The Liar Paradox — “this sentence is false” — is usually treated as a bug in language, a pathological sentence to be quarantined by type hierarchies or paraconsistent logics. It breaks classical semantics because assigning it “true” makes it false, and assigning it “false” makes it true. The standard response is to restrict self-reference or accept the contradiction.
Singh and Singh Parmar (arXiv:2603.24527) take the opposite approach: self-reference is a feature, not a bug, and the Liar Paradox reveals a fundamental mechanism of semantic informativeness. They introduce incongruent normal form (INF), which transforms a self-referential sentence into a family of non-self-referential sentences that are individually satisfiable but collectively incompatible. Each member of the family is well-behaved in isolation. The contradiction lives in the combination.
The key claim: semantic completeness prevents informativeness, while incongruence maintains it. A semantically complete theory — one where every sentence is either provably true or provably false — carries no information in the Shannon sense, because there's nothing left to learn. Incongruence, the property that consistent extensions of a theory can be mutually incompatible, is what preserves the capacity to say something meaningful.
The result extends beyond paradoxes. Any consistent incomplete first-order theory can produce finite incongruent families from its incompatible complete extensions. Gödel's incompleteness theorems guaranteed that interesting theories are incomplete. This work adds that the incompleteness itself — the existence of multiple incompatible ways to complete the theory — is what makes the theory informative. Completeness and informativeness are structurally opposed.
The authors develop a quantitative framework using Boolean functions and Fourier analysis, deriving bounds that connect semantic determinacy, informativeness, and spectral simplicity. The mathematics formalizes an intuition: the sentences that carry the most information are the ones whose truth value is most sensitive to context. Self-referential sentences are the extreme case — their truth value depends on their own truth value, making them maximally context-sensitive and maximally informative in the technical sense.
Whether this vindicates or merely reframes the Liar depends on whether you think “informative” is a good thing for a sentence to be. But the structural insight — that contradiction in the aggregate is compatible with consistency in the parts, and that this aggregate inconsistency is the source of meaning — applies far beyond formal logic.