The Analytic–Synthetic Distinction

The analytic–synthetic distinction is a proposed division between truths grounded in meaning and truths that also depend on how the world is. It became central to empiricism because it seemed to explain how logic and mathematics could be necessary and knowable a priori without describing a realm beyond experience. Quine's attack challenged whether “truth in virtue of meaning” marks any principled boundary at all.

This chapter examines both the distinction and that challenge. Its central question is whether analyticity supplies an explanation of truth or only a classification relative to a specified language, theory of meaning, or framework. Even if no absolute boundary survives Quine's objections, a more modest framework-relative distinction may remain precise and useful.

The issue has consequences beyond the classification of sentences. If a principled distinction exists, it may explain how understanding alone can justify some beliefs, why logic and mathematics differ from empirical science, and how changing a meaning differs from correcting a factual error. If no such distinction exists, these differences must instead be explained through inferential role, theoretical entrenchment, or framework-relative practice. The dispute therefore affects epistemology, theories of meaning, the foundations of mathematics, and the scope of empirical revision.

Background: the distinction

Kant called a judgment analytic when its predicate is already contained in its subject-concept: all bachelors are unmarried. Denying it would conflict with the concepts involved. A synthetic judgment adds something not contained in the subject-concept: all bachelors are unhappy. Its truth, if it is true, depends on more than meaning.

Later accounts broadened the distinction beyond subject–predicate form. On a common modern formulation:

• a sentence is analytic if it is true solely in virtue of meaning, perhaps together with logic;
• it is synthetic if its truth also depends on non-linguistic facts.

Frege treated analytic truths as those derivable from logic and definitions. Logical positivists such as Carnap used a wider notion: logical and mathematical truths, as well as explicit definitions, hold by linguistic convention, while synthetic claims are answerable to experience. This supported a sharp division between a framework's rules and the empirical claims made within it.

The distinction should not be identified with the a priori–a posteriori distinction. Analyticity concerns what makes a claim true; apriority concerns how it can be known. Nor is it simply the necessary–contingent distinction, which concerns how a claim holds across possibilities. Although these divisions were often aligned, Kripke's necessary a posteriori and contingent a priori cases showed that they are not equivalent.

Views on analyticity and a priori knowledge

The relation between the analytic–synthetic and a priori–a posteriori distinctions has been understood in several ways. Here empirical means a posteriori: justified at least partly through experience.

Kant: overlap without identity. Analytic judgments are knowable a priori because explicating a concept requires no observation. Some synthetic judgments are empirical, but Kant also defended the synthetic a priori: claims that extend knowledge while being knowable independently of experience. He placed arithmetic, geometry, and basic principles of natural science in this category.

Logical positivism: a near-collapse of the distinctions. Positivists rejected Kant's synthetic a priori. For them, a priori truths are analytic—true by logic or linguistic convention—while informative synthetic truths are empirical. This protected empiricism from apparently non-empirical knowledge of the world, but made the analytic–synthetic boundary carry much of its weight.

Quine: no principled boundary. Quine's holism challenges both divisions as traditionally drawn. Logical and mathematical claims differ from ordinary empirical claims mainly in how centrally they stand within a revisable web of belief, not because they possess a wholly different source of truth or justification. On this view, “a priori” may mark exceptional resistance to revision rather than independence from experience in principle.

Post-Quinean separation. Many philosophers retain one distinction without identifying it with the other. A sentence may be analytic because of its meaning yet require empirical knowledge to apply its terms correctly; conversely, defenders of rational intuition or constitutive principles allow synthetic a priori knowledge. Kripke's work reinforces the general lesson that semantic, epistemic, and modal classifications answer different questions, even though his central examples concern necessity rather than analyticity directly.

The main options are therefore: identify the a priori with the analytic, allow a synthetic a priori, reject both boundaries, or preserve them as distinct but interacting classifications. Much of the dispute turns on whether meaning can itself justify belief, and whether empirical revisability is compatible with genuine apriority.

Quine's objection

In “Two Dogmas of Empiricism” (1951), W. V. O. Quine attacked both analyticity and the reductionist picture of empirical confirmation that supported it.

His first objection is that proposed explanations of analyticity appear circular. Analytic truth is explained through synonymy: all bachelors are unmarried is analytic because bachelor means the same as unmarried man. But synonymy is then explained through analyticity, definition, or necessary interchangeability. Definitions normally record prior synonymies rather than create them, while interchangeability depends on intensional notions such as necessity that are no clearer than analyticity. Appeals to “semantic rules” identify which sentences a language treats as analytic but, Quine argues, do not explain what analyticity itself consists in.

His second objection is confirmation holism. Experience does not test most sentences one at a time. It bears on a network of assumptions—including observation claims, background theories, mathematics, and logic—whose members can in principle be retained or revised by adjusting other parts. Statements near the centre of the “web of belief” are harder to surrender, but no statement is absolutely immune to revision. The apparent boundary between truths of meaning and truths of fact is therefore, for Quine, a pragmatic difference of degree rather than a principled difference of kind.

The two objections reinforce one another. If individual sentences do not possess an independent stock of empirical content, it is unclear how some could be identified as having none. Quine does not merely claim that the boundary is vague; he doubts that “true solely in virtue of meaning” names a well-explained property at all.

Responses to Quine

Several responses preserve at least a limited distinction.

Ordinary linguistic competence. Grice and Strawson argue that speakers plainly distinguish changing a belief from changing a word's meaning. Someone who accepts that some bachelors are married has not made an unusual empirical discovery; they appear to use bachelor differently. Quine's demand for a non-circular reduction may be too strong: related semantic notions can be mutually illuminating without being reducible to non-semantic vocabulary.

Carnapian frameworks. Carnap's defenders treat analyticity as relative to a specified language or framework. Rules determine which expressions are synonymous and which sentences count as true under every admissible interpretation. The question is then not whether a sentence is analytic simpliciter, but whether it follows from the constitutive rules of a language. Quine may show that choosing and revising a framework is pragmatic, without showing that the internal distinction between rules and empirical claims is unreal.

Formal semantics. Model-theoretic and possible-worlds semantics can define analyticity as truth under every interpretation, or truth at every world compatible with the meanings of the expressions. This makes the notion precise once a semantic theory is given. Quine's deeper challenge remains: the formal machinery represents meanings but may not independently explain why one assignment, rather than another, gives the correct meanings.

Epistemic and metaphysical analyticity. Later philosophers distinguish a sentence's being justified merely by understanding it from its being made true by meaning. The epistemic notion may explain why competent speakers can reject a sentence only through confusion or semantic change, while avoiding the obscure claim that meanings themselves are truth-makers. This preserves a modest role for analyticity even if “truth by convention” is abandoned.

Holism without elimination. Revisability does not by itself erase the distinction. An analytic sentence may be surrendered by changing the language, whereas a synthetic sentence is normally surrendered while its meaning is held fixed. Likewise, evidence may confront theories holistically while some inferences remain constitutive of the concepts used in those theories. The difficult question is whether that contrast can be stated without presupposing the semantic notions Quine challenged.

The resulting consensus is mixed. Quine decisively undermined the idea of a wholly theory-neutral, unrevisable boundary and exposed the inadequacy of simple appeals to synonymy. He did not establish that no local, framework-relative, or epistemic notion of analyticity can be useful. The surviving distinction is therefore usually more modest than the positivists' foundational divide: it marks roles that sentences play within linguistic practices, not two permanently separated classes of truth.

Remark: Kant's $7+5=12$ example

Kant's standard example of the synthetic a priori is $7+5=12$. The concept the sum of seven and five, he argues, does not contain the concept twelve. To reach twelve we must perform a synthesis, perhaps by successively adding units in pure intuition. The result nevertheless holds universally and necessarily, so its justification is a priori rather than empirical.

The example receives different classifications because later writers changed what counts as “analytic”:

A contemporary mathematician or logician would normally formalize the claim rather than classify it using Kant's vocabulary. In Peano arithmetic, with $x+0=x$ and $x+S(y)=S(x+y)$, calculation gives

$$7+5=7+S(S(S(S(S(0)))))=S(S(S(S(S(7)))))=12.$$

Thus the equation is a short deductive consequence of the definitions and axioms. Most mathematicians would call it a proved arithmetic identity and, if pressed epistemologically, a priori; they would not usually need to decide whether it is analytic or synthetic. Constructivists also accept it because the calculation supplies a finite construction.

The formal derivation does not by itself refute Kant. It shows that the equation follows from an explicit formal system, whereas Kant was asking how arithmetic can extend human cognition and how its concepts acquire application. If analyticity means predicate containment, his classification remains intelligible; if it means derivability from logic and definitions, the logicist classification is natural. The dispute therefore concerns not the truth of $7+5=12$, but the source of its justification and the standard by which analyticity is judged.

Remark: the axioms of Infinity and Choice

The axiom of Infinity asserts that an inductive set exists and thereby supplies the natural-number structure $\omega$. The axiom of Choice (AC) asserts that every set of nonempty sets has a choice function. Both belong to ZFC, but neither is a theorem of first-order logic or a mere definition. ZF without Infinity admits a universe of hereditarily finite sets, while AC is independent of ZF, assuming consistency. Their formal independence shows that they add content to weaker theories; it does not by itself determine whether that content is analytic, synthetic, a priori, or empirical.

Mathematicians usually make a more modest classification. Infinity and Choice are axioms of a specified theory; theorems are then proved relative to them. Infinity is normally left implicit because ordinary arithmetic and analysis already require infinite structures. Choice is more often tracked explicitly because it has striking nonconstructive consequences and because useful mathematics can be developed without its full strength. In constructive foundations, full AC is especially consequential: by Diaconescu's theorem, it implies the law of excluded middle.

The comparison exposes a limit of the arithmetic example. Once definitions and rules are fixed, $7+5=12$ is derivable; neither Infinity nor Choice is similarly forced by the weaker background theory. Calling them “analytic” therefore requires a framework-relative notion on which adopted axioms count as constitutive. Calling them synthetic a priori instead treats them as substantive, non-empirical claims about mathematical reality or the conditions of mathematical reasoning. Formal logic maps the alternatives and their consequences, but does not choose between those philosophical descriptions.

Remark: analyticity, semantics, and theories of meaning

Analyticity cannot be classified independently of a theory of meaning. To say that a sentence is true “in virtue of meaning” presupposes an account of what meanings are, how they determine truth-conditions, and what fixes an expression's meaning rather than some alternative. Different answers yield different versions of the analytic–synthetic distinction.

Formal semantics therefore provides a conditional test: once an interpretation or set of meaning postulates is supplied, it can determine which sentences hold throughout the relevant models. It does not by itself answer the metasemantic question of why those postulates give the expressions their actual meanings. “All bachelors are unmarried” is analytic only if bachelor is fixed as synonymous with unmarried man; a model can represent that constraint, but cannot establish it merely by assigning denotations.

This is where Quine's objection reaches beyond a technical complaint about definitions. If synonymy, semantic rules, and constitutive inference have no independent foundation, then semantics may describe a language without grounding a sharp analytic–synthetic boundary within it. Defenders reply that this demand is excessive: a theory of meaning may explain analyticity as part of a network of semantic notions rather than reduce all of them to non-semantic facts.

The resulting dependence runs both ways. A theory of meaning determines what could count as analytic, while judgments about analyticity test whether that theory distinguishes semantic competence from factual belief. The distinction is sharpest in explicitly stipulated formal languages, more contestable in natural language, and least informative when “analytic” means only “derivable after these axioms and rules have been adopted.”

Remark: do Quine and conventionalists disagree?

Yes, but their disagreement is easily misstated. Quine does not merely ask conventionalists to write down a dictionary, an axiomatic system, or a scientific theory. He asks what makes some of its rules meaning-constitutive rather than highly entrenched claims about the world. Simply labelling a sentence “analytic,” “a meaning postulate,” or “true by convention” restates the distinction; it does not explain it.

Do they disagree about what “analytic” means? At the level of an initial gloss, not greatly. Quine takes the traditional proposal to be that analytic sentences are true in virtue of meaning and independently of fact. He does not primarily offer a rival definition. Instead, he argues that the proposed gloss has not been given a non-circular explication: “meaning,” “synonymy,” “necessity,” and “semantic rule” are repeatedly used to explain one another.

The disagreement appears when the gloss is made precise. A conventionalist treats analyticity as a genuine status conferred by meanings or framework rules. Quine doubts that linguistic practice determines a principled class with that status; apparent cases may instead be sentences protected from revision because of their central role in theory. Thus they can agree on the ordinary intension of the word—roughly, truth due to meaning—while disagreeing over whether it has a clear criterion of application, a nonempty and determinate extension, or enough explanatory content to mark a real distinction.

This also explains why merely proposing a new definition can miss Quine's point. One may define “analytic-in-$L$” as “derivable from the stipulated rules of language $L$.” Quine need not dispute that technical predicate. He disputes the further identification of those stipulated rules with the meanings of ordinary expressions, and the claim that this formal division recovers a fundamental boundary between language and fact.

A conventionalist can respond at three different levels:

1. Formal stipulation. In an explicitly constructed language, rules can stipulate that bachelor abbreviates unmarried man, or that certain axioms govern a symbol. This gives a precise framework-relative notion of analyticity. Quine can accept the formal result while denying that it explains analyticity in a previously existing natural language. An axiomatic system determines consequences conditional on its axioms; it does not by itself show that accepting those axioms is part of meaning rather than substantive theory.
2. A theory of meaning. A use-theoretic, inferentialist, or conventionalist metasemantics can say that communal practices or constitutive inference rules fix meanings. This directly addresses the circularity objection only if it independently distinguishes rules that constitute a concept from beliefs speakers strongly protect. Quine's behavioural and holistic doubts then become objections to that theory, not proof that no such theory is possible.
3. Carnapian tolerance. Carnap need not offer a reductive explanation of analyticity in Quine's preferred terms. He can take linguistic frameworks and their rules as the practical starting point: internal truths are settled relative to those rules, while adopting a framework is assessed for simplicity, fruitfulness, and convenience. This rejects Quine's demand for a deeper, framework-independent boundary rather than satisfying it.

The substantive disagreement has two parts. First, Carnap and other conventionalists treat some rules as constitutive of a language, whereas Quine sees a continuous web in which any sentence can in principle be revised. Second, conventionalists distinguish changing a theory within a fixed language from changing the language or framework itself; Quine doubts that this contrast marks a difference of kind, since revisions can be redescribed either way.

Providing a scientific theory is therefore not enough to settle the issue. For Quine, the theory is tested as a whole and its logical, mathematical, and semantic-looking components differ mainly in centrality. For a conventionalist, some components instead supply the standards by which empirical claims inside the theory are formulated and tested. The same formal structure can support either reading; the dispute concerns the status of its rules.

The fairest conclusion is conditional. If a theory of meaning can non-circularly explain why particular rules constitute meanings, it answers Quine's first objection to analyticity. It must still answer confirmation holism: why constitutive rules form a principled boundary rather than the most entrenched region of theory. If the conventionalist instead treats framework-relative analyticity as primitive, the view remains coherent, but it does not refute Quine; it declines his demand for reduction and defends the distinction by its explanatory usefulness.

Remark: is Quinean revision another stipulation?

Quine's web of belief can look like conventionalism under another name. When experience conflicts with theory, logic alone does not dictate which sentence to abandon; we choose a revision that preserves as much of the system as possible. If that choice fixes which claims survive, perhaps “centrality” is merely a convention, and apparent analyticity is whatever we stipulate not to revise.

That reading goes too far. The web is not normally designed or adopted in a single act. It is an inherited theory shaped by observation, prediction, language learning, and earlier revisions. Recalcitrant experience constrains the web as a whole, while simplicity, conservatism, explanatory power, and fruitfulness guide where adjustments are made. These criteria underdetermine revision, but underdetermination is not unconstrained stipulation. A community may choose among empirically adequate revisions without thereby making every retained sentence true by convention.

There is nevertheless a conventional element. Choices about notation, definitions, inferential policy, and which theoretical virtues to prioritize affect the resulting web. Quine's point is not that convention disappears, but that its contribution cannot be isolated into a privileged class of sentences whose truth is wholly conventional. Convention and empirical content permeate the system rather than dividing it into two kinds.

Is the account circular? It would be circular if “central” meant only “analytic” and analyticity were then explained as centrality. Quine instead proposes at least partly independent, graded indicators: how many other commitments depend on a sentence, how disruptive its rejection would be, how readily speakers revise it, and what explanatory or predictive work it performs. Logical laws and elementary arithmetic are exceptionally central on these measures, which explains their resistance to revision without first classifying them as analytic.

The explanation is therefore eliminative rather than reductive. Quine does not define analyticity as high entrenchment, since a deeply entrenched physical principle would then become analytic. He explains why sentences appear analytic—stability, generality, inferential centrality, and the high cost of revision—without positing a distinct semantic property shared by them. Differences traditionally represented as analytic versus synthetic become differences of degree within theory.

This avoids the original circle only by declining to provide a theory of analyticity at all. It remains open to two objections:

• Explanatory incompleteness. Revision behaviour may show which beliefs are protected without explaining linguistic competence, synonymy, or why rejecting “all bachelors are unmarried” sounds like changing the subject rather than correcting a belief.
• Normative dependence. Measures such as simplicity and minimal disruption require standards of good reasoning. If those standards covertly rely on constitutive meanings or unrevisable logic, the alleged explanation may presuppose part of what it seeks to replace.

Quine thus offers more than a stipulation but less than a conventional theory of meaning. The web supplies a naturalistic and pragmatic explanation of entrenchment; it explains away analyticity only if entrenchment, inferential role, and revision practice account for all the phenomena that motivated analyticity. Conventionalists deny that they do: for them, the difference between revising a belief and changing a meaning is itself a fact that the web metaphor leaves unexplained.

Remark: framework-relative analyticity within a revisable web

A conventionalist can accept Quine's claim that every sentence is revisable while retaining a local notion of analyticity. The proposal distinguishes a snapshot of a framework from its revision through time:

• Relative to a framework $F$, some rules $C_F$ fix the use of its vocabulary. A sentence is analytic-in-$F$ when it follows from those constitutive rules.
• Ordinary revisions preserve $C_F$ and change claims made within the framework.
• A framework revision replaces $F$ with $F^{\prime }$, changing some constitutive rules. A sentence analytic-in-$F$ may then become synthetic, false, or no longer expressible in $F^{\prime }$.

On this account, revisability and analyticity answer different questions. Revisability asks whether a community could alter a commitment; analyticity asks what role that commitment plays given its present framework. A rule can be constitutive now without being unrevisable forever. Replacing it changes both the accepted sentences and, partly, the meanings with which later inquiry proceeds.

The axiom of Choice illustrates the relativity cleanly. AC is a theorem-level commitment inside ZFC, unavailable in bare ZF, and rejected in ZF $+\neg \mathrm{AC}$. If “analytic” means derivable from adopted framework rules, its status varies with the theory. Similarly, a Carnapian can treat geometrical principles as constitutive within one framework while allowing scientific pressures to motivate replacing that framework. Revision occurs, but at the level that fixes what counts as an internal consequence.

This proposal is compatible with part of Quine's picture. It accepts holistic revision, denies absolute unrevisability, and treats framework choice as responsive to pragmatic and empirical pressures. Quine can also accept the precise formal relation “derivable from $C_F$.” What he rejects is the further claim that the web itself determines a unique boundary between revisions within a framework and revisions of it. The same transition can often be described either as changing a belief while holding meaning fixed or as changing a constitutive rule and hence changing meaning.

The hybrid therefore yields two possible notions:

The conventionalist has nevertheless gained something important. Quine's observation that a status can change under revision does not by itself refute a status indexed to a framework; it refutes only absolute analyticity understood as immunity to every possible revision. To recover a stronger distinction, however, the conventionalist must explain why one way of partitioning the web into framework and internal theory is privileged. Without that account, analytic-in-$F$ remains a useful tool for describing a chosen system, not a Quine-independent division inherent in belief as such.

Conclusion: the burden on both sides

The dialectical burden therefore falls on both sides. Anyone who affirms the analytic–synthetic distinction cannot let “this is true because it is analytic” end the justification or explanation. That verdict relocates the question: analytic relative to which framework and constitutive rules, what fixes those rules, and why should they be understood as determining meaning rather than expressing substantive commitments? Analyticity can classify a sentence's status within an account of meaning, but it cannot substitute for that account.

Conversely, anyone who denies the distinction should acknowledge the legitimate local notion. Once a framework and its rules have been fixed, whether a sentence follows from them can often be established precisely, and speakers do distinguish conclusions drawn within a framework from changes to the framework itself. The denial should therefore target an absolute or framework-independent boundary, not the possibility of framework-relative analyticity or the usefulness of drawing the distinction in a specified setting. Affirmation without further explanation is incomplete; denial without this qualification overlooks a genuine and familiar practice.