Documentation

Studies.ChemlaSpector2011

Chemla and Spector (2011): Experimental evidence for embedded scalar implicatures #

Chemla and Spector argue, against Geurts and Pouscoulous, that scalar inferences computed in embedded position are detectable. A graded truth-value judgment rates a picture higher when it satisfies more of a sentence's available readings (§3.2), and the readings a theory makes available are those its mechanism can derive. A restricted globalist computes the inference at the speech-act level only, a localist anywhere, and an unrestricted globalist wherever the result entails the literal reading. Under a universal the local reading entails the literal one, so Experiment 1 separates only the restricted globalist; under exactly one it does not, so Experiment 2 separates the localist from both globalists.

Main results #

Implementation notes #

References #

Readings and theories (§1) #

The three readings the two experiments cross; their entailments differ between the experiments, which the design exploits.

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

      The three families of theory compute the implicature at the speech-act level only (T1), in embedded position (T2), or globally but with alternatives that need not be stronger than the sentence (T3).

      Instances For
        @[instance_reducible]
        Equations
        def ChemlaSpector2011.instReprTheory.repr :
        Theory → ℕ → Std.Format
        Equations
        • One or more equations did not get rendered due to their size.
        Instances For
          def ChemlaSpector2011.GloballyDerivable {P : Type u_1} (readings : ReadingLabel → P → Prop) (ℓ : ReadingLabel) :

          A globalist derivation reaches a reading exactly when it entails the literal one, since strengthening the matrix meaning never yields something the literal reading does not follow from.

          Equations
          Instances For
            @[instance_reducible]
            instance ChemlaSpector2011.instDecidableGloballyDerivable {P : Type u_1} [Fintype P] (readings : ReadingLabel → P → Prop) [(ℓ : ReadingLabel) → (p : P) → Decidable (readings ℓ p)] (ℓ : ReadingLabel) :
            Decidable (GloballyDerivable readings ℓ)
            Equations
            • One or more equations did not get rendered due to their size.
            @[instance_reducible]
            instance ChemlaSpector2011.instDecidableAdmits {P : Type u_1} [Fintype P] (readings : ReadingLabel → P → Prop) [(ℓ : ReadingLabel) → (p : P) → Decidable (readings ℓ p)] (t : Theory) (ℓ : ReadingLabel) :
            Decidable (Theory.admits readings t ℓ)
            Equations
            def ChemlaSpector2011.availableAt {P : Type u_1} [Fintype P] (readings : ReadingLabel → P → Prop) [(ℓ : ReadingLabel) → (p : P) → Decidable (readings ℓ p)] (t : Theory) (p : P) :

            The readings a theory leaves available at a picture are those true there and within its reach; by the conjecture of §3.2 the rating reflects this set.

            Equations
            Instances For
              @[reducible, inline]
              abbrev ChemlaSpector2011.RatingsMonotone (data : List (ℕ × ℕ)) :

              The conjecture of §3.2 says that where strictly more of a sentence's available readings are true, the rating is higher; it is stated over pairs of a rating and a reading count.

              Equations
              Instances For

                Pictures (§3, Appendix 2) #

                @[reducible, inline]

                A six-letter picture places each letter at a vertex of the triangle of opposition, a falsifier being connected to none of its circles (E), a strong verifier to some but not all (IO), and a weak verifier to all (A) (Figure 14).

                Equations
                Instances For
                  @[reducible, inline]

                  A three-letter picture.

                  Equations
                  Instances For

                    Experiment 1: scalar items under a universal (§4) #

                    The three readings of (8) over any domain of letters are (10a), each letter is connected with at least one of its circles, (10b), that and not every letter with all of them, and (10c), every letter with some but not all of them.

                    Equations
                    Instances For
                      @[instance_reducible]
                      instance ChemlaSpector2011.Exp1Some.instDecidableReadingOfFintype {ι : Type u_1} [Fintype ι] (ℓ : ReadingLabel) (p : ι → Aristotelian.Triangle) :
                      Decidable (reading ℓ p)
                      Equations

                      The local reading is the literal one with all denied of every letter.

                      Under the universal the local reading entails the literal one, which keeps the unrestricted globalist abreast of the localist throughout Experiment 1.

                      The target conditions of Experiment 1 (§4.2.1) make no reading true, only the literal one, the literal and global ones, or all three.

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

                          The picture of each condition (Figure 4) has six falsifiers, six weak verifiers, four weak and two strong verifiers, or six strong verifiers.

                          Equations
                          Instances For

                            The readings true at a condition, read off its witness.

                            Equations
                            Instances For

                              Experiment 2: scalar items under exactly one (§5) #

                              @[reducible, inline]

                              Reading (19a) says that exactly one letter is connected with some or all of its circles, the others with none.

                              Equations
                              Instances For
                                @[reducible, inline]

                                Reading (19b) says that exactly one letter is connected with some but not all of its circles, the others with none.

                                Equations
                                Instances For
                                  @[reducible, inline]

                                  Reading (19c) says that exactly one letter is connected with some but not all of its circles, the others with none or all.

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

                                    Under exactly one the local reading no longer entails the literal one, since a strong verifier among weak verifiers makes it true and the literal reading false, which puts it beyond a globalist derivation's reach (§5.1).

                                    The target conditions of Experiment 2 (§5.3.1) make no reading true, only the literal one, only the local one, or all three.

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

                                        The readings true at a condition, read off its witness.

                                        Equations
                                        Instances For

                                          The downward-entailing controls (12) and (13) at the end of each experiment (§4.2.2) make no reading true, only the marginal local reading, or both readings.

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

                                              The rows #

                                              The two experiments.

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

                                                  The two scalar items, certains and ou.

                                                  Instances For
                                                    @[instance_reducible]
                                                    Equations
                                                    def ChemlaSpector2011.instReprItem.repr :
                                                    Item → ℕ → Std.Format
                                                    Equations
                                                    • One or more equations did not get rendered due to their size.
                                                    Instances For
                                                      def ChemlaSpector2011.rating (e : Experiment) (i : Item) (env cond : String) :
                                                      Option ℕ

                                                      The mean rating, in per-mille, of an item in a condition of an experiment.

                                                      Equations
                                                      • One or more equations did not get rendered due to their size.
                                                      Instances For
                                                        def ChemlaSpector2011.rating₁ (i : Item) (c : Exp1Condition) :
                                                        Option ℕ

                                                        The rating of a target condition of Experiment 1.

                                                        Equations
                                                        Instances For
                                                          def ChemlaSpector2011.rating₂ (i : Item) (c : Exp2Condition) :
                                                          Option ℕ

                                                          The rating of a target condition of Experiment 2.

                                                          Equations
                                                          Instances For

                                                            The rating of a downward-entailing control.

                                                            Equations
                                                            Instances For

                                                              Experiment 1 #

                                                              In Figure 5 the ratings of both items rise with the readings true at the condition's witness, the conjecture of §3.2 on Experiment 1, which subsumes the contrast between STRONG and WEAK that only the truth of the local reading separates.

                                                              The restricted globalist leaves the same readings available at STRONG as at WEAK, the local reading being true at STRONG but beyond a matrix-only mechanism, so it predicts equal ratings.

                                                              The unrestricted globalist agrees with the localist on every condition of Experiment 1, the local reading being within its reach there, which is why the experiment cannot separate them.

                                                              The local reading is reinforceable over the literal one, since at WEAK the literal reading holds and the local one fails.

                                                              Experiment 2 #

                                                              In Figure 12 the condition where only the local reading of certains is true is rated above the one where only the literal reading is, though the literal reading is false there.

                                                              In Figure 12 the local condition is rated far above the false one for both items.

                                                              The unrestricted globalist collapses at the diagnostic condition, where the one reading true at LOCAL is beyond its reach, so it leaves nothing available there, as at FALSE, and predicts equal ratings.

                                                              The downward-entailing controls (§4.4.4, §5.5.4) #

                                                              In Figures 6 and 13 the condition where only the marginal local reading is true is rated far below the one where both readings are, in both experiments and for both items.

                                                              The same controls are rated higher after the non-monotonic items of Experiment 2 than after Experiment 1, which §5.5.4 attributes to exposure to salient local readings.

                                                              Even primed, the local reading under negation stays below the local reading under exactly one (§5.5.4).

                                                              The marginal local reading under negation is rated far below the baseline, as the responses consistent with a local implicature fall far below the genuine-ambiguity baseline in the fourth experiment of Geurts and Pouscoulous.