Supervaluationism

Supervaluationism is a semantic theory for vague predicates that preserves classical logic while accommodating the phenomenon of borderline cases. The theory was developed by Kit Fine in 1975 and independently by Bas van Fraassen, though its roots trace to early twentieth-century discussions of the sorites paradox and the limits of classical bivalence.

The central idea is deceptively simple. A vague predicate like 'tall' or 'heap' does not have a single classical extension — a single set of things that satisfy it and a single set that do not. Instead, it has many precisifications — ways of making it precise that are all admissible, because the vague predicate does not itself determine which of the borderline cases belong to its extension. Each precisification assigns every object a definite truth-value. The predicate is super-true (true on all admissible precisifications) if it is true under every way of making it precise. It is super-false (false on all admissible precisifications) if it is false under every way of making it precise. Borderline cases are those where some precisifications make the predicate true and others make it false.

The payoff is immediate: classical logic is preserved at the level of super-truth. The law of excluded middle holds — every statement is either super-true or not super-true — because every precisification assigns a definite truth-value. The sorites argument fails because its inductive premise ('if n grains is not a heap, then n+1 grains is not a heap') is not super-true: there are precisifications where it is false, namely those that place the cutoff between n and n+1. The paradox dissolves not by rejecting classical logic but by rejecting the assumption that the sorites premise expresses a determinate truth.

The System of Precisifications

The formal structure of supervaluationism is a model-theoretic system. A supervaluation model consists of a set of classical models — the precisifications — constrained by a relation of admissibility. The constraints are not arbitrary: they must respect the facts that are definitely true. If it is definitely true that 10,000 grains of sand is a heap, then every admissible precisification must make '10,000 grains is a heap' true. If it is definitely true that 1 grain is not a heap, then every precisification must make '1 grain is not a heap' true. The borderline cases are the range where the precisifications disagree.

The set of admissible precisifications can be thought of as a partial order or a lattice, where each precisification is a point and the ordering reflects inclusion of truth-sets. The supervaluationist semantics is then a valuation over the lattice, determining super-truth by universal quantification over the lattice's points. This is the same mathematical structure that appears in possible worlds semantics, in Kripke semantics for modal logic, and in domain theory — a family of models, a notion of truth at each model, and a global notion of truth defined by aggregation across the family.

The structural parallel is not accidental. Supervaluationism is a multi-model semantics — it evaluates truth relative to a space of interpretations rather than a single interpretation. The space encodes the information that is determinate (the constraints shared by all precisifications) and the information that is indeterminate (the range of disagreement). This makes supervaluationism a formal model of partial information: a system that can represent what is known, what is unknown, and what is necessarily the case regardless of how the unknown is resolved.

The Logical Properties of Vagueness

Supervaluationism yields a distinctive logic for vague predicates. The key results:

Tautologies are preserved. Every classical tautology is super-valid — true on every admissible precisification in every model. This is because tautologies are true in every classical model, and super-truth is defined by universal quantification over classical models. Classical logic does not collapse; it is indexed to a level of determinacy.

The law of excluded middle holds. For any statement A, A ∨ ¬A is super-true. This is sometimes presented as a triumph of supervaluationism: it avoids the revisionism of fuzzy logic or intuitionistic logic, which reject excluded middle. But the triumph is technical, not philosophical. A ∨ ¬A is super-true even when A is borderline, but this does not mean that either A is true or A is false. It means that every precisification makes one of them true — which is trivially the case, since every precisification is classical. The supervaluationist can say 'A or not-A' is true without being able to say which disjunct is true. This is not classical logic as the classical logician understands it. It is classical logic with a gap between truth and determinacy.

No sharp boundaries are validated. A crucial feature of supervaluationism is that the statement 'there is a sharp boundary' — formally, ∃n (Heap(n) ∧ ¬Heap(n+1)) — is never super-true. This is because for any n, there are precisifications that make Heap(n) true and precisifications that make it false, and no single precisification makes both Heap(n) and ¬Heap(n+1) true for the same n. The supervaluationist can therefore claim that vagueness does not entail sharp boundaries: the theory validates the intuition that there is no exact number of grains where heaphood begins.

Criticisms and Responses

The most persistent criticism of supervaluationism is the epistemic objection: the theory does not explain why we cannot know where the boundary lies. If there are many admissible precisifications, and each one has a sharp boundary, then the boundary exists — it is just that we do not know which boundary is the correct one. But the supervaluationist denies that any precisification is 'correct' in the sense of being the uniquely intended interpretation. The vagueness of 'heap' is not a matter of epistemic ignorance. It is a matter of semantic indeterminacy: the predicate itself does not determine a unique extension.

This response is metaphysically contentious. The epistemicist — most notably Timothy Williamson — argues that vagueness is indeed a form of ignorance: there is a precise fact of the matter about how many grains make a heap, but this fact is unknowable due to the margins of error in our perceptual and cognitive systems. The epistemicist preserves classical logic and bivalence at the object level, paying the price of a counterintuitive epistemology. The supervaluationist preserves classical logic at the meta-level (super-truth), paying the price of a semantic indeterminacy that some philosophers find just as counterintuitive.

A second criticism concerns higher-order vagueness. The distinction between definitely true, borderline, and definitely false is itself vague. There are borderline cases of borderline cases: cases where it is unclear whether a case is borderline. Supervaluationism can be extended to handle this by introducing a hierarchy of precisifications — precisifications of the precisification relation itself — but the hierarchy raises the same questions about termination and precision that motivate the theory in the first place. The supervaluationist's response is that the hierarchy is not viciously regressive: each level is a well-defined mathematical structure, and the higher-order vagueness is managed by the same mechanism at each level. But the critics reply that this is an infinite regress dressed in formal clothing, not a solution.

Supervaluationism and Classification Systems

From a systems perspective, supervaluationism is a model of classification under uncertainty — a formal description of what it means for a category system to have boundaries that are not sharp but are not arbitrary either. The precisification model is a representation of a system that admits multiple consistent classifications, where consistency is determined by shared constraints and disagreement is localized to regions where the underlying concept is genuinely indeterminate.

This structure appears throughout systems that must classify continuous phenomena into discrete categories: medical diagnosis (where does 'high blood pressure' begin?), legal standards (what counts as 'reasonable doubt'?), engineering tolerances (when is a component 'defective'?), and machine learning classification (what is the decision boundary of a classifier?). In each case, the system faces a version of the sorites problem: the underlying variable is continuous, the category is discrete, and the boundary between them is not naturally determined. Supervaluationism does not solve the problem by finding the boundary. It solves it by making explicit that the boundary is not part of the system's determinate content — that the classification is partial, and the partiality is a feature, not a bug.

The supervaluationist insight for systems design: a classification system that does not specify every boundary is not necessarily a flawed system. It may be a system that correctly represents the indeterminacy of the domain it classifies. The task is not to eliminate the indeterminacy but to manage it — to specify what is determinate, to bound what is not, and to design the system's behavior so that its outputs are robust across the range of admissible precisifications. This is supervaluation as engineering: the art of building systems that work correctly even when their concepts are vague.

Supervaluationism is not a semantic theory about language. It is a theory about what it means for a system to be precise about what it knows and honest about what it does not. The precisification lattice is a map of epistemic boundaries — and the willingness to leave parts of the map blank is the hallmark of a system that knows its own limits.