Documentation

Studies.Spector2017

Spector (2017): The pragmatics of plural predication: Homogeneity and non-maximality within the Rational Speech Act model #

Spector reconstructs Križ and Spector's pragmatics of plural predication in the Rational Speech Act model. A plural definite is ambiguous between an existential and a universal reading, and the first speaker, unsure which reading the literal listener uses, weighs a message by its expected utility across the readings (S1_apply_singleton). A message false on some reading throughout the speaker's answer has utility minus infinity there, which derives Križ and Spector's truth on all interpretations (S1_apply_singleton_eq_zero_iff). The utility is the literal listener's probability of the speaker's answer to the question under discussion together with the question. When the question is what is the case, the students came and the students didn't come are used only where all or where none of the students came (homogeneity); when it is whether some of them came, both are existential (nonMaximality), and the listener who hears the students came entertains that only some came only under that question (L1_the_justSome_ne_zero_iff).

Implementation notes #

TODO #

References #

No student came, some but not all did, or all did.

Instances For
    @[instance_reducible]
    Equations
    @[instance_reducible]
    Equations
    def Spector2017.instReprWorld.repr :
    World → ℕ → Std.Format
    Equations
    Instances For
      @[instance_reducible]
      Equations

      The messages of the model.

      Instances For
        @[instance_reducible]
        Equations
        @[instance_reducible]
        Equations
        def Spector2017.instReprMessage.repr :
        Message → ℕ → Std.Format
        Equations
        Instances For
          @[instance_reducible]
          Equations

          The plural definite is read existentially or universally.

          Instances For
            @[instance_reducible]
            Equations
            def Spector2017.instReprInterp.repr :
            Interp → ℕ → Std.Format
            Equations
            Instances For
              @[instance_reducible]
              Equations
              @[instance_reducible]
              Equations

              Did some of the students come, did all of them, and what is the case.

              Instances For
                @[instance_reducible]
                instance Spector2017.instDecidableEqQUD :
                DecidableEq QUD
                Equations
                def Spector2017.instReprQUD.repr :
                QUD → ℕ → Std.Format
                Equations
                Instances For
                  @[instance_reducible]
                  Equations
                  @[instance_reducible]
                  Equations
                  @[instance_reducible]
                  instance Spector2017.instMeasurableSpaceQUD :
                  MeasurableSpace QUD
                  Equations
                  instance Spector2017.instDiscreteMeasurableSpaceQUD :
                  DiscreteMeasurableSpace QUD
                  def Spector2017.QUD.answer :
                  QUD → World → Bool ⊕ World

                  The answer to a question at a world: yes or no to a polar question, the world itself to what is the case.

                  Equations
                  Instances For
                    @[instance_reducible]
                    instance Spector2017.instDecidablePredWorldMemSetMeaning (i : Interp) (u : Message) :
                    DecidablePred fun (x : World) => x ∈ meaning i u
                    Equations
                    • One or more equations did not get rendered due to their size.
                    def Spector2017.answer (p : World × QUD) :
                    (Bool ⊕ World) × QUD

                    The answer at a world and a question, together with the question.

                    Equations
                    Instances For
                      noncomputable def Spector2017.L0 (μ : MeasureTheory.Measure World) (ρ : MeasureTheory.Measure QUD) (i : Interp) :
                      ProbabilityTheory.Kernel Message (World × QUD)

                      The literal listener of an interpretation conditions the prior over worlds and questions on the message.

                      Equations
                      Instances For
                        instance Spector2017.instIsFiniteKernelMessageProdWorldQUDL0 (μ : MeasureTheory.Measure World) (ρ : MeasureTheory.Measure QUD) (i : Interp) :
                        ProbabilityTheory.IsFiniteKernel (L0 μ ρ i)
                        noncomputable def Spector2017.answerL0 (μ : MeasureTheory.Measure World) (ρ : MeasureTheory.Measure QUD) (u : Message) :
                        MeasureTheory.Measure (Interp × (Bool ⊕ World) × QUD)

                        The literal listeners' probabilities of the answers, one for each interpretation.

                        Equations
                        Instances For
                          instance Spector2017.instIsFiniteMeasureProdInterpSumBoolWorldQUDAnswerL0 (μ : MeasureTheory.Measure World) (ρ : MeasureTheory.Measure QUD) (u : Message) :
                          MeasureTheory.IsFiniteMeasure (answerL0 μ ρ u)
                          noncomputable def Spector2017.S1 (μ : MeasureTheory.Measure World) (ρ : MeasureTheory.Measure QUD) (ν : MeasureTheory.Measure Interp) (α : ℝ) :
                          ProbabilityTheory.Kernel (World × QUD) Message

                          The first speaker is uncertain about the interpretation of the literal listener and weighs a message by its expected utility across interpretations.

                          Equations
                          Instances For
                            instance Spector2017.instIsFiniteKernelProdWorldQUDMessageS1 (μ : MeasureTheory.Measure World) (ρ : MeasureTheory.Measure QUD) (ν : MeasureTheory.Measure Interp) (α : ℝ) :
                            ProbabilityTheory.IsFiniteKernel (S1 μ ρ ν α)
                            noncomputable def Spector2017.L1 (μ : MeasureTheory.Measure World) (ρ : MeasureTheory.Measure QUD) (ν : MeasureTheory.Measure Interp) (α : ℝ) :
                            ProbabilityTheory.Kernel Message (World × QUD)

                            The pragmatic listener inverts the first speaker against the prior over worlds and questions.

                            Equations
                            Instances For
                              theorem Spector2017.answerL0_apply_singleton (μ : MeasureTheory.Measure World) (ρ : MeasureTheory.Measure QUD) (u : Message) (i : Interp) (x : (Bool ⊕ World) × QUD) :
                              (answerL0 μ ρ u) {(i, x)} = (((L0 μ ρ i).map answer) u) {x}
                              theorem Spector2017.S1_apply_singleton (μ : MeasureTheory.Measure World) (ρ : MeasureTheory.Measure QUD) (ν : MeasureTheory.Measure Interp) (α : ℝ) (p : World × QUD) (u : Message) :
                              ((S1 μ ρ ν α) p) {u} = (ν.geomMean fun (i : Interp) => (((L0 μ ρ i).map answer) u) {answer p}) ^ α * ENNReal.ofReal (Real.exp (-(α * cost u))) / ∑ u' : Message, (ν.geomMean fun (i : Interp) => (((L0 μ ρ i).map answer) u') {answer p}) ^ α * ENNReal.ofReal (Real.exp (-(α * cost u')))

                              The first speaker weighs a message by the geometric mean over interpretations of the literal listener's probability of her answer, to the power of the rationality, times the cost weight: the exponential of the rationality times the expected utility across interpretations.

                              theorem Spector2017.answerL0_answer_eq_zero_iff {μ : MeasureTheory.Measure World} {ρ : MeasureTheory.Measure QUD} [MeasureTheory.IsFiniteMeasure μ] [MeasureTheory.IsFiniteMeasure ρ] (hμ : ∀ (w : World), μ {w} ≠ 0) (hρ : ∀ (Q : QUD), ρ {Q} ≠ 0) (u : Message) (i : Interp) (w : World) (Q : QUD) :
                              (answerL0 μ ρ u) {(i, answer (w, Q))} = 0 ↔ ∀ (v : World), Q.answer v = Q.answer w → v ∉ meaning i u

                              Under an interpretation, the literal listener gives no probability to an answer exactly when the message is false at every world with that answer.

                              theorem Spector2017.S1_apply_singleton_eq_zero_iff {μ : MeasureTheory.Measure World} {ρ : MeasureTheory.Measure QUD} {ν : MeasureTheory.Measure Interp} {α : ℝ} [MeasureTheory.IsFiniteMeasure μ] [MeasureTheory.IsFiniteMeasure ρ] (hμ : ∀ (w : World), μ {w} ≠ 0) (hρ : ∀ (Q : QUD), ρ {Q} ≠ 0) [MeasureTheory.IsFiniteMeasure ν] (hα : 0 < α) (w : World) (Q : QUD) (u : Message) :
                              ((S1 μ ρ ν α) (w, Q)) {u} = 0 ↔ ∃ (i : Interp), ν {i} ≠ 0 ∧ ∀ (v : World), Q.answer v = Q.answer w → v ∉ meaning i u

                              The first speaker never uses a message at a world when, under some interpretation, it is false at every world with her answer, the principle of truth on all interpretations.

                              theorem Spector2017.S1_apply_singleton_ne_zero_iff {μ : MeasureTheory.Measure World} {ρ : MeasureTheory.Measure QUD} {ν : MeasureTheory.Measure Interp} {α : ℝ} [MeasureTheory.IsFiniteMeasure μ] [MeasureTheory.IsFiniteMeasure ρ] (hμ : ∀ (w : World), μ {w} ≠ 0) (hρ : ∀ (Q : QUD), ρ {Q} ≠ 0) [MeasureTheory.IsFiniteMeasure ν] (hα : 0 < α) (hν : ∀ (i : Interp), ν {i} ≠ 0) (w : World) (Q : QUD) (u : Message) :
                              ((S1 μ ρ ν α) (w, Q)) {u} ≠ 0 ↔ ∀ (i : Interp), ∃ (v : World), Q.answer v = Q.answer w ∧ v ∈ meaning i u

                              With every interpretation possible, the first speaker uses a message exactly when, under each interpretation, it is true at some world with her answer.

                              theorem Spector2017.homogeneity {μ : MeasureTheory.Measure World} {ρ : MeasureTheory.Measure QUD} {ν : MeasureTheory.Measure Interp} {α : ℝ} [MeasureTheory.IsFiniteMeasure μ] [MeasureTheory.IsFiniteMeasure ρ] (hμ : ∀ (w : World), μ {w} ≠ 0) (hρ : ∀ (Q : QUD), ρ {Q} ≠ 0) [MeasureTheory.IsFiniteMeasure ν] (hα : 0 < α) (hν : ∀ (i : Interp), ν {i} ≠ 0) (w : World) :
                              (((S1 μ ρ ν α) (w, QUD.whatsTheCase)) {Message.the} ≠ 0 ↔ w = World.all) ∧ (((S1 μ ρ ν α) (w, QUD.whatsTheCase)) {Message.notThe} ≠ 0 ↔ w = World.none)

                              When the question is what is the case, the first speaker uses the only where all of the students came and not the only where none did, so neither where only some did.

                              theorem Spector2017.nonMaximality {μ : MeasureTheory.Measure World} {ρ : MeasureTheory.Measure QUD} {ν : MeasureTheory.Measure Interp} {α : ℝ} [MeasureTheory.IsFiniteMeasure μ] [MeasureTheory.IsFiniteMeasure ρ] (hμ : ∀ (w : World), μ {w} ≠ 0) (hρ : ∀ (Q : QUD), ρ {Q} ≠ 0) [MeasureTheory.IsFiniteMeasure ν] (hα : 0 < α) (hν : ∀ (i : Interp), ν {i} ≠ 0) (w : World) :
                              (((S1 μ ρ ν α) (w, QUD.isSome)) {Message.the} ≠ 0 ↔ w ≠ World.none) ∧ (((S1 μ ρ ν α) (w, QUD.isSome)) {Message.notThe} ≠ 0 ↔ w = World.none)

                              When the question is whether some of the students came, the first speaker uses the wherever some did and not the only where none did, as existential quantifiers.

                              theorem Spector2017.L1_the_justSome_ne_zero_iff {μ : MeasureTheory.Measure World} {ρ : MeasureTheory.Measure QUD} {ν : MeasureTheory.Measure Interp} {α : ℝ} [MeasureTheory.IsFiniteMeasure μ] [MeasureTheory.IsFiniteMeasure ρ] (hμ : ∀ (w : World), μ {w} ≠ 0) (hρ : ∀ (Q : QUD), ρ {Q} ≠ 0) [MeasureTheory.IsFiniteMeasure ν] (hα : 0 < α) (hν : ∀ (i : Interp), ν {i} ≠ 0) (Q : QUD) :
                              ((L1 μ ρ ν α) Message.the) {(World.justSome, Q)} ≠ 0 ↔ Q = QUD.isSome

                              Hearing the, the pragmatic listener takes it to be possible that only some of the students came exactly as an answer to whether some of them did.