Kamp (1975): Two Theories about Adjectives @cite{kamp-1975} #
In E. Keenan (ed.), Formal Semantics of Natural Languages, 123–155. Cambridge University Press.
Overview #
Kamp presents two theories of adjective semantics:
Theory 1 (§ 1–2): Adjectives as functions from properties to
properties (type ⟨⟨e,t⟩,⟨e,t⟩⟩). The classification hierarchy —
intersective, subsective, privative, extensional — is formalized in
Theories/Semantics/Lexical/Adjective/Classification.lean.
Theory 2 (§ 3–7): Vague/graded models. Kamp introduces vague
models ⟨M, S, F, p⟩ (partial model + completions + σ-field +
probability measure) and derives the comparative from quantification
over completions. This framework is the common ancestor of both
@cite{fine-1975}'s supervaluationism and @cite{klein-1980}'s delineation
approach. Theory 2 is not formalized here; § 2 captures only the
motivating argument (why truth-functional many-valued logic fails)
and § 3 formalizes the comparative definitions that descend from it.
Structure #
- § 1: Single-world specialization of
Classification.lean's hierarchy - § 2: Many-valued logic failure (motivation for Theory 2)
- § 3: Kamp → Klein lineage:
kampAtLeastAs↔kleinMoreThan - § 4: Concrete witnesses for each hierarchy class
Key Insight #
Kamp argues (p. 233 of the Brill reprint) that truth-functional
many-valued logic fails for natural language connectives: if
⟦φ⟧ = ½, then ⟦φ ∧ ¬φ⟧ should be 0 (contradictions are false),
but any truth-functional F(∧) satisfying F(∧)(½, ½) = 0 also
gives F(∧)(½, ½) = 0 for non-contradictory ⟦φ ∧ φ⟧. This
motivates the move to supervaluation / probability over completions.
Classification.lean defines the general intensional hierarchy:
isIntersective, isSubsective, isPrivative, isExtensional
over Property W E = W → E → Bool. The bridge theorems below show
that fixing a world reduces the intensional definitions to their
single-world extensional counterparts.
Single-world specialization: given a fixed world, the intensional hierarchy reduces to the extensional one.
If adj is intersective, then at any fixed world w, the function
N ↦ adj N w is intersective in the sense of Modification.lean
(there exists a fixed predicate Q(w) such that the result is
Q(w) ∩ N(w)).
Single-world specialization of subsective.
intersective_at_world and subsective_at_world show that fixing
a world reduces the intensional hierarchy to single-world properties:
∃ Q_w, adj(N)(w)(x) = Q_w(x) ∧ N(x) (intersective) and
adj(N)(w)(x) → N(x) (subsective).
@cite{kamp-1975} (p. 233) argues that truth-functional many-valued logic cannot adequately handle vague connectives. The key observation:
If `⟦φ⟧ = ½` (borderline), then `⟦¬φ⟧ = ½` (standard negation).
We want `⟦φ ∧ ¬φ⟧ = 0` (contradictions are false). But any
truth-functional `F(∧)` satisfying `F(∧)(½, ½) = 0` also gives
`F(∧)(½, ½) = 0` for the non-contradictory `φ ∧ φ`, since the
inputs are identical. This is wrong: `φ ∧ φ` should have the same
value as `φ`.
Strong Kleene logic (`Truth3.meet`) makes the symmetric choice:
`meet indet indet = indet`. This preserves `φ ∧ φ ≡ φ` but fails
to make contradictions false. Supervaluationism resolves both.
Strong Kleene conjunction of indet with itself is indet, not
false. This means φ ∧ φ is correctly handled (same as φ),
but φ ∧ ¬φ gets indet rather than the desired false.
Kamp's dilemma: Strong Kleene meet cannot distinguish φ ∧ φ
from φ ∧ ¬φ when φ is borderline, because both reduce to
meet indet indet.
Supervaluationism resolves this: φ ∧ ¬φ is super-false (false
on every precisification) while φ ∧ φ is indefinite (true on
some, false on others). See Fine1975.non_contradiction_superfalse.
Kamp's dilemma (the actual impossibility result): no
truth-functional binary operator can simultaneously be
idempotent (F(x,x) = x) and make borderline contradictions
false (F(½, ¬½) = 0).
Since ¬½ = ½ in any symmetric three-valued logic, the two
requirements conflict: idempotence demands F(½,½) = ½,
but the contradiction requirement demands F(½,½) = 0.
This is what motivates the move to supervaluation.
@cite{kamp-1975}'s definition (12) for the comparative:
u₁ is at least as A as u₂ iff for every completion M' ∈ S where
u₂ is in the extension of A, u₁ is also in the extension.
@cite{klein-1980} rephrases this with comparison classes: u₁ is
more A than u₂ iff there exists a comparison class C where u₁ is
A-in-C but u₂ is not.
These are contrapositives: Kamp's "∀ completions, u₂ ∈ ext → u₁ ∈ ext"
is equivalent to Klein's "¬∃ completion where u₂ ∈ ext ∧ u₁ ∉ ext",
and Klein's strict comparative adds the asymmetric witness.
Kamp's definition (12): u₁ is at least as A as u₂ iff every context that puts u₂ in the extension also puts u₁ in the extension. Parameterized by a set of "completions" (Kamp) or "comparison classes" (Klein).
Equations
Instances For
Klein's strict comparative: there exists a context that separates
the two entities. This is comparativeSem from
Theories/Semantics/Comparison/Delineation.lean.
Equations
Instances For
Kamp–Klein bridge: Klein's strict comparative is equivalent to
Kamp's "at least as" in one direction but not the other. Precisely:
kleinMoreThan u₁ u₂ implies ¬kampAtLeastAs u₂ u₁ (if u₁ is
strictly more A than u₂, then u₂ is NOT at least as A as u₁).
Kamp's strict comparative (asymmetric part of "at least as") implies Klein's: if u₁ is at least as A as u₂ but not vice versa, then there exists a separating context.
Each class in the hierarchy is non-empty. We construct explicit
adjective denotations that provably satisfy each definition from
Classification.lean, modeling the classic examples: "gray"
(intersective), "fake" (privative), "skillful" (subsective but not
extensional), and "alleged" (non-subsective/modal).
These are the formal counterparts of the informal entailment judgments
from the literature (e.g., "gray cat entails cat" ↔ `isSubsective`,
"skillful surgeon + violinist ⊬ skillful violinist" ↔ `¬isExtensional`).
@cite{partee-2010} argues that the privative class should be eliminated
in favor of subsective + noun coercion. The witness `fakeAdj` below
models the traditional analysis; see `Partee2010.lean` for the
coercion reanalysis.
Two worlds suffice to distinguish extensional from non-extensional.
Instances For
"gray": an intersective adjective. The extension of "gray N" is
{x | gray(x)} ∩ {x | N(w)(x)} — a fixed property independent of
the noun.
Models @cite{kamp-1975} definition (4). Entailment pattern: "gray cat" entails both "gray" and "cat"; "gray" + "cat" entails "gray cat".
Equations
Instances For
"fake": a privative adjective (traditional analysis). "Fake N" entities are never N.
Models @cite{kamp-1975} definition (5). Entailment pattern: "fake gun" entails "not a gun".
@cite{partee-2010} argues this class should be reanalyzed as
subsective with noun coercion — see Partee2010.lean.
Equations
Instances For
"skillful": a subsective adjective that is NOT extensional. Being a "skillful N" depends on N's intension — what counts as an N across worlds — not just who the N's are in this world.
Models @cite{kamp-1975} definition (6) without definition (7). Entailment pattern: "skillful surgeon" entails "surgeon" (subsective), but "skillful surgeon" + "violinist" does not entail "skillful violinist" (not intersective, because not extensional).
Equations
- One or more equations did not get rendered due to their size.
- Phenomena.Gradability.Studies.Kamp1975.skillfulAdj N w x = (N w x && false)
Instances For
"alleged": a non-subsective (modal) adjective. An "alleged N" may or may not be an N — the adjective creates an intensional context without entailing or anti-entailing the noun.
This is the complement class in the hierarchy: adjectives like "alleged", "potential", "putative" that carry no meaning postulate constraining the relationship between the modified and unmodified extension.
Equations
Instances For
"alleged N" does not entail "N" (not subsective).
"alleged N" does not entail "not N" (not privative).