Documentation

Linglib.Phenomena.Generics.Studies.AsherPelletier2013

@cite{asher-pelletier-2012} — More Truths about Generic Truth #

Nicholas Asher and Francis Jeffry Pelletier, ch. 12 of Genericity (Mari, Beyssade, Del Prete eds.), OUP, Oxford Studies in Theoretical Linguistics 43.

Core Claim #

Generics should be analyzed as modal quantifiers — ∀x(φ(x) > ψ(x)), where > is a weak conditional (from Asher & Morreau 1991). The key innovation is per-individual evaluation (§12.3): ψ is evaluated for each individual a only in those worlds where a is a normal φ.

"Birds fly" is true because for each bird a, the most normal a-bird worlds are ones where a flies. For Opus the penguin, the normal Opus-penguin worlds are NOT normal Opus-bird worlds, so "Penguins don't fly" correctly overrides "Birds fly" for penguins (exx. 7–8).

Chapter Sections Covered #

Connection to Existing Infrastructure #

  1. NormalityOrder (Core/Order/Normality.lean): preorder structure on worlds. A&P's per-individual normality is a normality ordering per entity and restrictor class.

  2. traditionalGEN (Lexical/Noun/Kind/Generics.lean): GEN's normal parameter is the Bool-level projection of a normality ordering.

  3. TweetyNixon (Phenomena/DefaultReasoning/TweetyNixon.lean): the Tweety Triangle data — the classic test case for A&P's system.

Refinement vs Specificity #

processDefault below uses NormalityOrder.refine (@cite{veltman-1996}'s operation), which intersects ordering constraints. For the Tweety Triangle, Veltman's refinement produces incomparability between penguinFlies and penguinNoFly (neither is ≤ the other, since each satisfies one default and violates the other). A&P's per-individual evaluation resolves this via specificity: the more specific "penguins don't fly" overrides "birds fly" for penguins. The tweetyLe ordering encodes this specificity-resolved result directly.

A default rule: "Normally, if restrictor then scope."

Generic sentences express default rules. "Birds fly" means "Normally, if x is a bird, then x flies."

  • restrictor : WProp
  • scope : WProp
Instances For

    Process a default rule as a refinement of a normality ordering.

    "Normally, if P then Q" refines the ordering to promote P∧Q worlds over P∧¬Q worlds via NormalityOrder.refine. This is @cite{veltman-1996}'s monotonic (intersection-based) operation.

    Note: A&P's actual system uses per-individual evaluation with specificity, which goes beyond simple refinement. See the module docstring caveat about refinement vs specificity.

    Equations
    Instances For

      Per-individual evaluation data from §12.3.

      The chapter's key innovation: for ∀x(φ(x) > ψ(x)), the consequent ψ is evaluated per-individual. For individual a in the domain, we look at normal a-φ worlds — worlds where a is a normal φ.

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

          Ex. (7): "Penguins don't fly" — evaluated for Opus. Normal Opus-PENGUIN worlds: Opus doesn't fly. Scope (¬fly) holds.

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

            Ex. (8): "Birds fly" — evaluated for Opus. Normal Opus-BIRD worlds: in those worlds, Opus is "definitely not a normal penguin." So Opus flies. Scope (fly) holds.

            Key insight: both "Birds fly" and "Penguins don't fly" are true SIMULTANEOUSLY for the same individual — evaluated at different normal worlds (normal-bird vs normal-penguin worlds).

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

              Both generics are simultaneously true for penguins. Per-individual evaluation evaluates at DIFFERENT normal worlds for each restrictor, so both can hold without contradiction.

              Specificity determines which default to apply for inference: the more specific "penguins don't fly" overrides "birds fly" for penguins. This is a property of defeasible inference with the > conditional, not of the per-individual evaluation itself.

              §12.3, ex. (9): "Turtles live to be 100."

              The notion of a "normal φ(a) world" has some "give" to it — different construals of normality yield different truth values. Under an Aristotelian/teleological conception (natural telos of a turtle: if everything goes right), it's true. Under a statistical conception (most turtles die within hours of hatching), it's false.

              A&P consider this context-dependence a virtue: discourse and contextual factors fix the normality construal.

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

                      The same generic has different truth values under different normality construals — A&P's "slop" in the normality notion.

                      Default 2: "Penguins normally don't fly."

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

                        Veltman-style refinement of both defaults.

                        This uses NormalityOrder.refine to process both defaults. The result makes penguinNoFly and penguinFlies incomparable — neither is ≤ the other — because the two defaults create crossing constraints. (penguinFlies satisfies "birds fly" but violates "penguins don't fly"; penguinNoFly does the reverse.)

                        This matches the Nixon Diamond behavior (conflicting defaults → agnosticism), not the Tweety Triangle (specificity resolution). A&P's per-individual evaluation resolves this via specificity, encoded in tweetyLe below.

                        The order of processing doesn't matter (NormalityOrder.refine_comm).

                        Equations
                        • One or more equations did not get rendered due to their size.
                        Instances For
                          def Phenomena.Generics.Studies.AsherPelletier2013.normalFromOrdering {W : Type u_1} (le_dec : WWBool) (domain : List W) :
                          WBool

                          The normal parameter in traditionalGEN is the Bool-level projection of a normality ordering: normal(s) = true iff s is among the most normal elements.

                          This bridges the abstract ordering theory to the concrete GEN operator. Requires decidable le to compute Bool values.

                          Equations
                          Instances For
                            theorem Phenomena.Generics.Studies.AsherPelletier2013.total_all_normal {W : Type u_1} (domain : List W) (w : W) :
                            w domainnormalFromOrdering (fun (x x_1 : W) => true) domain w = true

                            When the ordering is total (initial state), every world is normal.

                            Challenge examples from §12.2 that the simple modal analysis ∀x(φ > ψ) appears to predict wrongly.

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

                                Ex. (4a): "Ducks lay eggs" — the basic analysis predicts normal male ducks lay eggs, which is wrong.

                                Equations
                                Instances For

                                  Ex. (4b): "Cardinals are bright red" — predicts normal female cardinals are bright red, which is wrong.

                                  Equations
                                  Instances For

                                    Ex. (5): "Mosquitoes carry the West Nile Virus" — true despite a vanishingly small percentage of normal mosquitoes carrying WNV.

                                    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

                                        A&P's arguments against @cite{cohen-1999a}'s probabilistic semantics for generics.

                                        Cohen proposes generic truth ↔ Pr(B(a)|A(a)) > 0.5 (the "Cohen conditional"). A&P argue this is inadequate for three reasons.

                                        • tooWeak : AntiProbArgument

                                          Too weak: Pr(tails|coin-flip) ≈ 50.05% → "*This coin normally comes up tails" should NOT be a true generic.

                                        • wrongInference : AntiProbArgument

                                          Wrong inference pattern: probabilistic semantics validates Modus Ponens (Pr(B|A) > 0.5 and A(a) → B(a) more likely than not) but NOT Defeasible Modus Ponens. Generics support the latter ("Birds fly, Tweety is a bird, so Tweety flies") but the inference is defeasible.

                                        • embeddedGenerics : AntiProbArgument

                                          Embedded generics ("Dogs chase cats that chase mice") require higher-order probabilities, leading to triviality results.

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