Documentation

Linglib.Phenomena.Polysemy.Studies.Sutton2024

Sutton (2024) — Common Nouns as Predicates vs. Types #

@cite{sutton-2024} @cite{chatzikyriakidis-luo-2020} @cite{cooper-2023} @cite{luo-2012} @cite{pustejovsky-1995}

@cite{sutton-2024} surveys the choice between Montagovian predicate denotations (man : Entity → Bool) and TTR/MTT type denotations (Man : Type, with a : Man witnessing manhood). The choice is not a notational variant: it determines whether selectional restriction violations are type errors or merely false, and whether co-extensional predicates can be intensionally distinguished.

This study file owns both halves of the argument:

This file consolidates the former TypeTheoretic/CNsAsTypes.lean (§§1–4) and TypeTheoretic/Selectional.lean (§§5–6), absorbed here under their shared Sutton-2024 anchor. Neither had any external consumer beyond this study's argument.

Lift a Montague predicate E → Bool to a TTR ptype E → Type. The resulting type family is inhabited at e iff p e = true.

Equations
Instances For
    @[implicit_reducible]
    instance Phenomena.Polysemy.Studies.Sutton2024.predToPtype_decidable {E : Type} (p : EBool) (e : E) :
    Decidable (Nonempty (predToPtype p e))

    Inhabitation of predToPtype p e is decidable (since p e = true is).

    Equations
    def Phenomena.Polysemy.Studies.Sutton2024.ptypeToPred {E : Type} (P : Semantics.TypeTheoretic.PredType E) [(e : E) → Decidable (Nonempty (P e))] :
    EBool

    Project a TTR ptype back to a Montague predicate. Requires decidable inhabitation (satisfied by any finite ptype).

    Equations
    Instances For
      theorem Phenomena.Polysemy.Studies.Sutton2024.montague_ttr_equiv {E : Type} (p : EBool) (e : E) :
      p e = true Nonempty (predToPtype p e)

      Bridge theorem: Montague predication and TTR type inhabitation agree when connected by predToPtype.

      p e = true ↔ Nonempty (predToPtype p e) — the predicate holds of e iff the corresponding ptype is inhabited (has a witness).

      theorem Phenomena.Polysemy.Studies.Sutton2024.roundtrip_pred {E : Type} (p : EBool) (e : E) :

      The round-trip ptypeToPredpredToPtype recovers the original predicate.

      TTR propExtension coincides with our bridge on lifted predicates.

      In Montague's STT, every entity has the same type e. A verb like "laugh" takes any entity and returns true or false. There is no structural distinction between a category mistake ("the chair laughed") and contingent falsity ("Mary didn't laugh").

      In TTR/MTT, verbs can require arguments of specific types. "Laugh" takes Animate, not Entity. A chair is not Animate, so the predication doesn't compose — it's not false, it's ill-typed.

      @cite{sutton-2024} §4: "In MTT [...] common nouns are interpreted as types [...] selectional restrictions are type requirements."

      Ontological sorts (TTR types for entity subclasses).

      Instances For

        In Montague, everything is an Entity. Chairs and people cohabit.

        Instances For

          In Montague, "the chair laughed" composes and evaluates to false. It's the same kind of thing as "Mary didn't laugh" — both are false.

          There's no structural way to distinguish category mistakes from contingent falsity — both are just false.

          theorem Phenomena.Polysemy.Studies.Sutton2024.SelectionalDemo.ttr_distinguishes_sorts :
          (∀ (a : Animate), ∃ (b : Bool), laughs_ttr a = b) ∀ (e : FlatEntity), ∃ (b : Bool), laughs_montague e = b

          In TTR, contingent falsity (Mary doesn't laugh) is well-typed but false. Category mistakes (chair laughs) don't typecheck at all. The selectional restriction prevents composition, rather than evaluating to false.

          Montague predicates are functions E → Bool. Two predicates with the same extension are definitionally equal by funext. TTR types carry intensional identity — two types can have the same witnesses yet remain distinct.

          @cite{sutton-2024} §3.1: "there is nothing which prevents two types from being associated with exactly the same set of objects."

          theorem Phenomena.Polysemy.Studies.Sutton2024.HyperDemo.montague_extensional {E : Type} (p q : EBool) (h : ∀ (e : E), p e = q e) :
          p = q

          Two Montague predicates with the same extension are equal.

          theorem Phenomena.Polysemy.Studies.Sutton2024.HyperDemo.montague_collapses_attitudes (p_groundhog p_woodchuck : SelectionalDemo.FlatEntityBool) :
          (∀ (e : SelectionalDemo.FlatEntity), p_groundhog e = p_woodchuck e)p_groundhog = p_woodchuck

          The consequence for attitude reports: Montague cannot distinguish "John believes a groundhog is outside" from "John believes a woodchuck is outside" — the belief content is the same function. TTR can.

          theorem Phenomena.Polysemy.Studies.Sutton2024.HyperDemo.ttr_preserves_attitudes :
          have groundhog := { carrier := Unit, name := "groundhog" }; have woodchuck := { carrier := Unit, name := "woodchuck" }; groundhog.extEquiv woodchuck ¬groundhog.intEq woodchuck

          TTR preserves the distinction: groundhog ≠ woodchuck as ITypes, even when co-extensional. Attitude verbs can be sensitive to this.

          For basic extensional predication over a fixed domain without selectional restrictions or attitude contexts, Montague predicates and TTR ptypes are interchangeable. The bridge is predToPtype/ ptypeToPred, and the equivalence is montague_ttr_equiv.

          The approaches diverge exactly when:

          1. Selectional restrictions matter (§2)
          2. Hyperintensional contexts arise (§3)
          3. Copredication requires dot types (see Studies/Gotham2017.lean)

          These are precisely @cite{sutton-2024}'s arguments for rich type theories.

          theorem Phenomena.Polysemy.Studies.Sutton2024.equivalence_boundary {E : Type} (p : EBool) :
          (∀ (e : E), p e = true Nonempty (predToPtype p e)) ∀ (e : E), ptypeToPred (predToPtype p) e = p e

          Summary: the predicate/type duality is an equivalence modulo three phenomena. We can state the exact boundary.

          Makes SubtypeOf from TypeTheoretic/Core.lean load-bearing by wiring it into typed composition. Verbs declare the ontological sort they require; composition through restrict succeeds only when SubtypeOf A B is available. Removing a SubtypeOf instance breaks the downstream derivation — that is the "load" the instance bears.

          @[implicit_reducible]
          Equations
          Equations
          • One or more equations did not get rendered due to their size.
          Instances For
            @[implicit_reducible]
            Equations
            Equations
            • One or more equations did not get rendered due to their size.
            Instances For

              Animate: the union of Human and Animal. Domain of "laugh", "sleep".

              Instances For
                Equations
                • One or more equations did not get rendered due to their size.
                Instances For
                  Equations
                  • One or more equations did not get rendered due to their size.
                  Instances For
                    @[implicit_reducible]
                    Equations
                    Equations
                    • One or more equations did not get rendered due to their size.
                    Instances For
                      @[implicit_reducible]
                      Equations
                      Equations
                      • One or more equations did not get rendered due to their size.
                      Instances For

                        Top-level entity sort. Everything coerces here.

                        Instances For
                          Equations
                          • One or more equations did not get rendered due to their size.
                          Instances For
                            Equations
                            • One or more equations did not get rendered due to their size.
                            Instances For

                              Each instance is load-bearing: removing it breaks a named theorem below. The hierarchy:

                              HumanAnimateEntity
                              AnimalAnimate
                              PhysObjEntity
                              InfoEntity
                              

                              Restrict a property to a subtype domain. Given P : Ppty B and [SubtypeOf A B], produce Ppty A by coercing the argument up before applying P.

                              This is typed function application: the semantic analogue of checking that the argument type is a subtype of the parameter type. If no SubtypeOf instance exists, composition fails at the Lean type level — a category mistake, not a truth-value failure.

                              Equations
                              Instances For

                                "laugh" selects for Animate.

                                Equations
                                • One or more equations did not get rendered due to their size.
                                Instances For

                                  Each theorem exercises a specific SubtypeOf instance. Removing the instance makes the theorem fail to typecheck — the instance literally bears the load of the derivation.

                                  Compositional derivations with typed NPs #

                                  "john laughs" = ⟦john⟧(restrict ⟦laugh⟧) Uses instHumanAnimate to compose Human quantifier with Animate VP.

                                  Equations
                                  • One or more equations did not get rendered due to their size.
                                  Instances For

                                    "fido laughs" = ⟦fido⟧(restrict ⟦laugh⟧) Uses instAnimalAnimate.

                                    Equations
                                    • One or more equations did not get rendered due to their size.
                                    Instances For

                                      "a human who thinks laughs" = ⟦a human who thinks⟧(restrict ⟦laugh⟧) Uses instHumanAnimate. The ExistWitness requires a Human who both thinks (restrictor) and laughs (scope, via coercion).

                                      Equations
                                      • One or more equations did not get rendered due to their size.
                                      Instances For

                                        Direct Human ⊑ Entity and chained Human ⊑ Animate ⊑ Entity produce the same result. This is the content of the coherence requirement on subtyping: the diagram commutes.

                                        Unlike SubtypeOf (identity-preserving: a human IS an animate entity), complement coercion is meaning-changing: the coerced entity is a DIFFERENT thing. "John enjoyed the book" doesn't mean John enjoyed a physical object — it means he enjoyed reading it. The book is coerced to an event via the telic quale.

                                        This is a distinct mechanism from subtyping, and conflating them would lose the meaning change.

                                        Equations
                                        • One or more equations did not get rendered due to their size.
                                        Instances For
                                          Equations
                                          • One or more equations did not get rendered due to their size.
                                          Instances For

                                            Complement coercion: a meaning-changing type shift triggered by type mismatch. The coerce function produces a new entity in the target type, not a view of the old one.

                                            • coerce : AB
                                            Instances

                                              restrict ≠ coerceApply #

                                              The structural distinction: restrict composes with SubtypeOf.up (the argument IS-A member of the supertype), while coerceApply composes with ComplementCoercion.coerce (the argument is MAPPED TO a different entity). Both produce Ppty A from Ppty B, but the semantic relationship is different.

                                              Consequence: "John fell" (restrict: John is an entity) and "John enjoyed the book" (coerceApply: the book became a reading event) are composed by different mechanisms, and the type system keeps them apart.

                                              coerceApply changes identity: coercion is a type-shift.