Documentation

Linglib.Phenomena.Presupposition.Studies.DegenTonhauser2021

@cite{degen-tonhauser-2021}: Prior Beliefs Modulate Projection #

@cite{qing-goodman-lassiter-2016}

Prior beliefs modulate projection. Open Mind 5:59–70.

Core Finding #

Higher prior probability of complement content leads to stronger projection inferences, at both the group and individual participant level.

This is the first systematic demonstration that subjective prior beliefs about the world modulate presupposition projection across a wide range of clause-embedding predicates.

Experiments #

Experiment 1 (within-participant, N=286) #

Experiment 2 (between-participant replication) #

Replication #

By-predicate projection ranking replicates Tonhauser & Degen 2020 with Spearman r = .991.

Theoretical Significance #

The prior-belief effect motivates probabilistic models of projection (e.g., @cite{qing-goodman-lassiter-2016}). Both existing probabilistic projection models (Qing et al. 2016, Stevens et al. 2017) are couched within the RSA framework, which standardly assumes utterance interpretation is modulated by listeners' prior beliefs.

source
inductive Phenomena.Presupposition.Studies.DegenTonhauser2021.Predicate :
Type

The 20 clause-embedding predicates investigated in D&T 2021, listed alphabetically as in Figure 1C. The set spans cognitive (know), emotive (be annoyed), communication (announce), and inferential (prove) predicates. For the traditional factive/nonfactive classification of these predicates, see DegenTonhauser2022.traditionalClass.

Instances For
    source
    instance Phenomena.Presupposition.Studies.DegenTonhauser2021.instDecidableEqPredicate :
    DecidableEq Predicate
    Equations
    source
    instance Phenomena.Presupposition.Studies.DegenTonhauser2021.instReprPredicate :
    Repr Predicate
    Equations
    source
    def Phenomena.Presupposition.Studies.DegenTonhauser2021.instReprPredicate.repr :
    Predicate → ℕ → Std.Format
    Equations
    • One or more equations did not get rendered due to their size.
    Instances For
      source
      structure Phenomena.Presupposition.Studies.DegenTonhauser2021.RegressionEffect :
      Type

      A regression coefficient from a mixed-effects model.

      Instances For
        source
        instance Phenomena.Presupposition.Studies.DegenTonhauser2021.instReprRegressionEffect :
        Repr RegressionEffect
        Equations
        source
        def Phenomena.Presupposition.Studies.DegenTonhauser2021.instReprRegressionEffect.repr :
        RegressionEffect → ℕ → Std.Format
        Equations
        • One or more equations did not get rendered due to their size.
        Instances For
          source
          inductive Phenomena.Presupposition.Studies.DegenTonhauser2021.PriorLevel :
          Type

          The three levels of prior-belief predictor tested in Experiment 1. The individual-level model captures the most variance (lowest BIC). Experiment 2 is between-participant, so only categorical and group-level are tested there (individual priors unavailable).

          • categorical : PriorLevel

            Binary: high vs low fact condition.

          • groupLevel : PriorLevel

            Continuous: group-level mean prior probability by item.

          • individualLevel : PriorLevel

            Continuous: each participant's own prior probability rating. Only available in within-participant designs (Experiment 1).

          Instances For
            source
            instance Phenomena.Presupposition.Studies.DegenTonhauser2021.instDecidableEqPriorLevel :
            DecidableEq PriorLevel
            Equations
            source
            def Phenomena.Presupposition.Studies.DegenTonhauser2021.instReprPriorLevel.repr :
            PriorLevel → ℕ → Std.Format
            Equations
            • One or more equations did not get rendered due to their size.
            Instances For
              source
              instance Phenomena.Presupposition.Studies.DegenTonhauser2021.instReprPriorLevel :
              Repr PriorLevel
              Equations
              source
              def Phenomena.Presupposition.Studies.DegenTonhauser2021.exp1_priorManipulation :
              RegressionEffect

              Experiment 1: prior manipulation was successful. Higher-probability facts yielded higher prior ratings than lower-probability facts (β = 0.45, SE = 0.01, t = 31.12).

              Equations
              Instances For
                source
                def Phenomena.Presupposition.Studies.DegenTonhauser2021.exp1_priorEffect :
                PriorLevel → RegressionEffect

                Experiment 1 regression: prior predicts projection at all three levels.

                Equations
                Instances For
                  source
                  def Phenomena.Presupposition.Studies.DegenTonhauser2021.exp2b_priorEffect :
                  PriorLevel → Option RegressionEffect

                  Experiment 2b regression: between-participant replication. Only categorical and group-level predictors are available — Exp 2 is between-participant, so individual priors are not measured alongside projection.

                  Equations
                  Instances For
                    source
                    def Phenomena.Presupposition.Studies.DegenTonhauser2021.exp1_bic :
                    PriorLevel → Float

                    BIC values for the three prior-level models in Experiment 1. Lower = better fit. Individual-level wins decisively.

                    Equations
                    Instances For
                      source
                      theorem Phenomena.Presupposition.Studies.DegenTonhauser2021.individual_best_by_bic :
                      exp1_bic PriorLevel.individualLevel < exp1_bic PriorLevel.groupLevel ∧ exp1_bic PriorLevel.groupLevel < exp1_bic PriorLevel.categorical

                      Individual-level prior beliefs predict projection better than group-level beliefs, which predict better than categorical (binary) beliefs. This is the key model-comparison result.

                      source
                      theorem Phenomena.Presupposition.Studies.DegenTonhauser2021.prior_manipulation_successful :
                      exp1_priorManipulation.β > 0

                      The prior manipulation was successful: β > 0.

                      source
                      theorem Phenomena.Presupposition.Studies.DegenTonhauser2021.prior_effect_positive_exp1 (level : PriorLevel) :
                      (exp1_priorEffect level).β > 0

                      The prior effect is positive at every level of analysis in Experiment 1. A positive β means higher prior → stronger projection.

                      source
                      theorem Phenomena.Presupposition.Studies.DegenTonhauser2021.prior_effect_positive_exp2b_cat :
                      exp2b_priorEffect PriorLevel.categorical = some { β := 0.18, se := 1e-2, t := 12.81 }

                      The prior effect replicates in the between-participant design at both applicable levels (categorical and group-level).

                      source
                      theorem Phenomena.Presupposition.Studies.DegenTonhauser2021.prior_effect_positive_exp2b_grp :
                      exp2b_priorEffect PriorLevel.groupLevel = some { β := 0.34, se := 3e-2, t := 13.27 }
                      source
                      theorem Phenomena.Presupposition.Studies.DegenTonhauser2021.exp2b_no_individual :
                      exp2b_priorEffect PriorLevel.individualLevel = none

                      Individual-level prior is not available in the between-participant design.

                      source
                      theorem Phenomena.Presupposition.Studies.DegenTonhauser2021.prior_effect_replicates :
                      (exp1_priorEffect PriorLevel.categorical).β > 0 ∧ exp2b_priorEffect PriorLevel.categorical = some { β := 0.18, se := 1e-2, t := 12.81 }

                      The prior effect replicates across experiments: Exp 1 (within-participant) and Exp 2b (between-participant) show the same direction at the categorical level.

                      source
                      def Phenomena.Presupposition.Studies.DegenTonhauser2021.replication_r_exp1_vs_td2020 :
                      Float

                      Spearman rank correlation between Exp 1 by-predicate certainty ratings (collapsing over facts) and Tonhauser & Degen 2020 Experiment 1a.

                      Equations
                      Instances For
                        source
                        def Phenomena.Presupposition.Studies.DegenTonhauser2021.replication_r_exp2a_vs_exp1 :
                        Float

                        Spearman rank correlation between Exp 2a and Exp 1 prior ratings.

                        Equations
                        Instances For
                          source
                          theorem Phenomena.Presupposition.Studies.DegenTonhauser2021.replication_robust :
                          replication_r_exp1_vs_td2020 > 0.95 ∧ replication_r_exp2a_vs_exp1 > 0.95

                          Both replication correlations are very high (> 0.95), confirming that the by-predicate ranking is robust across experiments.

                          source
                          def Phenomena.Presupposition.Studies.DegenTonhauser2021.projectionProfile :
                          Predicate → Gradience.ProjectionProfile

                          Mean certainty ratings by predicate and fact condition from Experiment 1. Values computed from the data at github.com/judith-tonhauser/projective-probability (results/9-prior-projection), rounded to 2 decimal places. Ordered by overall projection strength (Figure 3 x-axis).

                          Equations
                          Instances For
                            source
                            theorem Phenomena.Presupposition.Studies.DegenTonhauser2021.prior_modulates_all (p : Predicate) :
                            Gradience.priorBeliefModulatesProjection (projectionProfile p)

                            The core finding: prior beliefs modulate projection for EVERY predicate. High-prior content projects more strongly than low-prior content across all 20 predicates.

                            source
                            def Phenomena.Presupposition.Studies.DegenTonhauser2021.toVerbEntry :
                            Predicate → Option Fragments.English.Predicates.Verbal.VerbEntry

                            Map each D&T predicate to its Fragment verb entry (18 of 20). "be annoyed" and "be right" are copular — use toPredicateCore for full coverage.

                            Equations
                            Instances For
                              source
                              def Phenomena.Presupposition.Studies.DegenTonhauser2021.toPredicateCore :
                              Predicate → Core.Verbs.VerbCore

                              Map each D&T predicate to its VerbCore — the semantic spine shared by verbal and copular entries. Covers all 20 predicates. Copular entries go through ClauseEmbeddingAdj.toVerbCore (English-specific realization: copula + adjective).

                              Equations
                              Instances For
                                source
                                def Phenomena.Presupposition.Studies.DegenTonhauser2021.allPredicates :
                                List Predicate

                                All 20 D&T predicates (alphabetical).

                                Equations
                                • One or more equations did not get rendered due to their size.
                                Instances For
                                  source
                                  theorem Phenomena.Presupposition.Studies.DegenTonhauser2021.full_coverage :
                                  allPredicates.length = 20

                                  All 20 predicates have Fragment entries via toPredicateCore.

                                  source
                                  theorem Phenomena.Presupposition.Studies.DegenTonhauser2021.verbEntry_coverage :
                                  (List.filter (fun (p : Predicate) => (toVerbEntry p).isSome) allPredicates).length = 18

                                  18 of 20 predicates have VerbEntry entries (all except copular "be annoyed" and "be right").

                                  source
                                  theorem Phenomena.Presupposition.Studies.DegenTonhauser2021.all_predicates_take_clause_complement (p : Predicate) :
                                  (toPredicateCore p).complementType = Core.Verbs.ComplementType.finiteClause ∨ (toPredicateCore p).altComplementType = some Core.Verbs.ComplementType.finiteClause

                                  Every predicate takes a finite clause complement (as primary or alternate frame), matching D&T's experimental design.