tke an “axiom of uncountabiliy”, that i, it must
have 2 formula with uncountable models but no
countable models, If the uncountable is no logical
cancept, one is tempted to regard this consequence
asa proof that there are no complete logics extend-
ing elementary logic. However that would be a
mistake. Elementary logic alresdy has “axioms of
infinity” in a similar sense, That is, there are
formulas with models of all infinite cardinalities
bat with no finite models, even though infinite”
isnot a logical concept in the sense of there being a
‘quantifier expressing “for infinitely many 2”
Being able to distinguish the uncountable by
‘means of axioms of uncountabilty is evidently 2
‘much weaker property than having a quantifier
“for uncountably many 2”. Further, if one allows
even weiker criteria, elementary logic is already
able to distinguish the uncountable from the coun-
table, Call theory with infinite models categorical
in the cardinal ¥ if all models of cardinality 8 are
isomorphic, It ie known that there also theories
categorical in Ny but in no uncountable cardinal
‘There are also theories categorical in all uncoun-
table cardinals but not in No; theories categorical
in all infinite cardinals; and theories categorical in
io infinite cardinals. Morley’s theorem demon-
strates that for infinite cardinals, these are the
only possible categories. In this sense, elementary
logic cannot distinguish two uncountable cardi
nals, although it can distinguish the countably
infinite from the uncountably infinite.
"From this extensional viewpoint, Shelab’s quan~
tier “cofinality w” fares considerably better than
the quantifier (Us). One has axioms of uncount-
ability in precisely the sme sense one has axioms of
infinity, aamely formulas with predicate letters
satisfiable in, and only in, the uncountable uni
verses. Other gross properties seem reasonable
also: just as one cannot have formulas satisfiable
cxactly in finite universes, this logic has no formu
las satisfiable exactly in countable universes. The
‘obvious objection against the Sheleh loge is that
fone would never have supposed that a technical
notion like cofinality was a logical concept. Unless
it could be shown equivalent to some more
palatable notion, dhis must remain a serious objec
Continuity of the Standard Quantifiers
fone considers, instead of the entire logic EL, the
standard quantifiers, it would appear that there
Which Logic is the Right Logic?
must be some sense in which V and 3 are very
Simple and primitive. Ie may be posible to state
some natral condition which expreses this sim
plisiy, and which, at che same time, rules out
{uantfiers one feels are no part of gi. To start
With, Vand may be egarded as extrapolations of
the trath functional connectives A and V oinfnite
domains. To see how one can make such extrapo-
Invions, take the infinite fist of sententialletrs
PyP,Pay ony and suppose one i given an arbi
tray truth assignment that fonction which
assigns atrath valu (2) to cach lever P, By the
truth ble res, ¢ may be extended to assign a
ale (sf) fo each formula of of semen logic
Consider, for increasing m the wl (Pp AP)
‘A. MPa) asiged othe iit conjunctions, One
Seesat once that no matter what assignment 5, the
Timi im n+ 00 1€Pp A Py A.» Na) bas a clear
vale there ea Gite point N’ such thatthe eon-
Junction (Py Py A.A Py) is assigned a value
HPAP) AcvAPy)y and for all m preter
than MiP APL Ao APy).i6 the same as
HOA Py Aree Py) Tes ely sen that exactly
the same propery is rue offsite disjunctions.
‘Compare another familiar binary connective,
the bicondonal , Tt is also commattve and
associative so that parentheses may be dropped,
and (Po ++ Py +
formula, However inthis case there i no evident
limit ofthe values (Pp w+ Pj n+... Py) 35-8
inereases, This isnot merely due tothe unfani-
larity of the construction. If for example, (P) is
“L(llsehood) for all” then the value of
Hyer Phew oe Pa) is T for even, and 1
for m odd. There is simply no wel-defined Lit
value to be wed fran infinite quaniaton
“These considerations canbe estatd in the lan-
rage of quuntfers and predicates. Suppose 2
motel As given which interes a one-place
Jeter F. In each Gite submodel J with universe
{{do, .-+e},¥xF(z) has a truth value, the same,
aloe as (F (lo) A Flat) A.A ley) has in J.
js continuous in the sense that for cach model
theresa nite submodel Jin which F(x) takes
a truth valve and holds that truth value forall
submodels K between 7 and M, including M.
‘Thus if one thinks of the model Mas being
reveled step by step, there i faite portion at
‘hich VeP(s) assumes a tuth value and holds
that truth ale no mater how moch the model
further reveled, and even if is woally revealed
"This is the Continuity condition we wish w
ibolate, and of course the quansifer 3, a5 wel as
@
+ Py) considered to be 2