Documentation

Linglib.Phenomena.Modality.Studies.Alsop2024

@cite{alsop-2024} — Free Choice Any as GI-RSA #

@cite{alsop-2024} @cite{champollion-alsop-grosu-2019} @cite{franke-bergen-2020} @cite{tessler-franke-2019}

"Disjunction, Free Choice, and Exhaustification" (Chapter 4)

The Model #

Domain: "You may take any class" with 2 items {S, P}. 7 states based on permission structure (which baskets are permitted). 4 utterances. 2 global interpretation functions (weak/Szabolcsi vs strong/Dayal), following the GI-RSA architecture of @cite{franke-bergen-2020}.

Parameters: α = 2, uniform interpretation prior, configurable world prior.

Qualitative Findings #

#FindingTheorem
1Exclusiveness derivedexclusiveness_derived
2Exclusiveness robust to priorexclusiveness_robust
3Not-every holds under uniform priornot_every_uniform
4Not-every weakened under biased priornot_every_weakened
5Hearing "may S" → S is permittedliteral_s_correct
6Hearing "may every" → both permittedevery_permBoth
7Ambiguity essential for FCexclusiveness_requires_ambiguity
8No FC under negationno_fc_under_negation

The 7 states from @cite{alsop-2024} for a 2-item domain {S, P}. Each state is defined by which baskets are permitted: w0 (nothing), wS (S only), wP (P only), wSP (both).

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

      The 4 utterances.

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

          Two global interpretation functions (GI-RSA). Each assigns a meaning to every utterance simultaneously.

          Instances For
            Equations
            Instances For

              ◇take(S)_liberal: S is obtainable (alone or via both).

              Equations
              Instances For

                ◇take(P)_liberal: P is obtainable (alone or via both).

                Equations
                Instances For

                  Exclusiveness: each item is individually (strictly) permitted. ∀x[◇take(x)_strict]. True at {only1, anyNum}.

                  Equations
                  Instances For

                    Weak (Szabolcsi) interpretation: unexhaustified meanings.

                    • May S: ◇take(S)_liberal (6 states, all except onlyP)
                    • May P: ◇take(P)_liberal (6 states, all except onlyS)
                    • May Any: ∃x[◇take(x)] (7 states, always true)
                    • May Every: ◇take(S∧P) (4 states: anyNum, only2, sOrBoth, pOrBoth)
                    Equations
                    Instances For

                      Strong (Dayal) interpretation: exhaustified meanings.

                      • May S: {onlyS} (only S permitted, not P, not both)
                      • May P: {onlyP} (only P permitted, not S, not both)
                      • May Any: {only1, anyNum} (∀x[◇take(x)_strict], exclusiveness)
                      • May Every: {only2} (must take both, neither alone)
                      Equations
                      Instances For

                        The strong interpretation characterizes exclusiveness exactly.

                        The weak interpretation is always true for "may any".

                        Exclusiveness = ∀x[◇take(x)_strict].

                        The strong interpretation refines the weak for all utterances.

                        noncomputable def RSA.FCIAny.cfg (worldPr : FCIState) (hp : ∀ (w : FCIState), 0 worldPr w) :

                        @cite{alsop-2024} GI-RSA model for free choice any. Two global interpretations serve as latent variables. S1 score is rpow(L0, α) — standard belief-based RSA.

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

                          Uniform prior: all states equally likely.

                          Equations
                          Instances For
                            @[reducible, inline]

                            Biased prior: P(anyNum) = 3, others = 1. Biases toward the state where exclusiveness holds but not-every does not, testing prior sensitivity of the two inferences.

                            Equations
                            Instances For

                              Exclusiveness is derived: L1 assigns more mass to exclusiveness states than non-exclusiveness states upon hearing "may any".

                              Exclusiveness is robust: holds even under a prior biased toward anyNum.

                              Hearing "may S", the listener infers S is (strictly) permitted.

                              Hearing "may every", the listener infers both are permitted.

                              Counterfactual: both interpretations use weak meaning (no ambiguity). Without the informativity gap between weak (7 states for "may any") and strong (2 states), S1 cannot discriminate exclusiveness states.

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

                                Without interpretation ambiguity, exclusiveness is NOT derived. The informativity gap between weak (7 states) and strong (2 states) is what drives L1 toward exclusiveness states. Without a strong parse, "may any" is uninformative and the prior dominates: 2/7 exclusiveness states vs 5/7 non-exclusiveness states.

                                Extended utterances including negation of "may any".

                                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.

                                    RSAConfig for the extended model with negation.

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

                                      Free choice does NOT emerge under negation. Under negation, the weak interpretation is vacuous (false everywhere) and the strong interpretation supports only non-exclusiveness states. The informativity gap that drives FC in the positive case disappears.

                                      The 8 qualitative findings from @cite{alsop-2024}.

                                      Instances For
                                        Equations
                                        • One or more equations did not get rendered due to their size.
                                        Instances For
                                          noncomputable def RSA.FCIAny.formalize :

                                          Map each finding to its RSA formalization.

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

                                            The RSA model accounts for all 8 findings from @cite{alsop-2024}.

                                            Bridge content (merged from RSA_Alsop2024Bridge.lean) #

                                            Bridge: RSA Free Choice Any → Phenomena Data #

                                            @cite{alsop-2024}

                                            Connects the RSA free choice any model from @cite{alsop-2024} to empirical data in Phenomena.Modality.FreeChoice.

                                            Bridge Theorems #

                                            Connection to Phenomena #

                                            The model predicts the patterns in Phenomena.Modality.FreeChoice:

                                            1. FCI Any (anyClass, anyFruit):

                                              • "You may take any class" → permission for each class specifically
                                              • Derived: L1 assigns ~100% to exclusiveness states
                                            2. Robustness to priors:

                                              • Exclusiveness holds even with unfavorable priors
                                              • Parallels FCI robustness in disjunction
                                            3. Not-every is cancelable:

                                              • "You may take any class (in fact, you must take all of them)"
                                              • The "not every" inference can be cancelled, unlike exclusiveness

                                            Exclusiveness is robust to priors (as recorded in the data)