Lecture 1: Tarski on Truth

Philosophy of Logic and Language — HT 2016-17

Jonny McIntosh
[email protected]

Alfred Tarski (1901-1983) was a Polish (and later, American) mathematician, logician,
and philosopher.1 In the 1930s, he published two classic papers: ‘The Concept of Truth
in Formalized Languages’ (1933) and ‘On the Concept of Logical Consequence’ (1936).
He gives a definition of truth for formal languages of logic and mathematics in the first
paper, and the essentials of the model-theoretic definition of logical consequence in the
second. Over the course of the next few lectures, we’ll look at each of these in turn.

1 Background
The notion of truth seems to lie at the centre of a range of other notions that are central
to theorising about the formal languages of logic and mathematics: e.g. validity, con-
sistency, and completeness. But the notion of truth seems to give rise to contradiction:

Let sentence (1) = ‘sentence (1) is not true’. Then:

1. ‘sentence (1) is not true’ is true IFF sentence (1) is not true

2. sentence (1) = ‘sentence (1) is not true’

3. So, sentence (1) is true IFF sentence (1) is not true

4. So, sentence (1) is true and sentence (1) is not true

Tarski is worried that, unless the paradox can be resolved, metatheoretical results in-
voking the notion of truth or other notions that depend on it will be remain suspect. But
how can it be resolved? The second premise is undeniable, and the first premise is an
instance of a schema that seems central to the concept of truth, namely the following:

‘S’ is true IFF S,

1
The eventful story of Tarski’s life is told by Anita and Solomon Feferman in their wonderful biography,
Alfred Tarski: Life and Logic (CUP, 2004).

1
According to Tarski, the problem arises out of the fact that natural languages are seman-
tically closed: for each sentence S, they contain another sentence S’ that attributes truth
(untruth) to S. In order to establish the sorts of metatheoretical results that Tarski is in-
terested in, however, we need to be able to talk about the true and untrue sentences of
formal languages. How are we do that unless those languages are semantically closed?

Tarski’s solution: talk about the true and untrue sentences of one language, called the
object language, in another language, called the metalanguage. Tarski shows how, given
an object language L and metalanguage M that meet certain requirements, we can de-
fine a predicate T rueL in M that applies to all and only the true sentences of L.

2 The Shape of the Problem

2.1 Formal Correctness
Tarski imposes various requirements on a satisfactory definition of T rueL . First, it must
be formally correct, i.e. it must be (or be provably equivalent to) a sentence of the form,

∀x, T rueL (x) if and only if φ(x),

where φ doesn’t contain the predicate T rueL or any other predicates expressing other-
wise obscure notions. (Definitions taking this form are called explicit definitions.)

This requirement ensures that the definition is not circular. However, it’s obviously not
enough to ensure that what is defined is a truth predicate. Consider, for example:

∀x, T rueL (x) if and only if x = x.

This is formally correct, but obviously inadequate: it counts every sentence of L as true.

2.2 Material Adequacy

The definition must also be materially adequate. This is the requirement that the defini-
tion capture the core of the meaning of the word ‘true’. To do this, it must at the very
least be extensionally adequate. That is to say, unlike the definition above, it should yield
a predicate T rueL that applies to all and only the true sentences of the object language.2

Tarski suggests that a (formally correct) definition of T rueL for a given object language
L is materially adequate IFF it entails, for each sentence of L, an instance of the schema:

(T) ‘X’ is T rueL IFF p,

2
It is controversial whether Tarski intended or — as we will see below — achieved anything more than
extensional adequacy.

2
where ‘X’ is replaced by a name of the sentence and ‘p’ by a translation of that sentence
in M . (If M is just an extension of L, ‘p’ will be replaced by the sentence itself.)

Tarski’s suggestion here is his famous Convention T, though a better name for it would
be Criterion T. To see the idea, suppose teenagers start using the predicate ‘peng’.

On the one hand, it seems that ‘peng’ can only be a truth predicate for English if each
instance of the following schema, where what replaces ‘X’ is a name of an English sen-
tence and what replaces ‘p’ is either that sentence or a paraphrase of it, is true:

(P) ‘X’ is peng IFF p,

For example, it will have to be the case that ‘snow is white’ is peng IFF snow is white.

This seems to show that it is a necessary condition on ‘peng’ being a truth predicate
for English that each relevant instance of schema (P) is true. On the other hand, it
also seems to be a sufficient condition. Put roughly, the thought is this. Suppose each
relevant instance of (P) is true. This tells us that, if a sentence of English says that p,
then that sentence is peng IFF p. Yet it also seems that, if a sentence of English says that
p, it is a true sentence of English IFF p. It thus follows that it is peng IFF it is a true.

2.3 Tarski’s Hierarchy

Various constraints on the relationship between L and M fall out of Convention T:

• M has to contain resources for referring to each of the sentences of L.

• M has to contain the sentences of L (or their translations).

Convention T doesn’t in and of itself rule out the possibility that M is identical to T .
But if it is, T will be semantically closed. And we saw at the outset that Tarski thought
that this leads to the Liar Paradox. So Tarski also thinks that M has to be distinct from T .

Here’s a way of seeing the problem Tarski has in mind. Suppose that M is identical to
T . Then T rueL can be defined in L itself and, so long as L contains a negation operator,
there will be a sentence in L, whose name in L is λ, and which is of the form, ¬T rue(λ)
– i.e. a sentence that “says of itself” that it is not true. In accordance with Convention T,
the definition of T rueL will entail a theorem of the form T rue(λ) ≡ ¬T rue(λ). Assum-
ing classical logic, the Liar Paradox follows. If we want to (a) retain classical logic and
(b) allow for the possibility of languages that contain the resources for referring to their
own sentences (and a negation operator), we have to reject the assumption that M = L.

Similarly, M cannot contain its own truth predicate. So to talk about the true sentences
of M , we need a distinct metametalanguage M’, giving rise to a hierarchy of languages.

3
If we distinguish between object languages and their metalanguages, the Liar Paradox
is avoided. The truth predicate defined in the metalanguage will only apply to sen-
tences of the object language, and so to sentences that do not contain that predicate.

3 How Tarski Solves It

3.1 Finite Languages
Tarski also shows how to construct definitions along these lines for various sorts of
languages. It’s very easy in some simple cases. Consider L1 , which contains just two
sentences: ‘1 + 1 = 2’ and ‘1 + 1 = 3’. Assuming these have their ordinary meanings, a
definition of a truth predicate for L1 , using English as our metalanguage, might be:
∀s(s is trueL1 IFF
((s is ‘1 + 1 = 2’ and one plus one is two) OR
(s is ‘1 + 1 = 3’ and one plus one is three)))
• Is this formally correct?
• Is it also materially adequate?

3.2 Recursive Definitions

For ‘trueL1 ’, we just listed out the truth conditions of the various sentences of L1 . So
long as the language contains only a finite number of sentences, we can always take this
sort of approach. But what about when it doesn’t contain a finite number of sentences?

Suppose, for example, we consider L2 , which extends L1 with the sentential connec-
tives, ‘¬’ and ‘∧’ — where these have the meanings that you’re familiar with from 1st
year logic. These operators can be iterated, so L2 contains infinitely many sentences.
In this case, Tarski suggests that we give a recursive definition: we first define the truth
predicate for a certain basic set of sentences of the language, the atomic sentences, and
then extend the definition to all the other sentences in terms of what we have said about
the basic set. For example, a recursive definition of a truth predicate for L2 might be:
∀s(s is trueL2 IFF
((s is ‘1 + 1 = 2’ and one plus one is two) OR
(s is ‘1 + 1 = 3’ and one plus one is three) OR
(s is formed by prefixing a sentence s1 with ‘¬’ and s1 is not trueL2 ) OR
(s is formed by placing ‘∧’ between sentences s1 and s2 and both s1 and s2
are trueL2 )))
• Is this formally correct?
We might worry it’s not, as ‘trueL2 ’ occurs on the RHS. But as Frege and Dedekind
showed, given some mathematical machinery, it’s possible to turn recursive defi-
nitions like this into explicit, eliminative definitions.

4
 • Is it materially adequate?

3.3 Quantification
Quantifiers pose a further problem. We cannot specify the truth conditions of quan-
tified sentences, such as ‘∃x(x + x = 2)’, in terms of the truth values of their parts, as
their parts don’t generally have truth values. Tarski offers two methods for dealing with
this. The first involves replacing variables with names of the objects in the domain of
quantification. But this won’t always work, as in general some objects in the domain of
quantification won’t have names. So for the general case, Tarski developed the method
that will be familiar from the semantics of predicate logic. This involves the notion of a
variable assignment, and treats the variables themselves as a kind of temporary name.

4 Philosophical Significance
At the very least, Tarski shows us how to give explicit definitions of predicates that are
co-extensive with various substitution instances of the predicate ‘is a true sentence of
L’. This in itself is an impressive achievement. But does he do more?

4.1 Defining the Concept of Truth?

Does Tarski provide the means for defining the concept of truth, i.e. for defining predi-
cates that have the same meaning as, say, the English predicate ‘is true’?

Problems:

• The predicate ‘is true’ applies to propositions, not sentences.

OK. But if so, we can ask whether Tarski provides the means for defining predi-
cates that have the same meaning as the English predicate ‘expresses a truth’.

• The English predicate ‘is true’ (or ‘expresses a truth’) applies to sentences of a
range of languages, including English! Tarski’s predicates only apply to one lan-
guage, and do not apply to sentences of the language to which they belong. If
Tarski’s diagnosis of the Liar Paradox is right, this is a serious objection: it means
his predicates cannot be co-extensive with ‘is true’ or ‘expresses a truth’.

4.2 Explicating the Concept of Truth?

Perhaps we could more modestly take Tarski as providing the means for explicating the
concept of truth, i.e. for defining predicates that can replace the predicate ‘is true’ (or
‘expresses a truth’) in all legitimate theoretical contexts, but that lack of any of its de-
fects — and in particular, predicates that don’t give rise to the Liar Paradox.

Problems:

5
• Tarskian definitions, such as the definition of ‘trueL2 ’, might be regarded in either
of two ways: as (1) telling us what the predicate being defined means or (2) as
telling us what the sentences of the object language mean. The second way of
regarding them depends on the idea that to understand a sentence is to know its
truth conditions, an idea that lies at the heart of truth conditional semantics.
The problem is that it seems that a definition of truth cannot accomplish both tasks
at the same time. To accomplish the second, we would have to already know what
the point of describing a sentence as ‘trueL2 ’ is. (Contrast: a theory that tells us
under what conditions the sentences of a language are peng!) But telling us that is
part of the first task. We can call this the problem of truth conditional semantics.
One way to solve the problem of truth conditional semantics is to reject the idea
that to know what the sentences of a language mean is to know their truth condi-
tions. This idea will be discussed in more detail next term’s lectures for 108.

• Various philosophers have raised the following problem of modal difference. Con-
sider the definition of ‘trueL1 ’:
∀s(s is trueL1 IFF
((s is ‘1 + 1 = 2’ and one plus one is two) OR
(s is ‘1 + 1 = 3’ and one plus one is three)))
Now, if we replace ‘trueL1 ’ in
‘1 + 1 = 2’ is trueL1
with its definiens, we get:
(‘1 + 1 = 2’ is ‘1 + 1 = 2’ and one plus one is two) OR (‘1 + 1 = 2’ is ‘1 + 1
= 3’ and one plus one is three)
But now, ‘1 + 1 = 2’ has this property in all worlds in which one plus one is two, i.e.
in all worlds. But it is surely not true in all worlds: for it is surely false in worlds
in which ‘2’ refers to three and all the other words have their actual meaning?
But the languages for which Tarski’s definitions are given are individuated by
their semantics as well as their syntax. So it is a necessary truth that ‘1 + 1 = 2’
means in L1 that one plus one is two. So it is true in all worlds after all.

• Max Black first raised a problem of non-projectability: definitions like those of ‘trueL1 ’
and ‘trueL2 ’ don’t project. They don’t tell us under what conditions other truth
predicates, for other languages, hold of the sentences of those languages.

• There is also a problem of epistemic difference. Knowing the truth conditions of a

sentence, whether or not it’s sufficient for knowing what that sentence means,
at least gives some negative information about its meaning. But knowing the
conditions under which, say, ‘1 + 1 = 2’ is trueL1 doesn’t even provide that.

6
Selected Bibliography
Starred items (*) are more introductory, and good places to start. Tarski (1933) is his
first proper treatment of the topic, but his (1944) is a more accessible presentation. Even
more accessible still is his (1969). Of the rest, Field (1972), Etchemendy (1988), Soames
(1999), and Künne (2003) will be particularly useful in starting out on essays.

Max Black (1948) ‘The Semantic Definition of Truth’ in Analysis 8(4), pp. 49-63.
*Alexis P. Burgess and John P. Burgess (2010) Truth (Princeton UP), Ch. 2
Donald Davidson (1990) ‘The Structure and Content of Truth’ in Journal of Philosophy
87(6), pp. 279-328.
Michael Dummett (1959) ‘Truth’ in Proceedings of the Aristotelian Society 59, pp. 141-162.
John Etchemendy (1988) ‘Tarski on Truth and Logical Consequence’ in The Journal of
Symbolic Logic, 53(1), pp. 51-79, §1.
Hartry Field (1972) ‘Tarski’s Theory of Truth’ in Journal of Philosophy, 69(13), pp. 347-
375.
*Mario Gómez-Torrente (2006) ‘Alfred Tarski’ in E. Zalta, ed. Stanford Encyclopedia of
Philosophy: https://plato.stanford.edu/entries/tarski/
Richard Heck, Jr. (1997) ‘’ The Philosophical Review 106(4), 533-554.
*Wilfrid Hodges (2001) ‘Tarski’s Truth Definitions’ in E. Zalta, ed. Stanford Encyclopedia
of Philosophy: http://plato.stanford.edu/entries/tarski-truth/
*Richard Kirkham (1992) Theories of Truth (MIT Press), Ch. 5 and 6.
Wolfgang Künne (2003) Conceptions of Truth (OUP), §4.2.
John McDowell (1980) ‘Physicalism and Primitive Denotation: Field on Tarski’, reprinted
in his (1998) Meaning, Knowledge, & Reality (Harvard UP).
Douglas Patterson (2011) Alfred Tarski: Philosophy of Language and Logic (Palgrave Macmil-
lan), Ch. 4.
Hilary Putnam (1985) ‘A Comparison of Something with Something Else’ in New Liter-
ary History 17(1), pp. 61-79.
Gila Sher (1999) ‘What is Tarski’s Theory of Truth?’ in Topoi 18(2), pp. 149-166.
Scott Soames (1999) Understanding Truth (OUP), Ch. 3 and 4.
Alfred Tarski (1933) ‘The Concept of Truth in Formalized Languages’ in his (1983) Logic,
Semantics, and Metamathematics, 2nd revised edition (Hackett).
— (1944) ‘The Semantic Conception of Truth’ in Philosophy and Phenomenological Research
4(3), pp. 341-376.
— (1969) ‘Truth and Proof’ in Scientific American 220, pp. 63-77.