Category: logic
Type: paradox
Associated thinker(s): Ancient Megarian logicians, Alfred Tarski, Saul Kripke
Original publication or approximate period: Ancient tradition; influential modern semantic treatments in the twentieth century
Difficulty: Intermediate
Themes: truth, self-reference, semantics
Related experiments: Russell’s Paradox; The Unexpected Hanging / Surprise Test
Consider a sentence L that says of itself: L is not true. If L is true, what it says holds, so L is not true. If L is not true, what it says appears to hold, so it appears true. Ordinary principles seem to drive us in both directions.
This formulation uses “not true” rather than “a lie.” Lying involves intention and belief, which distract from the semantic difficulty. Also distinguish the simple sentence “this sentence is false” from strengthened versions: a proposal that leaves one sentence without a truth value must explain what happens when the sentence instead says it is not true.
How can a language talk about truth without licensing contradictions or giving up important expressive resources?
The puzzle concerns the interaction of reference, truth, and inference. It challenges the assumption that every meaningful declarative sentence can be placed without complication into exactly one of two boxes. A successful account needs more than a verdict on one curiosity: it must explain what ordinary truth talk is allowed to do.
Hierarchy approaches: distinguish an object language from a stronger language used to attribute truth to it. Restricting which truth predicates apply to which sentences blocks the problematic construction, at a cost to unrestricted expressive freedom.
Truth-value-gap approaches: some self-referential sentences fail to be true or false. Strengthened liars then test the limits of the proposed language.
Paraconsistent approaches: permit some contradictions without allowing every proposition to follow. This is not a license for arbitrary inference; carefully restricted logical rules do substantial work.
Not every self-referential sentence is paradoxical: “this sentence contains five words” merely invites counting. The claim that a Cretan says all Cretans lie is not automatically the precise self-referential construction needed here. Finally, rejecting one unrestricted truth principle does not show that ordinary factual inquiry collapses.
Liar cycles spread reference across several sentences. Strengthened liars target notions such as non-truth or lack of groundedness. Their purpose is to test whether a proposed repair handles the underlying mechanism instead of only banning one conspicuous wording.
The article avoids claiming a single securely identified ancient inventor. The retrieved SEP survey supplies the historical cautions and modern response families. No particular theory is described as a final consensus solution.
Research check: 2026-10-30. References above were retrieved for the support described; this is an original educational explanation, not a substitute for the primary texts. Additional speculative comparisons are teaching prompts, not established historical variations.