Tarski’s Concept of Truth

In this post I want to introduce at a high level Alfred Tarski’s semantic theory of truth for formal languages. This is the third in a three-part series on the early period of analytic philosophy. The first post covered the logic of Wittgenstein’s Tractatus. The second post was on Gödel’s proof on the incompleteness of logic and arithmetic. Tarski’s concept of truth closes out this early analytic period. He’s also a bridge to Carnap and Quine. But that’s a future topic.

First, I’m going to introduce the problem of contradictions in ordinary language. This will segue into formalized languages and Tarski’s logical constants. Then we’ll get to the “chief problem” of his paper, which is his definition of a true sentence, the object language, metalanguage, and how true paradox is avoided.

A couple is having an argument. It gets heated. Finally, one turns to the other and says, “I’m not speaking to you!” This is an example of a performative contradiction in ordinary language. The speaker, in word or deed, must affirm the very thing that is being denied. The key to seeing through such utterances is to separate the surface grammar from the underlying meaning itself. Once we realize the person having an argument means to say something like “I’m angry and I can’t continue,” there is no contradiction at all. Colloquial language is filled with performative contradictions.

But what of true contradictions? The gold standard is the Liar’s Paradox that says this sentence is false. It can be true if and only if it’s false. But if it’s false then it’s true. Another classic example in the ancient world was posed by Eubulides of Miletus where a man says what I am saying now is a lie. If it’s true, he’s lying; if it’s false, he’s telling the truth. These self-referential (or indexical) paradoxes cause headaches in logic. Natural everyday language is too flexible. We make puns and jokes that have double meanings. Why did the tomato turn red? It saw the salad dressing. Silence can be deafening. White noise can be quiet. In one war “it became necessary to destroy the town to save it.” People park on driveways but drive on parkways. The wonderful and messy ambiguity of speech goes on and on.

Because of self-reference, Tarski concludes that we can’t construct a formal definition of truth in ordinary language:

“The attempt to set up a structural definition of the term ‘true sentence’—applicable to colloquial language is confronted with insuperable difficulties…. I now abandon the attempt to solve our problem for the language of everyday life and restrict myself henceforth entirely to formalized languages” (p. 164-165).

Tarski’s system borrows its logical constants from the work of Frege, Peano, Russell, and Whitehead before him. In four axioms he recognizes the expressions—the building blocks of variables, constants, and operators—needed to build sentences that evaluate to a truth value. A fifth axiom asserts the principle of induction.

Variables are symbolic names that stand for values. By convention in mathematical logic, letters like n, x, y, and z are used as variable names. They are either free or bound. Given the formula F(n) the variable n is free because it’s a placeholder for some value or a range of values. The formula F(x, y) has two free variables. I might say that x is any integer less than 0 and y is any integer greater than 1. Suppose x = -13 and y = 2. Both are now bound to concrete values. If all the variables in a formula are bound, the formula is said to be closed.

Tarski has four names for the symbols of constants in his system. Axiom 1 lists them as ng, sm, un, and in. These four expressions are negation (~), logical disjunction (⋁), universal quantification (∀), and strict inclusion (⊆), respectively. No two are identical. A frown symbol “⌢” is used to concatenate them in order to create complex descriptions.

For example, take Tarski’s symbolic expression “((ng⌢in)⌢v1)⌢v2” (p. 172). This complex name describes the statement “it is not the case that set v1 is included in set v2.” Let’s assign some concrete values to the free variables. Suppose v1 is the set of all prime numbers and v2 is the set of all even numbers. This would then say it is not the case that the set of all prime numbers is included in the set of all even numbers.

If the symbolic expression were “((in)⌢v1)⌢v2” where v1 is the set of all prime numbers and v2 is the set of all odd numbers, then the expression would say the set of all prime numbers is included in the set of all odd numbers.

In mathematical logic a sentence is a well-formed, closed formula that evaluates to a truth value. Consider this open formula:

(∃x)(x = Sy)

Which says there is an x such that x is the immediate successor of y. This is not a closed formula because y remains free. However, if I say:

(∀y)(∃x)(x = Sy)

This is a closed formula because all variables are bound. This says for any y there exists an x such that x is the immediate successor of y. Given any integer for y, there will always be some x that is its successor (y + 1). Therefore, the sentence evaluates to true. I can describe this formula in Tarski’s syntax by:

(ng⌢un)⌢x⌢(ng)⌢S(y)

Notice that the existential quantifier is modeled by (ng⌢un)⌢x⌢(ng). Like Frege before him, Tarski’s verbose syntax didn’t catch on. But he was definitely ahead of his time. As John Corcoran notes, Tarski created a generative grammar long before linguists like Noam Chomsky in the fifties (p. xxi). With just four primitives, variables, and the frown symbol, Tarski can concatenate strings that can be structured in an abstract syntax tree. I’ve written an F# recursive function that walks such trees, tokenizing and evaluating each primitive for its truth value. I won’t go into such detail in this post.

Now we come to the “chief problem,” namely, “the construction of the definition of a true sentence” using set theory as the object language (p. 186). Tarski uses the symbol Tr “to denote the class of all true sentences” (p. 187). Today we would say “set” instead of classes. This leads to the definition of a true sentence which Tarski calls Convention T (for truth). It has two parts:

(𝝰) x ∈ Tr if and only if p
(𝞫) for any x, if x ∈ Tr then x ∈ S

The first part (𝝰) says x is an element of truth if and only if p, which is to say x is true if and only if p, where x is a structural-descriptive name of any sentence in an object language and p is a translation of the sentence into an expression in the metalanguage. The second part (𝞫) is a mere formality which states that if x is true then x is a true sentence. Recall that a sentence is a well-formed closed formula.

Since x is being replaced by a structural-descriptive name I’m going to put it in quotes, which gives it the final canonical form:

“x” is true if and only if p

The quotation-mark “x” is the description of a sentence made in the object language. The object language could be a sentence of logic, or an ordinary proposition in a language like English or German. The right-side p is the sentence translated into the metalanguage. They must correspond. For example:

“Snow is white” is true if and only if snow is white.

If you’re seeing this for the first time it can be confusing. Is this just some sort of obvious statement or tautology that doesn’t really say anything at all? No, and here’s why. The right-hand p (no quotes) is the sentence expressed in the formal metalanguage. This is the state of affairs in reality where snow is white. But the left-hand “x” is a quotation-mark name that points to the sentence in the object language. The phrase “‘x’ is true” makes a truth claim and this claim asserts that p is true. The object language could have been German in which case:

“Schnee ist weiß” is true if and only if snow is white.

You can see this is at root a correspondence theory of truth. Suppose we’re out walking and I say to you, “that’s a white oak tree”. In Tarski’s formal syntax I can use the quotation-mark name “x” as the description of my spoken sentence. This sentence is then held up against reality (p). My sentence is true if and only if there actually is a white oak tree in reality. In other words, if the state of affairs in reality matches my named sentence, then the truth claim holds. If not, then it is false.

Consider this code snippet, written in the functional programming language F#, that implements Tarski’s Convention T:

Line 4 reflects the fact in reality that snow is white. The structural-descriptive name ‘x’ is a sentence in the object language that “snow is white.” The function on line 7 implements the correspondence between the truth claim and the metalanguage fact. It returns a true value and the result is then printed to the output console.

Tarski thus avoids paradox by strictly separating the language making the statement from the language evaluating its truth. The problem with ordinary English (or any object language) isn’t that they have quirky idioms like “a nickel ain’t worth a dime anymore.” The real danger is semantic closure. This is when the sentence is evaluated for its own truth. When that happens, self-reference can lead straight to the Liar’s Paradox. In Tarski’s formalized language, a truth-schema sentence cannot judge itself. The left-side object language supplies the sentence, but the right-side metalanguage steps outside to judge it against reality. In this way Tarski avoids the problems that plagued Frege and Russell. In previous posts I wrote about Russell’s theory of types. Why did that fail where Tarski’s system does not? Because Russell’s hierarchy of types are like steps in a staircase. They look segregated at first glance. But they’re all still in the same “house” (the same object language). Tarski says the only way to avoid self-reference is to step outside the house in order to evaluate the sentence for truth.

Earlier I asserted that in reality snow is white. Tarski does not concern himself with ontological questions about how we can know if snow is white, or on what basis we form justified true beliefs, or even whether reality is really real at all. He’s a logician. His task is to show how we can adjudicate truth claims in the metalanguage while avoiding paradox. He leaves it up to science to sort fact from fiction.

But what about the metalanguage? Isn’t it subject to the same problems of the object language? Doesn’t it require a meta-metalanguage to adjudicate its own truth values, leading to infinite regress? Absolutely. Tarski proved that there is a way to satisfy truth values within a formal language. But he knows that there can never be an “ultimate” consistent language that gets around Gödel’s proof of incompleteness. Gödel himself seemed to fall back on Platonism—the idea that numbers were in the mind of God or otherwise encoded in reality itself. The alternative is to see mathematics in anti-realist terms. On this view, numbers don’t exist independently of the human mind that constructs them. Mathematics is a useful tool that “just works.” Thus, we can be satisfied with small “t” truth and leave the capital “T” truth to the theologians.

Notes

Tarski, Alfred. “The Concept of Truth in Formalized Languages.” Logic, Semantics, Metamathematics: Papers from 1923 to 1938. Translated by J. H. Woodger, Edited by John Corcoran, 2nd Ed., Hackett Publishing Company, 1983.

Leave a Reply