Liar Paradox

The setup

Take the sentence:

L: This sentence is false.

Two assumptions are doing the heavy lifting. First, bivalence — every meaningful declarative sentence is either true or false, with no third option. Second, the T-schema — for any sentence p, "p" is true if and only if p. Aristotle put it crisply in Metaphysics Γ.7: "to say of what is that it is, and of what is not that it is not, is true." Tarski formalized the schema in 1933.

Now run L through both assumptions. Suppose L is true. Then what L says is the case — namely, that L is false. So L is false. Suppose L is false. Then what L says fails — but what L says is that L is false, and that failure means L is not false, i.e. L is true. Either branch of the assumed bivalence gives the opposite of itself.

The paradox does not depend on tone, intent, or context. It survives translation. It survives being whispered. It is a structural collision between three apparently innocent commitments: a language that contains its own truth predicate, a T-schema that fixes its meaning, and bivalence over the resulting vocabulary.

A short history

Eubulides of Miletus, working in the Megarian school in the fourth century BCE, listed seven paradoxes including the Liar, the Sorites (heap), the Horns, and the Hooded Man. Cicero notes in Academica that Chrysippus the Stoic wrote multiple treatises on the liar. The medieval logicians compiled the so-called insolubilia — Bradwardine, Buridan, and Albert of Saxony each proposed solutions, with Buridan's Sophismata (c. 1350) often being singled out as the most sophisticated pre-modern attempt.

The often-quoted "Epimenides paradox" — the Cretan poet who said "all Cretans are liars" — is older still, cited by St. Paul in Titus 1:12 and traceable through Callimachus to Epimenides himself in the seventh century BCE. But strictly read, it isn't paradoxical: Epimenides could simply be wrong, with at least one Cretan having uttered at least one truth. Eubulides' single-sentence form is what bites.

The modern revival begins with the antinomies that Russell discovered in the foundations of mathematics. Russell's paradox of the set of all sets that don't contain themselves (1901) is the set-theoretic cousin; the liar is the semantic version. Russell's 1908 type theory — "Mathematical Logic as Based on the Theory of Types" — was an attempt to block both at once by stratifying entities into a hierarchy of types. Tarski sharpened this for semantics specifically in his 1933 monograph "The Concept of Truth in Formalized Languages."

The Tarski schema and the hierarchy

Tarski's diagnosis is that natural language is semantically closed — it contains its own truth predicate ("is true" applies to its own sentences) — and this is the disease. Any semantically closed language with classical logic is inconsistent. The cure is to refuse semantic closure.

Concretely, Tarski stratifies languages. An object-language L₀ contains sentences but no truth predicate for itself. A metalanguage L₁ contains a predicate "true-in-L₀" plus a quotation mechanism. A meta-metalanguage L₂ contains "true-in-L₁", and so on. The T-schema is preserved at each level: for any sentence S of Lₙ, the metalanguage Lₙ₊₁ contains the biconditional "S is true-in-Lₙ ↔ ". The liar can't be formed because the truth predicate it would need lives one level up — beyond the reach of self-reference.

This works for formal languages. The cost is that natural English, which lets us cheerfully say "the previous sentence is true," cannot be modeled as a single coherent language on Tarski's view. He himself acknowledged this and treated his result as a no-go theorem for formalizing truth in unrestricted natural language.

Approaches to the liar compared

Approach	Champion	Move	Cost
Tarski hierarchy	Tarski (1933)	Stratify languages; no language contains its own truth predicate	Natural language can't be modeled as one language
Truth-value gaps	Kripke (1975)	The liar is neither true nor false (undefined fixed point)	Strengthened liar — "this sentence is not true" — re-creates the problem
Truth-value gluts (dialetheism)	Priest (1979–87)	Accept that the liar is both true and false; switch to paraconsistent logic LP	Must abandon classical inference rules; many find true contradictions implausible
Contextualism	Parsons, Burge, Glanzberg	The liar shifts context as you reason about it; truth values change with context	Spelling out the parameter rigorously is hard
Revision theory of truth	Gupta & Belnap (1993)	Truth is a circular definition whose values revise across hypothetical iterations	Mathematically heavy; status of "ultimate" verdict unclear
Meaningless / category mistake	Goldstein, Prior, some Wittgensteinians	The liar fails to express a proposition	Must explain what's wrong without circularity; analogous mathematical paradoxes still stand
Substructural logics	Ripley, Beall (2010s)	Reject contraction or transitivity rather than non-contradiction	Counterintuitive consequences in unrelated arguments

None of these is settled. Each survives in the literature because each captures something the others miss.

Kripke's fixed-point construction

Saul Kripke's 1975 "Outline of a Theory of Truth" gives the most influential post-Tarskian solution. He drops bivalence: sentences can be true, false, or undefined. Starting from a base interpretation where the truth predicate has empty extension and anti-extension, he iteratively adds sentences whose truth values are settled by what's already in. The construction reaches a fixed point: at that stage, the truth predicate's extension and anti-extension are stable. Ordinary sentences ("snow is white") get their normal values. The liar lands in the gap — neither in the extension nor the anti-extension of "true." It is, in Kripke's term, ungrounded.

Kripke's construction handles plain liars and many self-referential sentences elegantly. The trouble is the strengthened liar: "This sentence is not true." Now if the sentence is in the gap, then it's not true — and that's exactly what it says, so it should be true. The classical paradox returns one floor up. Kripke's response is that "true" inside the formal theory is the grounded truth predicate, and the strengthened liar uses an external metalinguistic notion that the theory doesn't claim to capture.

Yablo's paradox: self-reference is not the disease

Stephen Yablo's 1993 variant is the cleanest argument that self-reference cannot be the source. Consider an infinite sequence of sentences:

S₁: For all k > 1, Sₖ is false.
S₂: For all k > 2, Sₖ is false.
S₃: For all k > 3, Sₖ is false.
… and so on.

No sentence refers to itself. Each Sₙ refers only to sentences with larger indices. Yet a contradiction arises identically. Suppose some Sₙ is true. Then all later sentences are false. So Sₙ₊₁ is false — but Sₙ₊₁ says all sentences after it are false, which (given Sₙ's truth) is true, so Sₙ₊₁ is true, contradiction. Suppose Sₙ is false. Then some later Sₘ is true, and the same regress kicks off.

This matters because if the liar were just a self-reference glitch, Yablo's sequence wouldn't paradoxify. But it does. So solutions that work by banning self-reference (early Russell, naïve readings of Tarski) are addressing a symptom, not the cause. The real cause appears to be a kind of unbounded negative dependency.

Why the paradox matters

The liar is not a parlor trick. It powers two of the most important results in twentieth-century logic.

Gödel's first incompleteness theorem (1931). Gödel constructs an arithmetic sentence G that effectively says "G is not provable in this system." If the system is consistent, G is true but unprovable. The construction is a precise arithmetic analog of the liar — replacing "false" with "unprovable" sidesteps the contradiction and yields incompleteness instead.

Tarski's undefinability theorem (1933). Building on the same self-reference machinery, Tarski proves that for any sufficiently rich formal language with classical logic, the predicate "is a true sentence of L" cannot itself be defined inside L. Truth is essentially metalinguistic.

Beyond logic, the liar appears in computer science (the halting problem uses an analog), in self-reference in computing systems, and in epistemology around revenge problems for theories of belief.

Counterarguments

The Tarski-style hierarchy is the orthodox response, but it has serious critics. Three lines stand out.

The natural-language objection (Kripke, Priest). Ordinary speakers freely use "true" of their own sentences. If the cure for the liar is to deny that we can ever do this coherently, the cure is worse than the disease.

The revenge problem. Every solution proposed faces a strengthened liar that uses the solution's own resources. Gap theorists face "this sentence is not true." Glut theorists face "this sentence is just false (and not also true)." Contextualists face sentences that quantify over all contexts. Whether revenge is fatal or merely a feature of any genuinely deep concept is itself contested.

The paraconsistent challenge. Graham Priest's dialetheism takes the liar at face value and argues that some contradictions are simply true. Combined with paraconsistent logic LP (which blocks the inference from any contradiction to everything), this avoids the explosion that classical logic would suffer. Critics including JC Beall, Tim Maudlin, and others have argued back, but the position is durable.

Variants and cousins

• The strengthened liar. "This sentence is not true." Defeats gap theories that rely on a third value.
• Curry's paradox. "If this sentence is true, then God exists." Yields any conclusion you like and doesn't even need negation — it works in positive logics, which makes it harder to dodge than the liar.
• The card paradox. Side A: "The sentence on side B is true." Side B: "The sentence on side A is false." Distributed self-reference that needs no single self-referring sentence.
• The pseudo-scotus paradox. "If this argument is valid, then the moon is made of cheese." A medieval relative that makes the same explosion threat as Curry.
• Yablo's paradox. Infinite descent without any self-reference, as covered above.
• Berry's paradox. "The smallest positive integer not definable in fewer than twelve words." A semantic paradox involving definability rather than truth.

Common confusions

• "It just doesn't mean anything." Easy to say, hard to defend. Why does the syntactically identical "this sentence is in English" make perfect sense? Any meaninglessness verdict needs principled criteria.
• "Self-reference is banned." Yablo's paradox falsifies this. So does the card paradox. The disease isn't self-reference per se — it's a structural pattern of negative dependency that self-reference is one way of producing.
• "Tarski solved it." Tarski solved it for formal languages. He explicitly held that natural language is semantically closed and therefore strictly inconsistent. Most contemporary work tries to do better.
• The liar is not the same as the Epimenides paradox. Epimenides has a stable resolution (he was just wrong); the liar does not.
• The liar is not the sorites. The sorites is about vague predicates — when does a heap stop being a heap? It's a different paradox with different solutions, even though both are old Megarian puzzles.