Morphisms between gradability representations #
[Kam75] [Kle80] [Ken99] [Ken07] [Sco14] [BS22]
The maps between the three framework objects for gradable predicates,
with their faithfulness theorems — the degree-semantic analogue of the
representation maps in Phonology/Autosegmental (AR ↔ tone strings):
Klein (Delineation) — most general
↑ measureDelineation
Kennedy (measure functions) — specialization: single linear scale
↑ DimensionedMeasure.apply
Scontras / Bale & Schwarz (typed measurement)
Each map is an embedding (ordering_faithful,
measurement_refines_degree); the composite is measure_to_delineation.
Strictness: delineation expresses nonlinear adjectives ("clever") that
no degree function induces (delineation_strictly_more_general,
nonlinear_delineation_exists).
What each framework adds #
| Framework | Ontology | Comparative | Unique capacity |
|---|---|---|---|
| Klein | No degrees | ∃C. A(x,C) ∧ ¬A(y,C) | Nonlinear adjectives |
| Kennedy | Degrees (D,≤) | μ(x) > μ(y) | Measure phrases, DegP |
| Measurement | Degrees + dim | μ_d(x) > μ_d(y) | Typed dimensions, CARD |
Theorems in this file #
- measure_to_degree: every
DimensionedMeasureforgets to a plain degree function - degree_to_delineation: every degree function induces a Klein delineation
- ordering_faithful: the induced delineation's ordering = degree comparison
- degree_delineations_are_linear: all degree-induced delineations are linear
- nonlinear_delineation_exists: a concrete nonlinear delineation witness
- monotone_excludes_nonlinear: monotone delineations are never nonlinear
- delineation_strictly_more_general: delineation ⊋ degree (strict containment)
- nlDel_not_degree_representable: no degree function can induce the nonlinear witness
- nondistinct_iff_equal_measure: Klein's emergent degrees = actual degree equality
- degree_delineation_strict_weak_order: degree orderings are strict weak orders
- very_degree_chain: Klein's
very= two-step degree chain
Measurement → Degree → Delineation #
The maps themselves carry no new definitions: measurement forgets to a
bare degree function by the DimensionedMeasure.apply projection
([Sco14]'s insight that measure terms and CARD are one
degree-assigning operation), and any degree function μ over a linear
order induces a Klein delineation via Degree.Delineation.measureDelineation
— the embedding of measure-function degree semantics ([Ken99],
developed in [Ken07]) into [Kle80]'s framework. The embedding is faithful
(ordering_iff_degree: Klein's ordering under the induced delineation
is exactly degree comparison) and lands in the monotone, linear
fragment (measureDelineation_monotone, measureDelineation_is_linear).
Strict Separation: Delineation > Degree #
Klein's delineation framework is STRICTLY more general than degree semantics. The key witness: nonlinear adjectives like "clever" produce cyclic orderings (both a > b and b > a for different comparison classes). This is impossible for any degree-induced delineation, since degree orderings are asymmetric.
See `Studies/Klein1980.lean` for the empirical
motivation and the concrete "clever" witness. Here we prove the
theoretical separation at the framework level.
Monotone delineations cannot be nonlinear: monotonicity forces asymmetry, which excludes cycles. This is the core constraint that degree semantics imposes — and that Klein's framework relaxes.
The strict separation theorem: Klein's delineation framework is strictly more general than degree-based frameworks.
Forward: every degree function induces a monotone delineation
(measureDelineation_monotone).
Backward FAILS: there exist delineations (nonlinear ones) that no degree function can induce, because degree-induced delineations are always monotone, and monotonicity excludes nonlinearity.
This is the formal content of Klein's critique of degree semantics: multi-criteria adjectives like "clever" require the richer delineation framework.
Degree = Monotone Delineation (Characterization) #
The degree-based frameworks correspond EXACTLY to the monotone fragment of Klein's delineation theory. This is not a coincidence: monotonicity is what ensures a delineation induces a well-behaved ordering (strict weak order), which is exactly what a degree scale provides.
- Forward: degree → monotone delineation (`measureDelineation_monotone`)
- Backward: monotone delineation → degree-recoverable ([klein-1980] §4.2,
proved in `Klein1980.lean` as `kleinDegree_measureDelineation`)
Together: `degree semantics = monotone delineation semantics`.
Klein's full framework adds the non-monotone fragment for
multi-criteria adjectives.
Degree functions always yield monotone delineations AND the ordering is faithful. This characterizes exactly what degree semantics buys you within the delineation framework.
Measurement = Degree + Dimension Typing #
The relationship between measurement semantics ([Sco14], [BS22]) and degree semantics ([Ken07]) is simple: measurement adds typed dimensions to degree functions.
A `DimensionedMeasure E` is a degree function `apply : E → ℚ` PLUS a
`dimension : Dimension` label. The degree function is recoverable
via `DimensionedMeasure.toHasDegree`, but the dimension label is lost.
What dimension typing buys you:
- Multiple measure functions per entity (weight AND volume AND count)
- The No Division Hypothesis: compositional operations respect dimension types
- Measure term semantics: ⟦kilo⟧ = λn.λx. μ_kg(x) = n, typed to mass
What it does NOT buy you: any new ordering structure. Measurement
adjectives are still degree adjectives under the hood.
The degree construction ([Cre76] §4, [Bal08]) #
Degrees built from comparisons rather than assumed: [Cre76]
(4.1) quotients an arbitrary comparison relation φ by two-sided
φ-indistinguishability, and (4.2) shows the induced comparison on
classes is well-defined. On a preorder the construction coincides with
mathlib's Antisymmetrization (cresswellSetoid_le_iff). [Bal08]
then maps any finite scale into the universal scale Ω ≅ ℚ ∩ (0, 1] by
relative position (relativeRank), the homomorphism that licenses
indirect cross-scale comparison.
Two-sided indistinguishability under a comparison φ
([Cre76] (4.1)): same φ-profile on the left and right.
Equations
- Degree.cresswellSetoid φ = { r := fun (a b : E) => (∀ (c : E), φ a c ↔ φ b c) ∧ ∀ (c : E), φ c a ↔ φ c b, iseqv := ⋯ }
Instances For
Degrees of comparison as φ-equivalence classes ([Cre76] (4.1)).
Equations
- Degree.CresswellDegree φ = Quotient (Degree.cresswellSetoid φ)
Instances For
The induced comparison on degrees; well-definedness is [Cre76]'s own consistency proof for (4.2).
Equations
- Degree.CresswellDegree.rel = Quotient.lift₂ φ ⋯
Instances For
On a preorder, φ-indistinguishability under ≤ is mathlib's
AntisymmRel: the Cresswell quotient IS Antisymmetrization.
[Bal08]'s universal-degree homomorphism on a finite scale: the
relative position of d, valued in an order-isomorphic model of
Ω (the paper takes Ω to be isomorphic to ℚ ∩ [0, 1]; only the order
on the values is ever consumed — universal degrees are ordinal, not
arithmetic). Defined on whatever carrier plays the primary scale —
in Bale's regime the quotient, so equivalent individuals share a
universal degree by construction and the value counts equivalence
classes, not individuals.
Equations
- Degree.relativeRank d = ↑{x : D | x ≤ d}.card / ↑(Fintype.card D)
Instances For
The universal-degree map preserves the primary scale's order ([Bal08]: ℌ preserves ≥_δ).
Universal degrees land in (0, 1].
Transport: which operators are natural in the scale #
The functoriality table for degree operators under change of scale
representation (a StrictMono map between scales — precisely the
admissibleMeasure condition, so an admissible measure IS a
scale-morphism): comparatives, equatives, and the max-quantified
comparative are invariant; the positive form transports only if the
threshold rides along. This derives the classic observation that
comparatives are context-independent while the positive form needs a
contextually fixed standard: pos is the one non-natural operator.
Comparatives are invariant under change of scale representation.
Equatives are invariant under change of scale representation.
The max-quantified comparative is invariant under change of scale
representation. Not immediate: thanDegrees is a downset and images
of downsets need not be downsets, but the greatest element rides
along (f δ is greatest in the transported set, and conversely any
greatest transported degree is f of a witness measure).
The positive form transports only as a pair: rescaling the measure commutes with membership when the threshold is rescaled too.
With a fixed threshold the positive form is not natural: some strictly monotone rescaling changes the verdict. The one non-natural operator in the table — the formal face of the positive form's context-dependence.
Universal property of the degree construction: any φ-invariant map
factors through CresswellDegree φ — the quotient is the initial
scale a comparison relation determines.