of the theory are f ound and l i sted; 2) the remai ni ng
concepts are defi ned i n t erms of the basi c concepts; 3) all the basi c
concepts are unambi guousl y and exhausti vel y characteri zed by means
of initial assumpti ons (postul ates or axi oms); 4) these initial l aws l ead on
to the remai ni ng statements of the theory.
Axi omati cs i s nowadays taken for granted inl ogi c, set theory, al gebra
and topol ogy, where i t has become standard practi ce to f ormul ate the
bases or f oundati ons of theori es as axi omati c defi ni ti ons, i.e. as defi ni ti ons
by means of axi oms.
Example. Tis a topol ogi cal space if and onl y if Tis an ordered coupl e
{ X, I) where Xis a set and l i s a unary operator such that:
1) if A c X, then I(A) c X,
2) I(x) = x
3) for every A c X, I(A) c A,
4) for every A c X, I(I(A)) = I(A)
5) for every A, B c X, I(A n B) = I(A) n I(B).
,
-
These five axi oms gi ve j oi ntl y an i mpl i ci t defi ni ti on of the i nteri or
operator I, and are necessary and suffi ci ent to devel op the general
theory of topol ogi cal spaces. But surel y thi s i s not how the theory was
born. Moreover, the f ormul ati on i s so abstract that i t is usel ess to anyone
who has not got the feel of i t - an i ntui ti ve understandi ng that can onl y
come by worki ng on exampl es and appl i cati ons. Neverthel ess, the axi oms
manage to convey some of the i ntui ti ons behi nd the f ormal i sm. Thus,
one can read I as an operator that peel s off a set, and I(A) is hel pful l y
cal l ed an open set. The ai mof axi omati zati on i s not to kill i ntui ti on but
to expl oi t i t to the ful l , by renderi ng expl i ci t all the taci t component s of
i ntui ti ve reasoni ng, whi l e at the same time avoi di ng bei ng mi sl ed by
i ntui ti on. In short, axi omati cs equal s expl i ci tness pl us exactness pl us order,
234 Mario Bunge
The possi bi l i ty of axi omati zi ng a theory is i ndependent of i ts content,
if any. I n pri nci pl e, i ndeed, every cl ear and wel l -kni t body of i deas can be
axi omati zed. Even theori es i n pol i ti cal sci ence can be axi omati zed
provi ded they are fi rst mathemati zed. However , axi omati zati on can be
understood ei ther restri cti vel y or broadl y. I n the f ormer axi omati zati on
means not onl y the orderl y arrangement of all the component s (basi c
concepts and initial assumpti ons) but al so the extracti on of all non-
mathemati cal content, i.e. the conversi on of the gi ven theory i nto a
purel y f ormal structure. Thus if mechani cs is axi omati zed, forces will be
descri bed as vectors, negl ecti ng the physi cal meani ngs. I f however the
factual content is taken i nto account, then axi omati cs can be attempted
inthe broad sense, taki ng care of meani ngs, at l east parti al l y, by addi ng
rul es of desi gnati on and assumpti ons for the referents of the symbol s in
questi on. One may thus speak of physi cal , bi ol ogi cal , or soci ol ogi cal
axi omati cs. A f ew factual theori es, mai nl y in physi cs, engi neeri ng,
chemi stry, bi ol ogy, and psychol ogy, have been f ormul ated axi omati cal l y.
The formal i zati on process can be carri ed beyond axi omati zati on by 1)
suppl ementi ng the symbol i c l anguage so as to mi ni mi ze the functi on of
ordi nary l anguage, and 2) stati ng or at l east menti oni ng all the assump-
ti ons and presupposi ti ons, i ncl udi ng the l ogi cal ones. Thi s compl ete
f ormal i zati on is si mpl y cal l ed formal i zati on , al though i t mi ght mor e
si gni fi cantl y be cal l ed full formalization. A fewel ementary mathemati cal
theori es have been ful l y formal i zed. A ful l formal i zati on of any of the
ri cher theori es is practi cal l y i mpossi bl e. Rati onal i smmust be content to
know that there are no obstacl es inpri nci pl e to a full formal i zati on of any
theory - but that i t is mor e rewardi ng toi nvent new theori es, even if they
are onl y hal f-formal i zed.
4. STRUCTURE OF THEORIES
The general characteri zati on of a theory hi nges on the metal ogi cal noti on
of deduci bi l i ty or entai l ment. But this is i nsuffi ci ent, for i t does not tel l
us anythi ng about a theory as a structured whol e. A deeper viewcal l s for
certai n mathemati cal noti ons. Three possi bl e mathemati cal treatments of
the structure of theori es ingeneral are set out bel ow.
The most el ementary way of handl i ng the concept of a theory is to
regard i t as a set S of statements parti al l y ordered by the rel ati on I- of
Metatheory 235
entai l ment [T = ( S, I- > 1. Thi s set-theoretic approach al l ows a number
of metatheoreti cal noti ons of the syntacti c ki nd to be f ormal i zed. I n
parti cul ar, the i dea of the uni on of t wo (axi omati zabl e) theori es can be
cl ari fi ed. I f TI and Tz are theori es expressed i n certai n l anguages L(T1)
and L(T)z respecti vel y, then thei r union T = TI u TZ i s a thi rd theory T
such that i ts l anguage i s the uni on of the gi ven l anguages and the axi oms
of Tar e those of TI j oi ntl y wi th those of T:!. Li kewi se for the intersection
T = TI n TZ of t wo theori es. Fi nal l y, TI may be sai d to be a subtheory
of T2 if the set SI of f ormul ae of TI i s i ncl uded inthe set SZ of f ormul ae
of T2 and, inaddi ti on, TI is cl osed under deducti on. I f TI is a subtheory
of T2, then Tz i s cal l edan extension of TI.
The set S of f ormul ae of a theory is the col l ecti on of all the l ogi cal
consequences of i ts initial assumpti ons (axi oms) A, i.e. S = Cn (A). The
cl osure of S under deducti on i s expressed i n the metatheorem: S =
Cn (5). Not every set of f ormul ae has this property. Thus the uni ons1 u S,
of the f ormul ae of t wo total l y unrel ated theori es may not consti tute a
thi rd theory. The consequences of SI u S:! consti tute what is cal l ed the
l ogi cal sum SI 4 SZ of the t wo systems. A typi cal met at heoremi nvol vi ng
this new noti on i s the fol l owi ng: If A1 and A:! are t wo f ormul ae (e.g.
axi oms) of a gi ven theory, then the set of consequences of thei r conj unc-
ti on is the l ogi cal sumof thei r separate consequences. On the other hand,
the set of consequences of the di sj uncti on of Al and A:! equal s the common
part of thei r separate consequences. I n other words, conj uncti on en-
ri ches whi l e di sj uncti on i mpoveri shes. So much for the cal cul us of
systems.
A di fferent characteri zati on of a theory is gi ven in terms of the al gebrai c
concept of filter. Afilter I; ina parti al l y ordered set P is a structure such
that: 1) if x is a member of F and x precedes y, then y too i s inF; 2) the
i nf i mum(greatest l ower bound) of any t wo member s of I; is i nF, and 3)
the empt y set is not i ncl uded inF. On i nterpreti ng precedes as entai l s
and i nf i mum as conj uncti on , i t i s possi bl e to prove the central
metatheorem: every deductive systemis afilter. Fr omhere on the study
of the structure of theori es becomes the study of a parti cul ar i nter-
pretati on (model ) of fi l ter theory, a branch of abstract al gebra. Whi ch i s
one mor e exampl e of the al gebrai zati on of metal ogi c and met amat hemat -
i cs and, at the same ti me, of the mathemati zati on of phi l osophy. So
much for the al gebrai c approach to the anal ysi s of theori es. Let us now
menti on an even deeper approach to the same probl em.
236 Mario Bunge
Mathemati cal theori es are usual l y f ormul ated and anal yzed in set
theoreti c t er ms, i.e. they are speci fi ed as sets together wi th certai n
rel ati ons and/ or operati ons and/ or functi ons on those sets. Thus, a
l atti ce is sai d to be a parti al l y ordered set together wi th t wo bi nary
operati ons on that set, meeti ng certai n condi ti ons (axi oms). When
constructed inthis styl e, a theory can usual l y be f ormul ated inmor e than
one way. There i s no poi nt in aski ng whi ch f ormul ati on i s the best,
except for aestheti c, di dacti c, or computati onal purposes. The mer e
mul ti pl i ci ty of equi val ent f ormul ati ons makes i nter-theory compari son
difficult and even mi sl eadi ng, as the di fferences may be mi ni mi zed in
some cases and exaggerated in others. A deep characteri zati on of a
theory, or f ami l y of theori es, shoul d be free f romsuch ambi gui ti es: i t
shoul d sei ze on what i s essenti al to the theory beneath the acci dents of
i ts parti cul ar presentati ons, whi ch are often a matter of taste. If this is
so, one may suspect that any set theoreti c characteri zati on of a theory
i s marri ed to a parti cul ar presentati on of a deeper structure that remai ns
beyond the reach of that mode of theory bui l di ng but mi ght be di scl osed
by a deeper l ook. Thi s i s i ndeed the case: there now exi sts an al ternati ve
way of f ormul ati ng, hence of anal yzi ng al gebrai c theori es, and surel y
mor e compl ex theori es as wel l , whi ch is mor e abstract and therefore l ess
ambi guous. Thi s al ternati ve is offered by category theory, a new basi c
branch of mathemati cs.
The categorial (usual l y spel t ' categori cal ' ) approach consi sts in
emphasi zi ng the morphi sms (e.g. functi ons) at the expense of the el ements
of the cl asses. I n the categori al viewan al gebrai c theory, and perhaps
every theory, is a parti cul ar type of category, and an al gebra is a real i za-
ti on of that category. I n turn, a category C consi sts essenti al l y of: 1) a
cl ass ob C of obj ects (e.g. sets, groups, topol ogi cal spaces), 2) a cl ass
mor C of mor phi sms (structure preservi ng mappi ngs) f r omone obj ect to
another and 3) the composi ti on of morphi sms (a general i zati on of
functi on composi ti on), the whol e sati sfyi ng t wo axi oms:
- the associ ati vi ty l awfor morphi sms: i ff, g, h are any composabl e
morphi sms, then (5s) - h = f - (g.h);
- the i denti ty axi om: i ff i s any mor phi smf r omthe obj ect A to the
obj ect B, then IA f = f = f . IB, where IA : A -> A i s the i denti ty
mor phi smon A and IB the one on B.
Exampl es of categori es: 1) the category of sets: i ts obj ects are all sets
and i ts mor phi sms are all functi ons; 2) the category of l atti ces: the one
Metatheory 237
whose obj ects are all l atti ces and whose morphi sms are all lattice
homomor phi sms.
One can i nvesti gate rel ati ons among categori es that are not rel ated i n
obvi ous ways. I ndeed, under certai n restri cti ons, one can map whol e
categori es on other categori es. Every such mappi ng i s effected by a
functor, or mor phi smof categori es, i.e. by a functi on that takes obj ects
i nto obj ects and mor phi sms i nto morphi sms. Mor e preci sel y, gi ven t wo
categori es C and D, a functor F f romC to D is determi ned by t wo l aws:
one f r omob C to ob D, the other f r ommor C to mor Dy such that
1) morphi smcomposi ti on i s preserved, i.e. i ffand g are mor phi sms of C,
then the composi ti on F(f).F(g) of thei r i mages in D equal s the i mage
F(f.g) of thei r composi te, and 2) i denti ti es are prserved, i.e. if B is an
obj ect in C and IB the correspondi ng i denti ty morphi sm, then i ts i mage
F(Ig) i nD equal s the i denti ty morphi smIF(B) of the i mage obj ect. I n turn,
functors can be composed and, correspondi ngl y, whol e bunches of
categori es can be handl ed at one stroke. Thus the category of categories
is bui l t: i ts obj ects are all categori es and i ts morphi sms all functors. Thi s
i s the supreme mathemati cal obj ect.
Si nce in the categori a1 approach the accent i s on f or mand on the
whol e, i t is reasonabl e to expect that i t will el i ci t a deep and frui tful
i nsi ght i nto the structure of theori es and thei r mutual rel ati ons. I t may
end up by reorgani zi ng the whol e of mathemati cs and by re- shapi ng
metatheory.
5. DEGREES OF ABSTRACTION
Theori es vary inthei r degree of abstracti on. Thus the theory of parti al l y
ordered sets i s abstract as l ong as the nature of the sets and of the orderi ng
rel ati on are not speci fi ed. If the rel ati on i s i nterpreted as the i ncl usi on
rel ati on for sets but the set i s sti l l regarded as unspeci fi ed or abstract,
then a somewhat mor e concrete theory resul ts. Fi nal l y, if both the nature
of the member s of the set and the orderi ng rel ati on are fi xed, a concrete
theory, or model of the abstract theory, will resul t.
All the basi c theori es in moder n al gebra are abstract: they concern
uni nterpreted operati ons and rel ati ons on abstract sets. These theori es
often ori gi nated i n an effort to di scl ose the common essenti al structure
beneath a col l ecti on of speci fi c or concrete theori es: i n a way, then,
J
238 Mari o Bunge
those theori es emerged as a resul t of metatheoreti cal i nvesti gati ons. Thus
an abstract semi - group is a structure shared by all obj ects contai ni ng an
associ ati ve operati on. Cl earl y, such obj ects are f ound not onl y i n
mathemati cs but al so around us: for exampl e, the set of all physi cal
systems - inparti cul ar l i ngui sti c si gns - consti tutes a semi - group i n so
far as i ts i ndi vi dual s can be combi ned or concatenated associ ati vel y. An
abstract theory Tis l i ke a dummy, whi l e a model M of Tis likea set of
cl othes fittingT.
Consi der an abstract theory T formal i zed to the extent that all i ts
speci fi c or non-l ogi cal constants have been i denti fi ed and arranged ina
sequence CO, CI, CZ, ..., Cn, ... Now consi der a paral l el sequence
M = < U, SO, SI, .. . , S,, ... ) of defi ni te mathemati cal obj ects, each
wi th a defi ni te status: U is a set and every one of the Sts, for i = 1 , 2, . . . ,
i s defi ned on Ui n a way that corresponds tothe structure of the constants
in the abstract theory. Mor e preci sel y, if C, i s an i ndi vi dual constant (a
vari abl e of the l owest l ogi cal type), then the correspondi ng Sn is an
el ement of U. I f Cn is a unary predi cate, then Sn is a subset of U. I f Cn
i s a bi nary predi cate, then S, is a set of ordered pai rs of el ements of U -
and so on. Such a sequence M of speci fi c mathemati cal obj ects defi ned in
set-theoreti c t er ms i s cal l ed a possi bl e extensi onal mode (real i zati on)
of T. Whet her M is an actual model of T will depend on the f ormul ae
f or med in T wi th the hel p of i ts speci fi c constants Ct and the l ogi cal or
uni versal concepts (negati on, conj uncti on, exi stenti al quanti fi er, etc.).
A f ormul a i n T may or may not become true when every constant Ct
occurri ng in i t i s repl aced by the correspondi ng speci fi c concept St
occurri ng inM. If every val i d statement of Tis satisfied (or hol ds, or is
true) under such a speci fi cati on or i nterpretati on of the non-l ogi cal
constants of T, then Mi s sai d to be an (actual ) extensi onal model of T.
Every abstract theory has a number of model s. These model s may be
structural l y i denti cal (i somorphi c) to one another or not. If they all are,
the theory i s sai d to be categori cal ; otherwi se i t i s non-categori cal . Most
theori es are non-categori cal . I ndeed, i somorphi smi s harder to come by
than i t woul d seem. To rui n i somorphi smi t is enough to take non-
i somorphi c basi c sets ( domai ns of i ndi vi dual s). For exampl e, the theory of
parti al order may be real i zed by infinitely many pai rs < A, < > , < By 6 ) ,
... where A, By ... are sets wi th di fferent cardi nal i ti es, i.e. unequal l y
numerous. The best one can usual l y do i s to prove that the set of all
model s of a gi ven theory T i ncl udes a subset of model s that are i somor-
Metatheory 239
phi c to one another: a statement to thi s ei fect i s cal l ed a representation
(meta) theorem for T. But full categori ci ty i s both excepti onal and far
f r omdesi rabl e, as i t amount s to ri gi di ty. Onl y non-categori cal theori es
can have wi de fi el ds of appl i cati on.
The semanti c concept of model has proved to be i mportant, not onl y
for mathemati cs but al so for l ogi c. Thus, if a gi ven statement couched in
the l anguage of T is sati sfi edinevery possi bl e model M, that statement
i s sai d to be wld, or logically true, or analytic in the model -theoreti c
sense. In this sense, l ogi cal truth i s model - f ree, i.e. i t does not depend
on the parti cul ar i nterpretati on assi gned the functi onal symbol s con-
cerned. The mere fact that there are such i nterpretati on-free f ormul ae
and that they are uni versal l y val i d t hrows doubt on the thesi s that every
i dea i s concerned wi th facts. And the fact that the set of such model - f ree
f ormul ae, i.e. l ogi c, i s rati onal l y pri or to any speci al enqui ry, in parti cul ar
any i nvesti gati on concerni ng matters of fact, suffi ces to rui n al together
the doctri ne that whatever can be thought concerns the worl d.
The semanti c concept of model i s al so of i mportance inexact phi l os-
ophy, and inparti cul ar in methaphysi cs. For exampl e, i t enabl es us to
el uci date the noti on of structural i denti ty, or f ormal anal ogy, between
t wo fields of fact, such as mol ecul ar di ffusi on and human mi grati on, or
t wo concrete systems, such as organi sms and soci eti es. I ndeed, i t can be
sai d that t wo such concrete obj ects are formally analogous (structural l y
i denti cal ) if and only if they are adequatel y accounted for by t wo
structures that are i somorphi c model s of an underl yi ng abstract theory.
Anal ogy i n a weaker sense will obtai n if the morphi smconcerned i s a
homomor phi smof one of the model s i nto the other.
The semanti c concept of model di ffers f r omthe vari ous other senses of
the same word, except i n the trivial sense that every theoreti cal model ,
or speci fi c theory, of some sector of real i ty is in turn a model , i n the
semanti c sense, of some abstract theory. The theory of semanti cal model s
was adumbrated at the turn of the century i n connecti on wi th certai n
metamathemati cal questi ons, mai nl y consi stency and i ndependence, but
i t was not worked out systemati cal l y unti l the 1950 s. It i s now the most
advanced branch of semanti cs and a fl ouri shi ng area of research, but i t
remai ns practi cal l y l i mi ted to l ogi c and mathemati cs. The semanti cs of
factual theori es i s hardl y under way (see Secti on 7).
240 Mario Bunge
6. METAMATHEMATICAL PROPERTIES
A theory can have, or fail to possess, a number of gl obal properti es of
a f ormal ki nd. The study of these properti es and thei r mutual rel ati ons,
as wel l as the fi ndi ng of practi cal cri teri a for establ i shi ng whether a
gi ven theory does have any such properti es, consti tutes the very heart of
metamathematics.
Chi ef among all metamathemati cal properti es i s consistency or non-
contradi cti on. It i s mandat ory for every theory worth i ts name that i t
be consi stent or that i t can be repai red so as to become consi stent. Thi s,
not onl y because a bl atant fal si ty is a bl emi sh, but al so because a si ngl e
contradi cti on (l ogi cal l y fal se statement) will generate an arbi trary
number of arbi trary statements, i.e. i t will i nduce i rrel evant di gressi on.
The usual consi stency test is based on the fol l owi ng metatheorem: A
theory is consistent i f and only if it possesses a model. The actual test i s as
fol l ows. First, axi omati ze the theory. Next , stri p the axi oms of thei r
usual meani ng, l eavi ng an uni nterpreted (abstract) f ormal i sm. Fi nal l y,
try to fi nd a model of thi s abstract theory in a wel l - known theory the
consi stency of whi ch is ei ther establ i shed or assumed: that i s, try to fi nd
an i nterpretati on of the basi c symbol s under whi ch the f ormul ae of the
theory become f ormul ae of a fami l i ar theory whose consi stency is not
questi oned (at the moment). If such a model is f ound, then the theory can
be pronounced consi stent: for, if i t di d hi de a contradi cti on, the l atter
woul d show up unmi stakabl y inthe new i nterpretati on.
Noti ce that thi s procedure does not yi el d absol ute but rather relative
consistency, inthe sense that the gi ven theory i s pronounced consi stent
rel ati ve to some other theory, i.e. provi ded the l atter i s consi stent. There
woul d be nothi ng wrong wi th thi s if the base-l i ne i tsel f weredemonstrabl y
consi stent, but this is usual l y not the case. I ndeed, mor e often than not in
a consi stency proof, the symbol s inthe gi ven theory are gi ven number -
theoreti c meani ngs - but i t so happens that number theory has not been
proved consi stent. Not e also that the consi stency test we have menti oned
i s a semanti cal one: i t empl oys a semanti c means for achi evi ng a f ormal
(syntacti cal ) end. The same hol ds for most other metamathemati cal
properti es: ei ther purel y syntacti c (model -free) met hods are not avai l abl e,
or they are l abori ous.
A second metamathemati cal property of i mportance i s postulate
independence. It i s desi rabl e, though not necessary, that the axi oms of
Metatheory 241
a theory be mutual l y i ndependent, i.e. that none of thembe entai l ed by
the others. Thi s is not j ust a matter of el egance and economy but al so one
of practi cal i mportance. I ndeed, if one wi shes to subj ect a theory to
repai rs, one has to proceed step by step, and this i s possi bl e onl y if the
repl acement of any gi ven axi omby another f ormul a does not force changes
inall the others: au i ndependent axi omsystemi s one that does not rej ect
transpl ants.
Axi omi ndependence can be tested inthe f ol l owi ng way. Take all the
postul ates and si ngl e out one of t hem. Negate the l atter and adj oi n i t to
the remai ni ng postul ates. Then fi nd out whether the newaxi omsystem
is consi stent. I f i t i s, then the postul ate under scruti ni ty i s i ndependent -
so much so that its negati on has produced no contradi cti on. The same
procedure is then appl i ed to all the remai ni ng postul ates, one at a ti me.
If a non- i ndependent is spotted, i t is removed f r omthe postul ate set: a
theoremhas been won. I n the last resort thi s techni que, too, is of a
semanti c nature, as i t boi l s down to a number of consi stency tests. There
is an al ternati ve i ndependence test whi ch i s semanti c all the way t hrough,
as i t consi sts in rei nterpreti ng the basi c (pri mi ti ve) symbol s so that
all the postul ates except the suspect one are sati sfi ed. Lf such a model
is f ound then the candi date must be i ndependent: if i t were dependent
on (entai l ed by) some of the remai ni ng axi oms it, too, woul d be sati sfi ed
under the gi ven i nterpretati on.
A thi rd val uabl e metamathemati cal property is primitive indepen-
dence. It is desi rabl e but not mandat ory that the basi c (pri mi ti ve)
concepts of the theory be mutual l y i ndependent, i.e. that they be not
i nterdefi nabl e. Thi s i s mai nl y of phi l osophi cal i mportance: i t al l ows one
to spot defi ni ti ons and avoi d conf usi ng t hemwi th non- conventi onal
f ormul ae, e.g. l awstatements. The test i s agai n of a semanti c nature and
i s as fol l ows. First col l ect all the pri mi ti ves and then re-i nterpret the one
under exami nati on. Then check whether the postul ates still hol d after thi s
parti al re-i nterpretati on: if they do, the candi date has passed the test.
A f ourthmetamathemati cal property is completeness. A theory Ti s com-
pl ete if and onl y if every f ormul a couched in the l anguage of Tis ei ther
provabl e or refutabl e in T. Cl earl y, compl eteness amount s to i ncapaci ty
to growwhi l e remai ni ng consi stent. Test: take the postul ate set and
consi der a newpremi se couched inthe l anguage of the theory. There are
t wo possi bi l i ti es: ei ther the new premi se is entai l ed by ( contai ned in)
the axi oms, in whi ch case the latter are pronounced compl ete; or the
242 Mario Bunge
axi omset enri ched by the new premi se i s i nconsi stent, inwhi ch case, too,
the ori gi nal theory is compl ete. A di fferent test, of a semanti c nature, is
based on the metatheorem: Ifa theory iscategorical then it is complete.
One of Gdel s shatteri ng di scoveri es was that, if a recursi vel y
axi omati zabl e theory is moderatel y ri ch (e.g. if i t contai ns el ementary
ari thmeti c) and moreover consi stent (demonstrabl y so or by assumpti on),
then i t is i ncompl ete. And, not bei ng compl ete, i t can be extended wi thout
necessari l y becomi ng i nconsi stent. Thi s has someti mes been vi ewed as a
l i mi tati on on rati onal i ty. It may be construed inthe opposi te sense, by
j ust repl aci ng i ncompl ete by i ts synonymextensi bl e. I ndeed, if prac-
tically all theori es are i ncompl ete, then there is no end to the task of
bui l di ng stronger and stronger theori es, and appl yi ng them. Thi s is
parti cul arl y obvi ous in the case of factual theori es: i ndeed, in factual
sci ence, theori es must be abl e to gr owby the addi ti on of subsi di ary
hypotheses and data: i.e. they must be i ncompl ete. Otherwi se they cannot
be general , or they cannot be appl i ed to speci al cases nor, consequentl y,
subj ected to empi ri cal tests. I rrati onal i smshoul d not rej oi ce in the
essenti al i ncompl eteness of al most every theory: whati s i ncompl ete makes
further rati onal i nvesti gati on possi bl e and desi rabl e.
The fifth and l ast metamathemati cal characteri sti c we shal l touch on
i s decidability. A theory T is cal l ed deci dabl e if a deci si on procedure
for i t exi sts. And a decision procedure for T i s a general and effecti ve
(mechani cal ) met hod al l owi ng one (or a computer) to deci de whether any
gi ven f ormul a couched inthe l anguage of Tis i ndeed a bonajide member
of Tor j ust a stranger dressed up inthe same l i ngui sti c garb. Deci dabi l i ty,
i.e. the exi stence of such an effecti ve procedure, can often be proved
wi thout actual l y exhi bi ti ng the techni que. As i t happens, the vast
maj ori ty of mathemati cal (hence scientific) theori es are undeci dabl e. Not
even the el ementary theory of groups i s deci dabl e - l et al one ari thmeti c
and the theori es presupposi ng any such undeci dabl e theori es. Moreover,
i t can be proved that, if a (fi rst order) theory is deducti vel y compl ete,
then i t i s undeci dabl e if and onl y if i t i s not recursi vel y axi omati zabl e.
Thi s met at heoremis far l ess catastrophi c than i t l ooks. I ndeed, the com-
pl eteness cl ause i s sel domsati sfi ed, so the metatheoremdoes not al ways
appl y. Hence i t consti tutes no real l i mi tati on on the possi bi l i ty of
axi omati zi ng any branch of knowl edge.
Moreover, the non-exi stence of a deci si on procedure does not precl ude
admi tti ng or rej ecti ng any parti cul ar f ormul a as a member of a gi ven
Metatheory 243
theory T, by usi ng means obtai ned outsi de T, e. g. l emmas not proved
inT. Thi s i s, after all, the way proof s and counter- exampl es are usual l y
constructed in non- f ormal i zed mathemati cs, namel y, with the hel p of
auxi l i ary constructi ons and noti ons borrowed f romother chapters of
mathemati cs. Surel y by proceedi ng inthis way the crust of a f ormal i zed
theory i s pi erced, but everythi ng proceeds wi thi n the sel f -i mposed bounds
of reason. Reason cannot be confi ned wi thi n one theory or even wi thi n
one arbi trari l y demarcated field. Every f ormal i smi s artificially separated
f r omall the other theori es. I nventi ve reason puts together what f ormal -
i zi ng reason puts asunder.
7. THE SEMANTICS OF SCIENTIFIC THEORIES
Model theory, touched on in Secti on 5 and appl i ed in Secti on 6, has
become the bul k of the semanti cs of l ogi c and mathemati cs, as well as
an i mportant tool for provi ng theorems, or rather metatheorems, con-
cerni ng certai n aspects of any theory as a whol e.
But model theory cannot, wi thout further ado, be appl i ed to a se-
manti c anal ysi s of factual theori es, i.e. theori es witha factual reference;
and thi s for t wo mai n reasons: because bot h the concept of meani ng and
t he concept of truth suffer radi cal changes when enteri ng the terri tory of
factual (usual l y cal l ed ' empi ri cal ' ) sci ence. Thi s, i n turn, shoul d be
obvi ous: unl i ke pure mathemati cs, factual sci ence concerns concrete
thi ngs and is tested by checki ng i ts perf ormance in rel ati on to the
observabl e behavi our of concrete thi ngs.
Consi der fi rst the probl emof i nterpretati on, whi ch i s i nti matel y
rel ated to the one of meani ng. An i nterpreted mathemati cal theory (i.e.
a model of some abstract theory) is a structure i nterpreted i n some other
mathemati cal theory. For exampl e, ri ng theory can be i nterpreted i n
ari thmeti c terms. Her e i nterpretati on remai ns wi thi n mathemati cs and i t
consi sts inassi gni ng a speci fi c ari thmeti cal obj ect ( number or operati on)
to every pri mi ti ve symbol of the uni nterpreted theory. By contrast, a
theory wi th a factual content (reference) i s i nterpreted by ref erence to
extra-conceptual obj ects such as physi cal thi ngs, thei r states, and changes
inthese states. I n short, whereas a mathemati cal model i s yet another
f ormal i sm, a sci enti fi c theory is a f ormal i smendowed wi th an i nterpre-
tati on poi nti ng beyond the conceptual domai n. Two di fferent concepts of
i nterpretati on are then i nvol ved.
244 Mario Bunge
The di fference can be el uci dated roughl y i n the fol l owi ng way. An
i nterpretati on ensui ng in a mathemati cal model may be construed as a
mappi ng Im on the set S of symbol s of a theory i nto a set C of speci fi c
mathemati cal concepts. I n short, Im: S -+C. If we wi sh to produce a
factual theory we must assume that, in turn, every one of the basi c
concepts in C ref ers to some concrete obj ect (thi ng, property thereof,
event, etc.). That i s, we must i ntroduce a further map R, one assi gni ng
every member of C a set inthe fami l y P of cl asses of concrete obj ects. I n
bri ef, R: C + P. Fi nal l y we compound the t wo maps, R and Im, obtai ni ng
the map cal l ed factual interpretation, namel y If = lUm. The three
arrows consti tute a commut at i ve di agram. Thi s al l ows us to go di rectl y
f r omthe set S of basi c symbol s tothe fami l y P of referents of the theory,
i.e. If$ : S - +P. (The el uci dati on of the noti on of empi ri cal i nterpretati on
is even mor e compl ex, for i t i nvol ves further theori es as well as a reference
to facts as experi enced by some subj ect.)
Once the noti on of factual i nterpretati on has been el uci dated, one can
expl i cate the one of factual (e.g. physi cal ) meaning. For exampl e, one may
regard the meani ng of a symbol occurri ng ina factual theory as the sense
(connotati on) of the correspondi ng concept together wi th the reference
cl ass (set of i ndended referents) of the l atter. Thi s poses, of course, the
probl emof bui l di ng a cal cul us of senses to el uci date the el usi ve noti on of
connotati on: thi s i s di ffi cul t though possi bl e. But the probl emwill not
even be tackl ed if one bel i eves that current model theory fulfils the needs
of factual sci ence, or if one confuses factual meani ng with empi ri cal
meani ng or, worse, wi th empi ri cal test operati ons.
Fi nal l y, once the concept of factual meani ng has been el uci dated, one
can approach the task of bui l di ng a theory of factual truth or materi al
adequacy. Such a theory goes beyond the si mpl e concept of truth
empl oyed i n l ogi c and mathemati cs; 1) because factual truth i nvol ves
factual ref erence, and 2) because factual truth is rarel y total or compl ete:
in the best of cases, if i nteresti ng and deep i t i s approxi mate: onl y
trivialities can be squarel y true or fal se. Now, si nce a concept is best
el uci dated by a theory that contai ns it, we need an exact theory of inexact
factual truth. Onl y thus will the semanti cs of sci ence be real i sti c and hence
in a posi ti on to hel p sci enti sts, who every day empl oy the concept of
parti al truth.
So much for the outl i ne of a research proj ect in the semanti cs of sci ence.
Metatheory 245
8. THE EPISTEMOLOGY OF SCIENTIFIC THEORIES
Unl i ke mathemati cal theori es, sci enti fi c theori es are supposed to enri ch
our knowl edge of real i ty. Hence they must have certai n epi stemol ogi cal
properti es i naddi ti on to thei r syntacti cal and semanti cal characteri sti cs.
A fi rst epi stemol ogi cal condi ti on every scientific theory i s requi red to
meet i s external consistency or compati bi l i ty wi th the bul k of corroborated
data, hypotheses, and theori es. No matter how ori gi nal a theory, i t must
respect l ogi c and mathemati cs - not just by l eavi ng t hemal one but by
usi ng t hem- and i t must not questi on every itemof factual knowl edge.
For, if i t does cl ash withthe whol e of tradi ti on, then i t will be i mpossi bl e
to put i t tothe test, as every test of a gi ven theory cal l s for the col l abora-
ti on of further theori es, hypotheses, and data. (For exampl e, the test of
a psychol ogi cal theory requi res i tems of physi cs, somet i mes al so of
chemi stry and bi ol ogy, al ways some l ogi c and mathemati cal statistics.) The
surest si gn of pseudo- sci ence i s that i t makes a cl ean break wi th the whol e
of our f ormal and sci enti fi c background, thus rej ecti ng the normal mode
of testi ngfor truth.
For a theory to be testabl e, all of i ts predi cates must be scrutable: t h y
must be open to critical anal ysi s and they must somehow, however
devi ousl y, show up i nobservabl e facts. For exampl e, the phase of a wave
is scrutabl e (though not by di rect observati on) whereas a repressed
wi sh, one that does not mani fest itself, i s i nscrutabl e. The requi rement
that we keep onl y measurabl e predi cates (whi ch is somet i mes made inthe
name of an obsol ete vari ety of empi ri ci sm), woul d kill every hypothesi s,
hence every theory as well, for the i nark of a hypothesi s i s that i t goes
beyond experi ence. It is not di rect observabi l i ty that we must have but
scrutabi l i ty: thi s is necessary and suaci ent for empi ri cal testabi l i ty.
A speci fi c sci enti fi c theory shoul d be testabl e by observati on, measure-
ment, or experi ment: i t shoul d stick i ts neck out by betti ng that such and
such will be the case. It shoul d do so withpreci si on and bol dness. I n
short, speci fi c sci enti fi c theori es shoul d have an appreci abl e predictive
power composed of accuracy and ori gi nal i ty. The predi cti ve perf ormance
of a theory i s, together wi th i ts external consi stency, the mai n t hough not
the onl y factor al l owi ng us to esti mate i ts degree of truth, hence i ts
credence and i ts acceptabi l i ty. Thi s does not arisei nconnecti on withvery
general theori es: these are so many comprehensi ve f rameworks, each
consi stent wi th any number of speci fi c theori es, or theoreti cal model s of a
1
,
246 Mario Bunge
gi ven pi ece of real i ty. The rel ati vi sti c theory of gravi tati on, aut omat a
theory, and the theory of evol uti on are such comprehensi ve f rameworks:
they are untestabl e unl ess enri ched wi th detai l ed model s but, on the
other hand, they have an appreci abl e expl anatory power.
Predi cti ve power is not essenti al and, even when possessed by a theory,
predi cti ve power i s not enough: we want every theory to expl ai n the
facts i n some domai n and we want some theori es to expl ai n certai n
l ower l evel theori es. Every theory shoul d be abl e to expl ai n, at l east in
pri nci pl e, the bul k of what i t purports to cover: in short, i t shoul d have
an appreci abl e explanatory power. The expl anati on shoul d be as accurate
as possi bl e but i t shoul d preferabl y al so gi ve some i nsi ght i nto the
referents of the theory. A theory accounti ng for the behavi our of i ts
referents, wi thout di scl osi ng what makes t hemti ck, can be a good
scientific theory but i t i s not the best possi bl e theory. A ri cher theory
will be a deeper one, a theory hypothesi zi ng unseen mechani sms re-
sponsi bl e for the overt behavi our of the system. For exampl e, of thevari ous
theori es of bi l ogi cal evol uti on proposed so far, the syntheti c theory is the
deepest because i t tel l s us what the mai n evol uti onary mechani smi s,
namel y, the natural sel ecti on of spontaneous vari ati ons. (Thi s does not
mean that the theory i s fi nal , i .e. compl etel y true.) In any case, we want
our theori es to have a substanti al expl anatory power wi thout, however,
expl ai ni ng everythi ng under the sun, as pseudo- sci ence attempts to do.
Further, we di strust expl anati ons contri ved whol l y after the event: we do
not vahe a hi gh expl anatory power unl ess i t i s accompani ed by a reason-
abl e predi cti ve power.
The above features are some of the si gns or i ndi ces of the degree of
truth of a sci enti fi c theory, i.e. of the extent to whi ch the theory consti tutes
a fai thful or adequate conceptual reconstructi on, or symbol i c pi cture,
of a sector of real i ty. I n eval uati ng the degree of truth of a theory i t is
not enough to count i ts hits and misses and to compress this i nf ormati on
i nto a degree of conf i rmati on. A theory may often succeed, or fail
through no meri t or faul t of i ts own: the experi mental techni ques may
have been faul ty, the observati ons may have been carel ess, or the in-
i nferences f r omt hemmay have been i nval i d. observati on, measurement
and experi ment are sel domf ool - proof : they utilizeassumpti ons that may
be fal se; they i nvol ve a host of vari abl es that are not adequatel y
control l ed; and their outcoi nes may be mi si nterpreted. A careful anal ysi s
of the actual test of any sci enti fi c theory must take all these empi ri cal and
i
Metatheory 247
conceptual factors i nto account. Whi l e s wh cri ti cal anal yses of i ndi vi dual
cases are someti mes f ound inthe scientific l i terature, no general theory of
the testi ng of theori es is avai l abl e. Nor , for that matter, do we have mor e
than the begi nni ngs of theori es concerni ng the external consi stency,
testabi l i ty, predi cti ve power, and expl anatory power of theori es. This
whol e matter i s practi cal l y vi rgi n, and what metatheori es there are ini t
sel domconcern real sci ence, and, inparti cul ar, theoreti cal sci ence. And in
any case they have hardl y been subj ected to tests: they are usual l y
bel i eved on authori ty. I n short, the epi stemol ogy of scientific theory
- the supreme i t emof sci enti fi c knowl edge - is sti l l underdevel oped.
1
9. THE PRAGMATICS OF SCIENTIFIC THEORIES
Pragmati cs is concerned wi th the worki ngs of symbol i c f orms i nreal life.
I n parti cul ar, the busi ness of the pragmati cs of sci enti fi c theori es is to
i nvesti gate thei r ori gi n, growt h, use, and di suse, rather than to l ook at
t hemas if they wereready- made and sel f-exi stent.
The study of the ways i n whi ch theori es are i nvented and expanded,
accepted and cri ti ci zed, shoul d be an i nterdi sci pl i nary enterpri se, for i t
has psychol ogi cal , soci ol ogi cal , hi stori cal and methodol ogi cal aspects -
not to ment i on the aestheti c and ethi cal ones. Unf ortunatel y no such
col l ecti ve effort seems to have been attempted. There are onl y a number
of hal f - baked and untested opi ni ons on the life of theori es, and on the
process of theory constructi on and decay.
A popul ar viewon theory constructi on is that i t can be subj ected
to rul es, hence reduced to routi ne operati ons. Thi s vi ewrests on the
opi ni on that every sci enti fi c theory i s j ust an i nducti ve synthesi s of
empi ri cal data that coul d inpri nci pl e be made by a cl everl y pr ogr ammed
computer. The troubl e wi th thi s opi ni on is that i t mi stakes sci enti fi c
theori es for empi ri cal curves (or data i nterpol ati ons) and i t overl ooks
the fact that scientific data are col l ected i n the l i ght of theori es and in
response to probl ems posed wi thi n theori es. At the other end of the
spectrumwe fi nd the vi ewthat theori es are total l y f ree i nventi ons subj ect
to no canon whatever. Thi s opi ni on i s surel y ri ght in emphasi zi ng that
theoreti cal creati on is aki n to artistic creati on, inso far as, inei ther case,
the i magi nati on produces somethi ng radi cal l y newand often far beyond
the fronti ers of percepti on. Neverthel ess, one can di scern styl es and
248 Murio Bunge
i
standards in both cases. Thus we fi nd the phenomenol ogi cal (bl ack box
or ki neti c) styb on the one hand and the dynami cal on the other. And i t
is obvi ous that, no matter how i magi nati ve a theory may be, i t is con-
strai ned by certai n nor ms, such as l ogi cal consi stency, cohesi on, external
consi stency, scrutabi l i ty, predi cti ve power , and so on. Wer e thi s not so,
metatheory woul d be i mpossi bl e. In short, al though there are no rul es
for securi ng theoreti cal ori gi nal i ty and depth, there certai nl y are standards
whi ch every theory must respect if i t is tobe regarded as sci enti fi c rather
than as a pi ece of pop sci ence.
As to the use of sci enti fi c theori es for practi cal purposes, thi s i s a
central thei ne of the phi l osophy of the vari ous technol ogi es - physi cal ,
chemi cal , bi ol ogi cal , psychol ogi cal , and soci ol ogi cal . A fi rst probl emin
this regard i s to determi ne the way theori es must be prepared or worked
out in order to be appl i ed to the del i berate producti on or preventi on of
some practi cal out come, such as the manuf acture of a commodi t y or the
cure of a di sease. Thi s probl emi s rel ated to, but di fferent f rom, the
probl emof the preparati on of a theory for i ts empi ri cal test: inei ther case
one has got to deduce l ow-l evel theorems and summon t he hel p of
subsi di ary assumpti ons and auxi l i ary theori es but, i n the case of the
practi cal appl i cati on, one will l ook for maxi mumeffi ci ency rather than
for maxi mumtruth. Thi s will i nfl uence the actual choi ce of theory: for
practi cal purposes, one will usual l y prefer the theory whi ch is the easi er
to handl e, perhaps al so the one i nvol vi ng the l ower cost, even at the pri ce
of superfi ci al i ty and i naccuracy.
Open probl ems of thi s ki nd are pl enti ful , i ndi cati ng that the prag-
mati cs of theori es, once they are taken seri ousl y and wi thout pragmati st
prej udi ces agai nst theory, will g owconsi derabl y and enabl e us the better
to understand the i nteracti ons between theori zi ng and doi ng.
10. THE PHILOSOPHY OF SCIENTIFIC THEORIES
It i s cl ear that sci enti fi c research is gui ded by a number of heuri sti c cl ues
of a phi l osophi cal type about the nature of the worl d and our knowl edge
of il. Thus, one assumes that thi ngs are grouped i nto natural ki nds, that
facts fall i nto obj ecti ve patterns that can be known, even if onl y gradual l y,
that space and timeconsti tute the basi c structure of the worl d, that there
are no mi nds separate f rombodi es, and so on. I n short, scientific
Metatheory 249
i nvesti gati on presupposes a number of ontol ogi cal and epi stemol ogi cal
hypotheses - not to ment i on a good sl i ce of l ogi c - whi ch are in turn
suggested by sci ence and phi l osophy. Thi s i s hardl y deni abl e: the questi on
i s whether any such phi l osophi cal hypotheses are retai ned in sci enti fi c
theori es once they have been bui l t, or whether they di sappear withthe
scaffol di ng.
Thi s probl emcannot be sol ved wi thout regard to speci al cases: here,
as in every other buddi ng domai n of research, one shoul d proceed
empi ri cal l y. Moreover, an adequate answer cannot be offered before
axi omati zi ng the theory inquesti on. Thi s i s necessary, but may not yet
suffi ce: to di scl ose the phi l osophi cal i ngredi ents of a theory, if any, we
must subj ect i ts key concepts tophi l osophi cal anal ysi s. It i s thus that we
have come to l earn that every sci enti fi c theory deal i ng wi th compl ex
systems mor e or l ess overtl y empl oys the i nnocent- l ooki ng concept of
part - a concept that i s usual l y taken for granted but calls for a speci al
theory (mereol ogy), one that bel ongs to metaphysi cs and must be f ormul -
ated i nmathemati cal terms.
On the other hand, an axi omati c and phi l osophi cal anal ysi s may show
that certai n phi l osophi cal tenets commonl y assi gned to some sci enti fi c
theori es are real l y al i en to t hem. A f amous case i npoi nt i s the doctri ne
of absol ute space and t i me, whi ch we have become accust omed to as-
soci ate wi th Newt oni an mechani cs. We have onl y recentl y come to l earn
that cl assi cal mechani cs does not depend on thi s metaphysi cal hypoth-
esi s: i ndeed, whether space and time are construed as absol ute (self-
exi stent) or as rel ati onal (anchored to thi ngs and thei r changes), one gets
the same testabl e f ormul ae. Hence that metaphysi cal hypothesi s (or
i ts negate) i s redundant to cl assi cal mechani cs and i t must be j udged by
cri teri a other than empi ri cal success or fai l ure. Another cel ebrated case i s
the thesi s that quant ummechani cs i s al ways concerned wi th systems
under observati on and, mor e preci sel y, wi th unanal ysabl e uni ts myste-
ri ousl y consti tuted by physi cal obj ects and observers. Thi s metaphysi cal
thesi s, too, has recentl y been di ssoci ated f r omthe theory thanks to a
semanti cal anal ysi s of i ts basi c (pri mi ti ve) concepts and to the axi omati c
reconsti tuti on of the theory: i t turns out that the concept of observer
does not even occur inthe theory.
I n short, a critical anal ysi s of the axi omati c f oundati ons of any
sci enti fi c theory is bound to reveal a number of genui ne phi l osophi cal
i t ems - chi efl y metaphysi cal categori es l i ke those of system, change, and
250 Mario Bunge
cause - and to di scard certai n phi l osophi cal i deas that had been wrongl y
assumed to be countenanced by the theory. Moreover, a phi l osophi cal
l ook at current sci ence reveal s a number of sci enti fi c theori es that bel ong
by ri ght to metaphysi cs, as bei ng both extremel y general , and untestabl e
wi thout further ado: thi s i s the case of general systems theory, the
general theory of machi nes, automata theory, and i nf ormati on theory.
Paradoxi cal l y enough, hi gh- powered engi neeri ng has become genui ne
exact metaphysi cs. Thi s newmetaphysi cs, created i n the engi neeri ng
school s, may eventual l y repl ace the dyi ng metaphysi cs of the phi l osophy
departments.
1
SUMMARY
I n our timeall of mathemati cs and much of sci ence consi st of theori es.
Hence metatheory, or the theory of theori es, is at the heart of the anal ysi s
of mathemati cs and sci ence.
A theory, whether mathemati cal or factual , has certai n f ormal (l ogi cal
and al gebrai c) characteri sti cs that di sti ngui sh i t f roma body of opi ni ons.
These characteri sti cs are studi ed by metamathemati cs. This di sci pl i ne i s
al so concerned wi th the rel ati ons among theori es as far as thei r structure
and their extensi on are concerned. A powerf ul new tool for the met a-
mathemati cal anal ysi s of whol e cl asses of theori es and thei r rel ati ons
is category theory. Model theory, another newcomer , tackl es the probl em
of the i nterpretati on of abstract theori es and of the rel ati ons among the
model s of one and the same abstract theory; i n addi ti on, i t suppl i es
met hods for the metamathemati cal anal ysi s of theori es. I n the l i ght of
these studi es, most theori es prove to be i ncompl ete and undeci dabl e -
whi ch, far f romconsti tuti ng a defeat of reason, poi nts to boundl ess
possi bi l i ti es of expansi on of the real mof reason.
A theory wi th a factual content poses not onl y metamathemati cal and
model theoreti c probl ems but al so a number of metasci enti fi c probl ems:
semanti cal , epi stemol ogi cal , methodol ogi cal , pragmati c, and even
ontol ogi cal . Central among themare those of factual and empi ri cal
i nterpretati on, factual and parti al truth, empi ri cal scrutabi l i ty, predi cti ve
power, and metaphysi cal commi t ment . All branches of phi l osophy are
rel evant to the study of theoreti cal sci ence and both are bound to benefi t
f r omsuch contacts. After all, metatheory is the consci ence of theoreti cal
Metatheory 251
science and a test of philosophy. Only a philosophy which accounts for
the nature and central role of theory in contemporary science deserves to
be taken into consideration.
REFERENCES
ADDI SON, J. W. , HENKI N, L. and TARSKI , A. (eds.) (1965) The Theory of Models,
Amst er dam, Nort h- Hol l and Publ . Co.
BELL, J. L. and SLOMSON, A. B., (1969) Models and Ultraproducts: An Introduction,
Amst er dam, Nor t h- Hol l and Publ . Co.
BETH, E. W. , (1959) The Foundations of Mathematics, Amst er dam, Nort h- Hol l and
Publ . Co.
BUNGE, M. , (1967) Scientific Research, 2 vol s., Berl i n- Hei del berg- NewYor k, Spri nger-
Verl ag.
BUNGE, M. , (1967) Foundations of Physics, Berl i n- Hei del berg- NewYor k, Spri nger-
Verl ag.
BUNGE, M. , (f orthcomi ng) Meaning and Truth in Science.
CARNAP, R., (1958) Introduction to Symbolic Logic and its Applications, New York,
CURRY, H. B., (1963) Foundations of Mathematical Logic, New Yor k, McGr aw- Hi l l .
HATCHER, W. A. , (1968) Foundations of Mathematics, Phi l adel phi a, W. B. Saunders.
HENKI N, L. , SUPPES, P. and TARSKI , A. (eds.) (1959) The Axiomatic Method, Amst er -
HI LBERT, D. and BERNAYS, P., (1934-1939) Grundlagen der Mathematik, 2 vol s.,
KLEENE, S., (1952) Introduction to Metamathematics, Pri nceton (N.J.), Van Nost rand.
KLI BANSKY, R., (ed.) (1968) Contemporary Philosophy, vol s. I and I I , Fi renze, La
MACLANE, S. and BI RKHOFF, G. , (1967) Algebra, New York, Macmi l l an.
MOSTOWSKI , A. , (1965) Thi rty Year s of Foundati onal Studi es , Acta PhilasophiCa
POPPER, K. R., (1959) The Logic of Scientific Discovery, 2nd edi ti on, London, Hut chi n-
RASI OWA, H. and SI KORSKI , R., (1963) The Mathematics of Metamathematics,
ROBI NSON, A., (1956) Complete Theories, Amst er dam, Nor t h- Hol l and Publ . Co.
ROBI NSON, A. , (1963) Introduction to Model Theory and to the Metamathematics of
SMULLYAN, R. M. , (1961) Theory of Formal Systems, Pri nceton, Pri nceton Uni v. Press.
STOLL, R. R., (1961) Set Theory andlogic, San Franci sco and London, W. H. Fr eeman.
SUPPES, P., (1967) Set-Theoretical Structures in Science, Stanf ord (Calif.), I nsti tute f or
Dover.
dam, Nort h- Hol l and Publ. Co.
Berlin, Spri nger (Rev. ed. of Vol . I: 1968).
Nuova I tal i a Edi tri ce.
Fennica, XVII.
son.
War saw, Paf i st wowe Wydawni ct wo Naukowe.
Algebra, Amst er dam, Nort h- Hol l and Publ . Co.
Mathemati cal Studi es inthe Soci al Sci ences.
252 Mario Bunge
TARSICI, A., (1956) Logic, Semnntics, Metumatliematics, transl. by J. H. Woodger,
TARSICI, A,, MOSTOWSICI, A. and ROBINSON, R. M., (1953) Undecidable Theories, Am-
WOODGER, J. I., (1937) The Axiomatic Method in Biology, Cambridge (Mass.),
Oxford, Clarendon Press.
sterdam, North-Holland Publ. Co.
Cambridge Univ. Prcss.

You might also like