{
  "id": "liar-paradox",
  "title": "Liar Paradox",
  "lang": "en",
  "source_lang": "en",
  "summary": "The liar paradox exposes a conflict between self-referential statements about truth, unrestricted truth principles, and ordinary rules of logic.",
  "category": "philosophy",
  "subcategory": "metaphysics-epistemology",
  "type": "concept",
  "aliases": [
    "Liar antinomy",
    "Epimenides paradox"
  ],
  "generated_at": "2026-11-03",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/liar-paradox/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/liar-paradox/",
      "json": "https://aiwiki.page/zh/liar-paradox.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "The liar paradox is a semantic paradox exemplified by the sentence “This sentence is not true.” If that sentence is true, what it says must hold, so it is not true. If it is not true, it appears to say exactly how things stand, so it is true. The difficulty concerns the interaction of truth, negation, and self-reference, rather than lying in the ordinary sense of deliberately deceiving someone. No speaker, intention to deceive, or mistaken belief is required.\n\nThe paradox is important because its reasoning uses principles that otherwise appear fundamental to understanding language and logic. A sentence normally seems equivalent to an assertion that the sentence is true: “Snow is white” and “‘Snow is white’ is true” should agree. Yet applying this relationship without restriction to sentences that describe their own truth status produces contradiction. Proposed treatments therefore alter some combination of the language’s expressive resources, its account of truth, its rules of inference, or its treatment of context.\n\nKnown in antiquity and extensively studied by medieval logicians, the paradox became central to twentieth-century work on formal semantics and mathematical logic. Its modern forms connect ordinary linguistic puzzles with precise results about definability, formal proof, and the limits of languages capable of describing themselves."
    },
    {
      "heading": "The structure of the contradiction",
      "text": "A standard version of the paradox is the sentence:\n\nThis sentence is false.\n\nCall this sentence \\(L\\). Suppose first that \\(L\\) is true. Its content is that \\(L\\) is false, so truth leads to falsity. Suppose instead that \\(L\\) is false. Since being false is precisely what \\(L\\) asserts about itself, its assertion appears correct, making it true. Under the familiar assumption that every meaningful declarative sentence has exactly one of the two truth values, neither assignment is stable.\n\nThe phrase “is false” can obscure an important distinction. In a two-valued setting, falsity and lack of truth coincide. In theories allowing sentences that are neither true nor false, they do not. Consequently, the strengthened liar is often formulated as “This sentence is not true.” Simply classifying that sentence as neither true nor false appears to establish what it says: it is not true. This does not automatically refute every theory with truth-value gaps, but it shows why a third label is not, by itself, a complete treatment.\n\nThe underlying truth principle is commonly expressed through the T-schema:\n\n\\[T(\\ulcorner A\\urcorner)\\leftrightarrow A.\\]\n\nHere \\(T\\) is a truth predicate, \\(A\\) is a sentence, and \\(\\ulcorner A\\urcorner\\) is a name or code of that sentence. An instance is “‘Snow is white’ is true if and only if snow is white.” The marks around \\(A\\) distinguish mentioning a sentence from using it to make an assertion. If a liar sentence satisfies \\(L\\leftrightarrow\\neg T(\\ulcorner L\\urcorner)\\), applying the T-schema to \\(L\\) yields an equivalence between a claim and its own negation.\n\nSeveral principles must be distinguished. The principle of bivalence says that every relevant sentence is either true or false, with neither gaps nor overlaps. The law of excluded middle is the logical schema \\(A\\lor\\neg A\\). The principle of noncontradiction rejects \\(A\\land\\neg A\\). These principles have different roles, and abandoning one does not automatically require abandoning all the others. Some semantic approaches reject bivalence while retaining excluded middle; others permit certain contradictions while rejecting the inference from a contradiction to an arbitrary conclusion.\n\nSelf-reference alone is not sufficient for paradox. “This sentence is written in English” can be straightforwardly true. Even “This sentence is true,” sometimes called a truth-teller, does not generate the liar’s reversal: its truth status may be underdetermined, but assigning it truth does not force falsity. The distinctive problem arises when semantic self-reference is combined with a negative condition on the sentence’s own truth."
    },
    {
      "heading": "Ancient and medieval development",
      "text": "The liar is traditionally associated with Eubulides of Miletus, a fourth-century BCE philosopher connected with the Megarian school. Ancient reports credit him with several puzzles, including the liar, although the surviving evidence does not establish a single original wording. Diogenes Laertius records the attribution centuries later. The paradox also occupied Stoic logicians, notably Chrysippus, whose extensive writings on it are known chiefly through reports and titles rather than surviving complete treatises.\n\nAnother famous association is with Epimenides, a Cretan traditionally credited with saying that Cretans are always liars. A related statement appears in the New Testament, in the Letter to Titus, where a Cretan’s characterization of his own people is quoted. This is not equivalent to the strict self-referential liar. If a Cretan says “All Cretans always lie,” the statement can simply be false because some Cretan sometimes tells the truth. A genuine contradiction requires additional assumptions about the speaker, the range of the generalization, and what counts as lying. “Epimenides paradox” is therefore a historical alternative name, not a guarantee that every Cretan-liar anecdote instantiates the formal problem.\n\nAristotle discussed a related difficulty in Sophistical Refutations: someone may appear to speak truly when saying that he speaks falsely. His treatment concerns qualifications of predication and the difference between speaking truly in a particular respect and doing so without qualification. Ancient discussions thus already raised an enduring question: does the apparent contradiction reveal a defect in inference, an equivocation in language, or a genuine obstacle to a general theory of truth?\n\nIn medieval scholasticism, liar-like arguments were studied under the heading insolubilia, “insolubles.” This became an organized logical literature rather than a collection of isolated riddles. Authors examined cases such as a person whose only utterance is that he is speaking falsely, often specifying carefully which utterances exist and what they signify. Thomas Bradwardine, writing in the fourteenth century, developed an account on which a sentence can signify more than its explicit wording initially suggests. His treatment connected a sentence’s truth with everything it signifies and argued that insoluble sentences acquire additional semantic commitments that block the simple alternation between truth and falsity.\n\nJean Buridan developed a different treatment connecting assertions with implications concerning their own truth. William Heytesbury’s Rules for Solving Sophisms, completed in 1335, also treated insolubles through carefully specified semantic conditions. These authors did not share one uniform “medieval solution.” Their disagreements concerned signification, the consequences of asserting a sentence, and whether the stipulated description of an insoluble case could be accepted unchanged. The sophistication of these debates anticipates later disputes about whether a liar sentence has only its apparent content and whether its semantic setting can remain fixed throughout the argument."
    },
    {
      "heading": "Formal self-reference and diagonalization",
      "text": "Ordinary formulations use expressions such as “this sentence,” which may invite questions about what exactly the demonstrative refers to. Formal versions show that the problem does not depend on an ambiguous pointing expression. A language capable of representing enough elementary arithmetic can encode its expressions as numbers and describe operations on those codes. Sentences can then indirectly refer to themselves through arithmetic relationships among their numerical representations.\n\nThe key construction is the diagonal lemma. Roughly, for any suitable formula \\(\\varphi(x)\\) with one free variable, an adequately expressive formal theory can construct a sentence \\(G\\) such that:\n\n\\[G\\leftrightarrow\\varphi(\\ulcorner G\\urcorner).\\]\n\nThe equivalence is provable in the underlying theory. The sentence need not literally contain its own name as a visible part. Instead, coding and substitution generate the required self-referential relationship. Choosing \\(\\varphi(x)\\) to be \\(\\neg T(x)\\) produces a formal liar sentence \\(L\\) satisfying:\n\n\\[L\\leftrightarrow\\neg T(\\ulcorner L\\urcorner).\\]\n\nIf the theory also admits the unrestricted truth instance\n\n\\[T(\\ulcorner L\\urcorner)\\leftrightarrow L,\\]\n\nthen ordinary classical reasoning yields a contradiction. The significance is structural: a sufficiently expressive language cannot consistently combine all the relevant self-reference, unrestricted disquotation, and classical logic. Merely prohibiting phrases such as “this sentence” does not remove the arithmetic construction.\n\nThis argument underlies Tarski’s undefinability theorem. In its standard arithmetic setting, the theorem establishes that the set of truths of arithmetic cannot be defined by an arithmetic formula within arithmetic itself. If an arithmetic formula defined truth for all arithmetic sentences in the intended interpretation, diagonalization would produce a sentence that holds exactly when it does not satisfy that definition. The theorem concerns the limits of internal definability, not the impossibility of discussing arithmetic truth in a richer language.\n\nThe same technique also illuminates the Gödel incompleteness theorems, but a Gödel sentence is not simply a liar sentence. Its characteristic content concerns its own unprovability in a specified formal theory, not its own untruth. Provability is tied to formally checkable proofs, whereas truth in the intended interpretation is a semantic notion. Under the relevant hypotheses, a Gödel construction establishes limitations on what a theory can prove rather than deriving a contradiction in that theory. Confusing truth with provability would erase precisely the distinction that allows incompleteness to reveal a limitation without producing inconsistency."
    },
    {
      "heading": "Tarski’s hierarchy of languages",
      "text": "Alfred Tarski’s work in the 1930s made the liar central to the systematic study of formal truth definitions. His account distinguishes an object language, whose sentences are being evaluated, from a metalanguage, in which their truth conditions are stated. A truth definition for the object language is formulated in the metalanguage, which must have resources sufficient to describe the object language’s expressions and the structures in which they are interpreted.\n\nFor example, a simple object language might contain arithmetic sentences but no truth predicate. A richer metalanguage can then define what it is for an arithmetic sentence to be true. The metalanguage can express instances such as “‘\\(0=0\\)’ is true if and only if \\(0=0\\),” without allowing the object language to construct a sentence that denies its own truth under that same predicate. The object language contains the arithmetic claim; the semantic evaluation belongs to the higher level.\n\nThe central constraint concerns semantic closure: roughly, the ability of a language to contain names of its own sentences, an applicable truth predicate, and the resources needed to express the relevant semantic reasoning internally. Tarski’s analysis shows why sufficiently strong, semantically closed languages are problematic when governed by unrestricted classical truth principles. A hierarchy blocks the contradiction by distinguishing predicates according to which levels of language they evaluate.\n\nSchematically, \\(T_0\\) applies to sentences of a base language \\(L_0\\). A richer language \\(L_1\\) contains \\(T_0\\), while \\(T_1\\), introduced at a further level, evaluates sentences of \\(L_1\\). A purported sentence that says of itself that it is not true under \\(T_0\\) cannot simultaneously be an eligible sentence of \\(L_0\\) if its construction already requires the higher-level predicate. The apparent liar either crosses a level distinction or falls outside the intended domain of the predicate. This is a restriction on semantic applicability, not a claim that the sentence is both classically meaningful and ordinarily false.\n\nThe approach provides rigorous truth definitions for many formal languages and preserves classical reasoning within the relevant levels. Its difficulty as an account of everyday speech is explanatory rather than a defect in those formal results. Natural language routinely uses the single word “true” to discuss arithmetic, journalism, other truth claims, and statements about language itself. Speakers do not normally mark levels. Hierarchical approaches must therefore explain whether ordinary usage conceals level restrictions, requires contextual interpretation, or lacks a single fully coherent global truth predicate. The formal hierarchy solves a precise consistency problem while leaving open how closely that architecture should model unrestricted linguistic practice."
    },
    {
      "heading": "Truth-value gaps and Kripke’s fixed points",
      "text": "A different family of approaches keeps a language containing its own truth predicate but permits some sentences to be neither true nor false. Such truth-value gaps need not mean that a hidden answer exists but is unknown. They can instead indicate that the semantic construction assigns neither truth nor falsity. Saul Kripke gave an influential formulation in “Outline of a Theory of Truth,” published in 1975, using partial interpretations and fixed points.\n\nThe construction begins with an interpretation of the nonsemantic vocabulary. Ordinary sentences such as “Snow is white” receive their values from that interpretation. The truth predicate initially has only a partial interpretation; in the minimal construction, it starts without sentences assigned to its positive or negative extension. At subsequent stages, a sentence is classified as true under the truth predicate when its content has already been evaluated as true, and false when its content has already been evaluated as false. The process continues through successive stages, including transfinite stages where necessary.\n\nA common implementation uses the strong Kleene evaluation rules. Negation exchanges truth and falsity but leaves an undefined value undefined. Some compounds nevertheless have definite values despite an undefined component: a disjunction with a true disjunct is true, and a conjunction with a false conjunct is false. This matters because semantic dependence is determined by the evaluation rules, not simply by whether a sentence contains the word “true.”\n\nAt a fixed point, repeating the evaluation procedure adds no new truth or falsity assignments. Sentences whose values become settled in the minimal construction are called grounded. A simple chain beginning with an ordinary true statement and repeatedly asserting the truth of the preceding statement becomes grounded. The liar does not. If \\(L\\) says that \\(L\\) is not true, an undefined value for its truth attribution remains undefined under negation. The minimal fixed point therefore does not assign \\(L\\) either classical truth value. The truth-teller also remains ungrounded there, although other fixed points can assign truth-tellers values.\n\nThe strengthened liar poses a further expressive problem. From outside the construction, it is possible to say that the liar is not in the extension of the truth predicate. But the object language’s negation of its truth attribution is not thereby true; it remains undefined. The metalinguistic description “not assigned true” and the object-language expression “not true” have different semantic roles. Adding an unrestricted predicate for undefinedness or groundedness can generate new diagonal difficulties. Kripke’s approach thus supplies a mathematically precise account of partial semantic determination, while not providing an unrestricted internal language for every distinction visible in its metatheory."
    },
    {
      "heading": "Truth-value gluts and paraconsistent logic",
      "text": "Where gap approaches deny that the liar must have a classical truth value, dialetheism allows that some contradictions are true. Applied to the liar, a dialetheic treatment can accept both the sentence’s truth and its falsity. This is a truth-value glut, contrasting with a gap. Graham Priest has been a prominent advocate of this approach, notably in In Contradiction, first published in 1987.\n\nAccepting a true contradiction requires a distinction between contradiction and triviality. In classical logic, the principle of explosion allows any conclusion to follow from \\(A\\) and \\(\\neg A\\). A familiar derivation first infers \\(A\\lor B\\) from \\(A\\), then uses \\(\\neg A\\) to infer the arbitrary \\(B\\) by disjunctive syllogism. If every conclusion followed from the liar’s contradiction, its acceptance would eliminate meaningful discrimination between claims.\n\nA paraconsistent logic therefore blocks explosion. Priest’s Logic of Paradox, usually abbreviated LP, can be presented with three semantic statuses: true only, false only, and both true and false. Truth only and both count as designated statuses for evaluating consequence. Negation exchanges true only with false only and leaves both unchanged. An inconsistent premise set need not entail a conclusion that is false only. In this framework, disjunctive syllogism is not unrestrictedly valid: \\(A\\) may be both true and false while \\(B\\) is false only.\n\nParaconsistency and dialetheism are not synonymous. Paraconsistency is a feature of a consequence relation: contradictions do not automatically imply everything. Dialetheism is a philosophical thesis that some contradictions are true. A paraconsistent system can also be used to reason with inconsistent records or conflicting information without endorsing any contradiction as true of reality. Conversely, a dialetheist needs an account of consequence that prevents accepted contradictions from trivializing the theory.\n\nThe broader challenge is to combine a nonexplosive logic with a sufficiently expressive truth theory. Simply declaring the liar both true and false does not settle the behavior of conditionals, quantifiers, semantic predicates, and general reasoning principles. Some combinations that safely accommodate the basic liar remain vulnerable to other paradoxes. In particular, Curry-style arguments can produce arbitrary conclusions without relying on an explicit pair \\(A,\\neg A\\). Dialetheic theories must therefore specify more than the treatment of negation. Their distinctive proposal is that consistency need not be a universal requirement on truth, while the resulting theory must still draw substantive boundaries around what follows from an accepted contradiction."
    },
    {
      "heading": "Revision sequences and contextual approaches",
      "text": "Revision theories treat truth as a concept whose application can be repeatedly reassessed rather than immediately assigned one permanently settled extension. Anil Gupta and Nuel Belnap developed this approach systematically, culminating in The Revision Theory of Truth in 1993. The starting point is a hypothesis about which sentences are true. The truth biconditionals then determine how that hypothesis should be revised.\n\nFor a simplified liar sentence, let \\(v_n(L)\\) indicate its assigned truth value at stage \\(n\\). Revision produces the pattern:\n\n\\[v_{n+1}(L)=1-v_n(L).\\]\n\nIf the initial hypothesis classifies the liar as true, the next classifies it as false; if it begins as false, the next classifies it as true. The oscillation continues. An ordinary grounded sentence behaves differently: its value stabilizes independently of arbitrary starting assumptions. A truth-teller can remain true or remain false according to its initial assignment. Liar sentences, truth-tellers, and ordinary assertions can therefore exhibit different patterns even when a simpler static classification groups some of them together.\n\nThis picture differs from a three-valued semantics. Each individual revision hypothesis may use ordinary two-valued assignments, while the overall semantic account studies what persists or changes across sequences. The liar’s alternation is not normally understood as an event unfolding in physical time, as though a printed sentence changed truth value every second. It models the dependence of one semantic assessment on another. Full revision theories must specify treatment of limit stages, where an infinite sequence has no immediately preceding assignment, and define the relevant forms of stability.\n\nContextual approaches locate the crucial variation elsewhere. They hold that the interpretation or range of a truth claim may shift with the context in which it is evaluated. Charles Parsons and Tyler Burge developed influential contextual treatments in the 1970s. On such views, a liar argument may appear to maintain one fixed truth predicate while its reasoning actually requires a change in semantic perspective. An assessment that a sentence is not true relative to one context need not contradict its truth assessment in a richer context.\n\nBoth families seek to explain why the liar’s alternating reasoning feels compelling without accepting a single unrestricted classical truth predicate with an inconsistent extension. Their mechanisms nevertheless differ. Revision theory analyzes sequences generated by semantic rules; contextualism analyzes the interpretation of utterances and the circumstances of assessment. Contextual accounts must constrain shifts rather than introduce them only when contradiction threatens. Revision accounts must explain which stable or unstable patterns support legitimate assertion and inference. Neither task is completed merely by observing that the liar can be reconsidered from another viewpoint."
    },
    {
      "heading": "Related paradoxes and the problem of revenge",
      "text": "A proposed treatment of the liar often introduces a new distinction: meaningful versus meaningless, grounded versus ungrounded, true versus assertible, or paradoxical versus nonparadoxical. A revenge paradox uses that distinction to construct another sentence. If the original liar is called meaningless, for example, a strengthened construction may assert that it is either false or meaningless. The question is whether the proposed classification can be expressed inside the same language without recreating the difficulty.\n\nRevenge is not a single theorem showing that all solutions fail. Its force depends on the exact expressive and inferential resources of the theory. A classification available only in a stronger metalanguage may not support an internal diagonal sentence. A predicate introduced without unrestricted disquotation may also behave differently from ordinary truth. Nevertheless, revenge arguments expose a central tension: a theory may coherently explain its semantic limitations from outside while being unable to express that explanation internally without qualification.\n\nCurry’s paradox sharpens the problem by reducing the role of explicit negation. Let \\(C\\) say, in effect, “If \\(C\\) is true, then \\(Q\\),” where \\(Q\\) is an arbitrary claim. With unrestricted truth equivalences, this yields:\n\n\\[C\\leftrightarrow(C\\rightarrow Q).\\]\n\nUnder familiar conditional reasoning, assume \\(C\\). The equivalence gives \\(C\\rightarrow Q\\), and hence \\(Q\\). Discharging the assumption yields \\(C\\rightarrow Q\\); the reverse direction of the equivalence then yields \\(C\\), and therefore \\(Q\\). Since \\(Q\\) was arbitrary, the result is triviality. The precise derivation depends on the conditional and structural rules involved, but the example shows why rejecting explosion alone is insufficient for an unrestricted truth theory.\n\nYablo’s paradox, published by Stephen Yablo in 1993, uses an infinite sequence \\(S_0,S_1,S_2,\\ldots\\), each saying that all later sentences are untrue. If some \\(S_n\\) is true, all its successors are untrue; but then a successor appears to say something true. If none is true, each sentence’s claim that all later ones are untrue appears correct. No sentence explicitly refers to itself or to an earlier sentence. Whether this completely eliminates self-reference depends on how the infinite semantic dependencies are analyzed.\n\nOther relatives reveal both similarities and differences. Two sentences can form a liar loop if one says that the other is true and the other says that the first is false. Russell’s paradox concerns membership in the collection of all collections that do not contain themselves, rather than a truth predicate. These cases share a negative diagonal structure, but restrictions appropriate to set formation do not automatically provide an account of linguistic truth. The broader family makes clear that the basic liar is one expression of a structural difficulty, not merely an unfortunate English sentence."
    },
    {
      "heading": "Axiomatic truth and the limits of formal theories",
      "text": "Beyond particular semantic models, the liar motivates the study of axiomatic theories of truth. These theories extend a mathematical base, often Peano arithmetic, with a truth predicate and explicitly stated principles governing it. Instead of assuming that every intuitive truth principle can be accepted together, they ask which combinations are consistent, which consequences they have, and how much mathematical strength they add.\n\nAn important distinction separates compositional principles from unrestricted self-application. A compositional truth theory specifies how the truth of a compound sentence depends on its parts. Truth of a conjunction corresponds to truth of both conjuncts; truth of a negation corresponds to failure of truth of the negated sentence; quantified statements require suitable conditions concerning instances. When these clauses apply only to sentences of a base language without the added truth predicate, they avoid the unrestricted liar construction. They can still provide a unified way to discuss arbitrarily complex sentences of that base language.\n\nTruth predicates are useful because some generalizations cannot be conveniently replaced by a single assertion in the original language. “Every theorem of this theory is true” expresses a uniform semantic claim about a potentially infinite collection of sentences. Merely listing individual theorems does not express the same generalization. This explains why eliminating the word “true” from each isolated example does not remove the demand for truth predicates. The challenge is to preserve their generalizing function without granting unrestricted access to paradox-generating instances.\n\nFormal theories also differ in whether they are conservative over their mathematical base. A conservative extension proves no new sentences expressed solely in the original language. Some carefully restricted truth theories have this property; stronger truth principles, especially when combined with expanded induction or reflection principles, can yield new mathematical consequences. A reflection principle connects what a theory proves with what is true or otherwise warranted. Its strength depends on its form and scope, so adding “truth” is not automatically either mathematically harmless or overwhelmingly strong.\n\nThese investigations preserve a distinction between a theory and the metatheory used to study it. A consistency proof, model construction, or classification of semantic values is carried out with specified external resources. That does not by itself supply the studied theory with unrestricted access to the same reasoning about itself. The liar places this distinction under particular pressure because the target concept is truth itself: a theory intended to describe all correct assertions is naturally asked to describe its own. Formal research makes that demand precise by identifying which self-applications, truth clauses, and inference rules can coexist, and which combinations force contradiction or stronger limitations on expressibility."
    }
  ],
  "links": [
    "truth",
    "logic",
    "law-of-excluded-middle",
    "diogenes-laertius",
    "chrysippus",
    "new-testament",
    "aristotle",
    "scholasticism",
    "classical-logic",
    "godels-incompleteness-theorems",
    "saul-kripke",
    "disjunctive-syllogism",
    "paraconsistent-logic",
    "russells-paradox"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}