PY4804 Philosophy of Logic

Week 9: Bolzano and Tarski on Logical Consequence

1. Completeness

(i) (a) Weak completeness: if φ then φ

First-order logic is weakly complete.

(b) Gödel’s completeness theorem: if φ is logically true (in the restricted functional
calculus, i.e., first-order logic) then PM φ

(c) Strong completeness: if Σ φ then Σ φ

First-order logic is strongly complete.

(d) Model existence lemma: if Σ is simply consistent then Σ has a model (equivalent
to strong completeness). (Σ is simply consistent if Σ / ⊥ .)

(ii) Compactness:

(a) A logic is compact if any set of wffs has a model iff every finite subset has a
model.

(b) Strong completeness entails compactness.

(c) First-order logic is compact. Second-order logic is neither compact nor,

consequently, complete.

(iii) ω-completeness and ω-consistency:

(a) Σ is -complete if (∀x)φ ∈ Σ whenever φ(n) ∈ Σ for all n ∈ ω. (‘n’ is the

numeral in Σ denoting n.)

(b) Σ is -consistent if (∃x)φ ∉ Σ whenever ¬φ(n) ∈ Σ for all n ∈ ω.

(c) ω-consistency entails simple consistency, but not vice versa, and completeness
entails ω-completeness but not vice versa.

(iv) Gödel’s incompleteness theorem: in every system of arithmetic, Π, satisfying certain

plausible constraints, there is a wff G such that neither Π G nor Π ¬G, i.e. G is
undecidable in Π. G has the form (∀x)ψ, specifically, (∀x)(Tx ⊃ ¬Pr(x)), where T is
uniquely true of the Gödel number of G, and Pr(x) “means” (or represents) ‘x is
provable’ (i.e., if n is the Gödel number of φ and φ is provable then Π Pr(n), and if φ
is not provable, Π ¬Pr(n).). ψ(n/x) is provable in Π for every n.
(v) (a) the ω-rule, an infinitary rule of inference: from φ(0), φ(1), … , i.e. φ(n) for every
n ∈ ω, infer (∀x)φ.

(b) first-order arithmetic with the ω-rule is complete.

 2

(c) first-order arithmetic has non-standard models, models in which, e.g., the Gödel
sentence is false.

2. Satisfaction

(a) An interpretation of the language of predicate logic consists of a domain D, and

an assignment v = 〈v1,v2〉, where v1 maps individual constants to members of D,
and v2 maps (n-place) predicates to sets of (n-tuples of) members of D.

A sequence is a map from N (the natural numbers) to D, 〈s 1,s2,…〉.

· if A is an atomic wff, φt, the sequence s satisfies A (written s v A) if

{ svi1∈(a)v2∈(φ)v2(φ)
if t is the individual variable xi
if t is the individual name a

· s v ¬A if s / v A

· s v A ∨ B if s v A or s v B

· s v (∃xi)A if s´ v A, where s´ differs from s at most in its ith place,

i.e. sj = sj´ for all j ≠ i

A proposition A is true (under v) if s v A for all sequences s. (Cf. Handout

week 2, Appendix.)

(b) Ajdukiewicz’s syntactic categories: let ι be the type of individuals, and ο the
type of propositions. Then, if α, β are types, αβ is a type, specifically, the type of
functions from objects of type β to objects of type α. E.g., οι is the type of
predicates, i.e. from individuals to propositions, that is, if φ is of type οι and a is
of type ι, then φa is of type ο, i.e. a proposition. (n-place predicates are of type
οιι…ι with n ιs.)

Then Boolean connectives are functions from propositions to propositions, e.g.,

‘¬’ has type οο, and ‘∨’ has type οοο (i.e., (οο)ο—bracketing is always to the
left), which takes a proposition and yields a function from propositions to
propositions, one which takes a proposition and yields a proposition. E.g., A ∨ B
is analysed as (∨(B))(A), which reduces to A(∨B) and written A∨B. What are
usually called ‘functions’ are functions of type ιι(or more generaly, ιι…ι with
n+1 ιs, for an n-place function).

(c) The account of satifaction in (a) can now be generalized to the language of type
theory in (b). (Exercise, optional.) (See, e.g., L. Henkin, ‘Completeness in the
theory of types’, Journal of Symbolic Logic 15 (1950), 81-91. Note that the type
theory above follows Henkin and Church. Ajdukiewicz’s syntactic categories
s s
were s for sentences, n for names, n for predicates, ss for binary truth-functions
and so on. See K. Ajdukiewicz, ‘Syntactic connexion’, tr. in Polish Logic: 1920-
39, ed. S. McCall, 207-31.)
 3

(d) A sentence X is logically true w.r.t. F if X´ is satisfied by every sequence, where

X´ results from X by replacing every atomic expression in X by a variable of the
appropriate type. (Etchemendy p. 47)

X is a logical consequence of K w.r.t. F if 〈K´,X´〉 is satisfaction-preserving on

all sequences, where 〈K´,X´〉 results from 〈K,X〉 by replacing all atomic
expressions not in F by variables (of the appropriate type), and 〈K´,X´〉 is
satisfaction-preserving on the sequence s if either s satisfies X´ or s does not
satisfy some member of K´. (Etchemendy p. 49)

3. Tarski’s concept of logical consequence

(i) Intuitively, valid arguments are truth-preserving, that is, the conclusion is true
whenever the premises are true:

(a) material consequence: either on of the premises is false or the conclusion is true.

(b) logical consequence: necessarily, if the premises are true then the conclusion is
true.

(ii) Carnap (Logical Syntax of Language, 1934):

(a) “It is impossible by the aid of simple methods to frame a definition for the term
‘consequence’ in its full comprehension. Such a definition has never yet been
achieved in modern logic (nor, of course, in the older logic).” (p. 27)

(b) “[X] is called a consequence of [K] in language II [i.e. classical mathematics] if

[K ∪{¬X}] is contradictory.” (p. 117)

Cf. Tarski, ‘On the concept of logical consequence’: “the decisive element of the
above definition obviously is the concept ‘contradictory’. Carnap’s definition of
this concept is too complicated and special to be reproduced here without long
and troublesome explanations … [but is] materially inadequate, just because the
defined concepts depend essentially … on the richness of the language
investigated” (pp. 414, 416 n.)

(iii) Tarski, op. cit. (1936), p. 415:

(F) X follows logically from K only if X´ is true whenever every member of K´ is

true, where X´, K´ differ from X, K only by replacement of all constants except
logical constants.

This is a necessary condition. But is it sufficient? No, because the language may have
an insufficient stock of extra-logical constants.

(iv) Persistence (Etchemendy, The Concept of Logical Consequence, p. 30): “The

property of being logically true with respect to a given F [class of fixed logical terms]
should persist through simple expansions of the language … [and] the property of not
being logically true should persist through contractions of the language.”
 4

(v) Tarski’s definition of logical consequence (ibid. p. 417):

(a) ‘Let L be any class of sentences. We replace all extra-logical constants which
occur in sentences belonging to L by corresponding variables, like constants
being replaced by like variables, and unlike by unlike. In this way we obtain a
class L´ of sentential functions. An arbitrary sequence of objects which satisfies
every sentential function of the class L´ will be called a model or realization of
the class L of sentences.”

(b) “The sentence X follows logically from the sentences of the class K if and only if
every model of the class K is also a model of the sentence X.”

(c) “It is not difficult to reconcile the proposed definition with that of Carnap. For we
can agree to call a class of sentences contradictory if it possesses no model.”

(d) “Underlying our whole construction is the division of all terms of the language
discussed into logical and extra-logical. This division is certainly not quite
arbitrary.” (p. 418)

(vi) Hilbert’s thesis (Etchemendy, p. 6): “all and only logically valid arguments of a
given language are provable in a given deductive system.”

Cf. Church’s thesis : intuitively computable functions coincide with the precise class
of functions defined severally by Church, Gödel, Turing and others.

(vii) If Tarski’s analysis of logical consequence is right, then first-order logic is complete,
and the ω-rule is unsound. Etchemendy claims that Tarski’s definition both
undergenerates (it declares invalid certain inferences, e.g., the ω-rule, which are valid)
and overgenerates (it declares valid certain inferences which are invalid).

3. Bolzano’s definition of logical truth and logical consequence

(i) Bernard Bolzano (1781-1848):born in Prague, studied philosophy, mathematics and

theology, ordained a priest, appointed to the Chair of the Science of Religion at
Charles University in Prague in 1806. Accused of teaching political and religious
heresy in 1819 and removed from the Chair. Spent rest of his life writing on
mathematics and philosophy, much of it unpublished on his death (e.g., The
Paradoxes of the Infinite), and what was published, e.g., the Wissenschaftslehre
(Theory of Science) 1837, largely neglected until the twentieth century.

(ii) (a) “Among the definitions of [the concept of deducibility] … one of the best is that
of Aristotle: ‘a syllogism is a discourse in which, certain things being stated,
something other than what is stated follows of necessity from their being so.’
Since there can be no doubt that Aristotle assumed that the relation of
deducibility can hold between false propositions, the ‘follows of necessity’ can
hardly be interpreted in any other way than this: that the conclusion becomes true
whenever the premises are true. Now it is obvious that we cannot say of one and
the same class of propositions that one of them becomes true whenever the others
are true, unless we envisage some of their parts as variable. For propositions none
of whose parts change are not sometimes true and sometimes false; they are
always one or the other. Hence when it was said of certain propositions that one
 5

of them becomes true as soon as the others do, the actual reference was not to
these propositions themselves, but to a relation which holds between the
infinitely many propositions which can be generated from them, if certain of their
ideas are replaced by arbitrarily chosen other ideas.” (Wissenschaftslehre § 155
note 1)

(b) “We consider in a given proposition not merely whether it is itself true or false,
but also what relation to the truth follows for all the propositions that develop out
of it when we assume certain of the ideas present in it to be variable and permit
ourselves to exchange them for whatever other ideas … If the proposition A is of
such a character that the propositions that can be derived from it when only
[meaningful] propositions may be constructed and ideas i, j, … alone are looked
on as variable are all true, then … we can call the proposition a universally or
completely valid proposition.” (§ 147)

(c) “Suppose there is just a single idea in [a proposition] which can be arbitrarily
varied without disturbing its truth or falsity, i.e., … all the propositions produced
by substituting for this idea any other idea … are either true altogether or false
altogether, presupposing only that they have [meaning]. […] I permit myself,
then, to call propositions of this kind, borrowing an expression from Kant,
analytic.” (§ 148)

(d) “Compare several propositions A,B,C,D, … and regard certain ideas i, j, …

common to all of them as the arbitrary ones, then the question arises whether
there may well be some ideas substituted for i, j, … which are so constituted that
those propositions thereupon all become true at the same time […] I call the
relation obtaining among propositions A,B,C,D, … a relation of compatibility
[…] and the propositions themselves compatible.” (§ 154)

(e) “[If] there is a relation among the compatible propositions A,B,C,D, … , M,N,O,
… such that all of the ideas that make a certain section of these propositions true,
namely, A,B,C,D, … when substituted for i, j, … also have the property of
making some other section of them, namely, M,N,O, … true […] it puts us in the
position … to obtain immediately from the known truth of A,B,C,D, … the truth
of M, N, O, … as well. Consequently, I give the relationship which subsists
between propositions A,B,C,D, … on the one hand and M,N,O, … on the other
the title, a relationship of derivability.” (§ 155)

(iii) Bolzano’s criterion of logical consequence arguably does not satisfy persistence. It
depends on which ideas are expressible in the language. Tarski’s use of satisfaction
was designed to avoid this problem.