Stratified Reference [Cha17] #
Champollion's unified stratified-reference property SR_{d,γ} and the
three specializations linglib uses. Named in the namespace-carried scheme:
the Stratified namespace supplies the "stratified" qualifier, so the
core property is Stratified.Reference and the specializations are
Stratified.DistributiveReference / SubintervalReference /
MeasurementReference.
Main definitions #
Reference— Champollion's unifiedSR_{d,γ}(defined Ch 4 §4.6, motivated Ch 5);ReferenceUnivis its universal closure∀x, P x → Reference d γ P x.DistributiveReference/…Univ— Stratified Distributive Reference (Ch 4 §4.6 distributivity-via-θ; Ch 6 atelicity reuse):d = θ,γ = Atom.RelationalDistributiveReference/…Univ— the relational-role form (R : Entity → Event → Prop), composing withVerb.denote; coincides with the functional form on a role's graph.SubintervalReference/…Univ— Stratified Subinterval Reference (Ch 4 §4.6 atelicity-via-τ; Ch 5 §5.4 motivation):d = τ,γ = proper subinterval.MeasurementReference/…Univ— [Cha17]'s stratified measurement reference (Def 61 in §4.6; final Defs 52–53 in Ch 7 §7.4);d = μ,γ = strict less-than(substance pseudopartitives: thirty liters of water, five feet of snow).
Granularity signature: binary, not unary #
Champollion's primary schema has g : β → Prop (unary); his [[of]]
lexical entry constructs g via γ(M, x) := λd. d < M(x), closing over
the outer entity. Linglib uncurries: Reference takes γ : β → β → Prop
directly with inner-then-outer convention. Equivalent
post-closure-elimination but lets all specializations be genuine
instances of Reference.
Relation to Krifka's CUM/QUA (Champollion §2.7.2) #
The naive bridge SubintervalReferenceUniv τ P ↔ CUM P does not hold,
in either direction, by Champollion's own design:
CUM P → SubintervalReferenceUniv τ Pis false. Champollion §2.7.2 adopts lexical cumulativity as a foundational assumption: every verb is cumulative ([[V]] = *[[V]]). Counterexample: reach the summit is telic but cumulative; lacks subinterval reference along τ.SubintervalReferenceUniv τ P → CUM Pis false. Subinterval reference is about subdividing into τ-smallerP-parts — it does not entail closure under sum. Counterexample (linglib):P := λe. e.runtime.length ≤ 1over dense time has subinterval reference (split each event in halves) but fails CUM (two unit-length disjoint events sum to length 2).
The contrast with Krifka [Kri98] that Champollion makes in Ch 6 is
stratified reference vs. divisive-reference, not vs. CUM. Champollion
retains CUM as a baseline holding of all VPs and replaces Krifka's
≤_τ-divisiveness diagnostic with the strictly weaker stratified-reference
diagnostic. This makes Aspect/Stratified.lean (this file) and
Aspect/Cumulativity.lean (sibling) genuinely independent: stratified
reference is the atelicity diagnostic; CUM is the NP→VP propagation
property.
TODO #
- The stativity opposition along τ (Champollion's fourth opposition) is
realized in linglib by
HasSubintervalPropinSemantics/Aspect/SubintervalProperty.leanrather than as a decomposition form. The(τ, point-granularity)form has no current consumer. Studies/Champollion2017.leancovers distributivity (Ch 6) + a basic subinterval-reference↔Vendler atelicity bridge (§ 5), but does NOT yet formalize Champollion's distinctive Ch 6 push carts all the way to the store for fifty minutes contrast — the empirical case where stratified reference and Krifka's divisive-reference diagnostic make divergent predictions.
References #
Stratified Reference: the core unified property from [Cha17] eq. (16), with the binary-granularity convention from eq. (17)'s γ-helper inlined.
Reference d γ P x holds iff x can be decomposed into P-parts y
whose d-images stand in relation γ to d x.
d : α → β— the dimension (thematic role θ, runtime τ, measure μ, ...)γ : β → β → Prop— the granularity relating inner (d y) to outer (d x). Uncurried form of Champollion's eq. (17) γ-helperγ(M, x) := λd. d < M(x).P : α → Prop— the predicate under scrutiny ("the Share")x : α— the entity being decomposed
Reference d γ P x = *{y : P(y) ∧ γ (d y) (d x)}(x).
Equations
- Semantics.Aspect.Stratified.Reference d γ P x = Mereology.AlgClosure (fun (y : α) => P y ∧ γ (d y) (d x)) x
Instances For
Universal Stratified Reference #
Universal Stratified Reference: every P-entity has stratified
reference. ReferenceUniv d γ P := ∀ x, P x → Reference d γ P x. When
this holds, P is "stratified" along d at granularity γ.
Equations
- Semantics.Aspect.Stratified.ReferenceUniv d γ P = ∀ (x : α), P x → Semantics.Aspect.Stratified.Reference d γ P x
Instances For
Atomic granularity (shared γ) #
Atomic granularity for dimensions where [PartialOrder β] is
available: the inner d-image is an Atom in β. Used by
DistributiveReference (dimension = θ thematic role; entities have a
partial-order instance via the entity lattice).
For dimensions without a PartialOrder instance — notably the
runtime dimension (NonemptyInterval Time) used by stativity — atomicity
is expressed dimension-natively (e.g., NonemptyInterval.IsPoint for
NonemptyInterval Time). The unification is at the Reference
parameter-space level: both express "γ = inner is atomic in the
dimension's natural sense" at different concrete instantiations.
Equations
- Semantics.Aspect.Stratified.AtomicGranularity inner _outer = Mereology.Atom inner
Instances For
Stratified Distributive Reference: dimension is a thematic role θ,
granularity is Atom on the inner image (the outer is unused —
atomicity is an absolute property). [Cha17] eq. (24).
Captures distributivity: "The boys each saw a movie" distributes over atomic agents.
Genuine instance of Reference with γ := AtomicGranularity.
Equations
Instances For
Universal distributive reference: every P-entity distributes along θ.
Equations
- Semantics.Aspect.Stratified.DistributiveReferenceUniv θ P = ∀ (x : α), P x → Semantics.Aspect.Stratified.DistributiveReference θ P x
Instances For
Relational distributive reference (neo-Davidsonian roles) #
Champollion's DistributiveReference takes a functional thematic role
θ : α → Entity (the unique role-filler). linglib's neo-Davidsonian roles
are relational — ArgumentStructure.ThematicRel = Entity → Event → Prop,
Agent(a, e) — with thematic uniqueness available as
ArgumentStructure.UP. RelationalDistributiveReference is the relational
form, so the distributivity property composes directly with a
ThematicFrame / Verb.denote without picking a role function. It
coincides with the functional form on a role's graph
(relationalDistributiveReference_graph).
Relational Stratified Distributive Reference: the role is a
neo-Davidsonian relation R : Entity → α → Prop. A stratum y counts
iff it has an atomic R-filler. Under thematic uniqueness
(ArgumentStructure.UP R) that filler is unique, recovering "the
R-filler of y is atomic".
Equations
- Semantics.Aspect.Stratified.RelationalDistributiveReference R P x = Mereology.AlgClosure (fun (y : α) => P y ∧ ∃ (a : Entity), R a y ∧ Mereology.Atom a) x
Instances For
Universal relational distributive reference: every P-element distributes along R.
Equations
Instances For
Relational distributive reference is monotone in the predicate.
Relational distributive reference along a functional role's graph
coincides with the functional DistributiveReference — the bridge
justifying the relational form as the faithful generalization of
[Cha17]'s distributive reference.
Proper-subinterval granularity: inner runtime is a proper subinterval
of outer runtime. The binary γ for subinterval reference.
Equations
- Semantics.Aspect.Stratified.SubintervalGranularity inner outer = (inner < outer)
Instances For
Stratified Subinterval Reference: dimension is τ (runtime),
granularity is proper-subinterval. SubintervalReference P e holds
iff e can be built from P-parts with runtimes properly included in
τ e. [Cha17] eq. (38).
Captures atelicity: predicates compatible with for-adverbials have subinterval reference. "John ran for an hour" → run has it.
Genuine instance of Reference with d := τ and
γ := SubintervalGranularity.
Equations
- Semantics.Aspect.Stratified.SubintervalReference P e = Semantics.Aspect.Stratified.Reference (fun (e' : Event Time) => e'.runtime) Semantics.Aspect.Stratified.SubintervalGranularity P e
Instances For
Universal subinterval reference: every P-event has it.
Equations
- Semantics.Aspect.Stratified.SubintervalReferenceUniv P = ∀ (e : Event Time), P e → Semantics.Aspect.Stratified.SubintervalReference P e
Instances For
Stratified Measurement Reference #
MeasurementReference is [Cha17]'s stratified
measurement reference (his named property and abbreviation): Def 61
in the Ch 4 §4.6 unification, with final universal/restricted forms
Defs 52–53 in Ch 7 §7.4, written as SR_{μ, λd.d < μ(x)} (thirty
liters of water, five feet of snow, two degrees Celsius of global
warming).
The strict-less-than granularity is Champollion's faithful translation
of [schwarzschild-2006]'s monotonic measure-function predicate
(Ch 7 eq. 8: `μ` is monotonic iff `a < b → μ(a) < μ(b)`), with the
extensive/intensive measure-function distinction ([krifka-1998])
reduced to whether the resulting measurement-reference presupposition
is satisfiable on the given substance noun.
Stratified Measurement Reference: dimension is a measure function μ,
granularity is strict less-than on the scale.
MeasurementReference μ P x holds iff x can be decomposed into
P-parts with strictly smaller μ-values.
Genuine instance of Reference with γ := (· < ·).
Equations
- Semantics.Aspect.Stratified.MeasurementReference μ P x = Semantics.Aspect.Stratified.Reference μ (fun (x1 x2 : β) => x1 < x2) P x
Instances For
Universal measurement reference: every P-entity has it along μ.
Equations
- Semantics.Aspect.Stratified.MeasurementReferenceUniv μ P = ∀ (x : α), P x → Semantics.Aspect.Stratified.MeasurementReference μ P x
Instances For
Distributivity Constraint #
[Cha17] Ch 4 §4.6 Distributivity Constraint
(restated in Ch 7 §7.4 for the measurement chapter):
a distributive construction with Share S, Map M, granularity γ
describing entity x is acceptable iff SR_{M,γ}(S)(x). The same
constraint underlies adverbial-each, for-adverbials, and
pseudopartitives — they differ only in how M, γ, and S are set.
Equations
- Semantics.Aspect.Stratified.DistributivityConstraint Map gran Share x = Semantics.Aspect.Stratified.Reference Map gran Share x
Instances For
Construction Instances #
"each" distributes over atomic θ-fillers. Map = θ (thematic role), granularity = Atom (inner only).
Equations
- Semantics.Aspect.Stratified.eachConstr θ Share x = Semantics.Aspect.Stratified.DistributiveReference θ Share x
Instances For
"for"-adverbials require subinterval reference: the predicate must have stratified subinterval reference (atelicity). Map = τ, granularity = proper subinterval.
Equations
Instances For
Key Theorems #
ReferenceUniv entails Reference for any specific element.
Predicates have stratified reference for trivial granularity: every
P x is its own base-case stratum when γ is vacuously true. (No CUM
required — AlgClosure.base suffices.)
Distributive reference is monotone in the predicate.
Stratified reference is monotone in the predicate, dimension-
polymorphically. Generalizes distributiveReference_mono to any
dimension d and granularity γ.
Dimension-polymorphic substrate witness. Stratified reference with
reflexive granularity is satisfied by every P-element via the base
case. Quantifies over any d : α → β (no IsSumHom needed for this
direction, since the witness is structural).
The companion direction — closure under sums via IsSumHom d — is
reference_join below; together they establish that stratified
reference composes faithfully with the trace-function abstraction.
Stratified reference is closed under join when (i) the dimension is a
sum-homomorphism and (ii) the granularity is monotone in the outer
position w.r.t. ≤ on β. The substrate validation that the
trace-function abstraction ([IsSumHom d], applicable uniformly to τ,
σ, agentOf, patientOf, themeOf via the instances in Events/CEM.lean)
composes correctly with stratified reference.
The IsSumHom assumption ensures d (x ⊔ y) = d x ⊔ d y; the
monotonicity assumption on γ then carries the stratification witnesses
for x and y over to a witness for x ⊔ y.
Aspect Bridge (subinterval reference ↔ atelicity) #
for-adverbials require subinterval reference (Champollion Ch 5 §5.4). "John ran for an hour" is felicitous because "run" has it. "* John arrived for an hour" is infelicitous because "arrive" lacks it.
QUA and subinterval reference are directly incompatible: if P(e) and
SubintervalReference P e hold, then P cannot be quantized. The
AlgClosure decomposition yields a base element a with P(a) and
a.runtime ⊂ e.runtime. Since a ≤ e (from the join structure) and
a ≠ e (proper subinterval is irreflexive), we get a < e, contradicting
QUA.
Direct, not routed through CUM: the would-be route
SubintervalReferenceUniv → CUM → ¬QUA fails at the first step
(SubintervalReferenceUniv → CUM is false in general; counterexample:
P := λe. e.runtime.length ≤ 1 over dense time). See module docstring
"Relation to Krifka's CUM/QUA".
for-Adverbial Compatibility #
The "for"-adverbial adds a duration constraint on the event runtime and requires the predicate to have subinterval reference ([Cha17]'s for-adverbial entry, eq. (72), restated for for an hour as eq. (21); eq. (39) is the constraint on its Share). "V for δ" = λe. V(e) ∧ τ(e) = δ ∧ SubintervalReference V e.
Equations
- Semantics.Aspect.Stratified.forAdverbialMeaning V duration e = (V e ∧ e.runtime = duration ∧ Semantics.Aspect.Stratified.SubintervalReference V e)
Instances For
"in"-adverbials are incompatible with subinterval reference (they require telicity). "V in δ" requires QUA, which is incompatible with subinterval reference. Any P-event with subinterval reference has a strict P-part, contradicting QUA.