Documentation

Linglib.Phenomena.Modality.Studies.BarLevFox2020

Bar-Lev & Fox (2020) — Free Choice via Innocent Inclusion #

@cite{bar-lev-fox-2020} @cite{fox-2007} @cite{spector-2016}

Worked example of @cite{bar-lev-fox-2020} "Free choice, simplification, and Innocent Inclusion" (Natural Language Semantics 28:175–223) over a five-world toy modal model.

What this file does #

The abstract theory of Innocent Exclusion (IE), Innocent Inclusion (II), the cell-of-the-induced-partition (cell), and the cell-identification theorem mem_II_of_cell_witness lives in Theories/Semantics/Exhaustification/Operators/Basic.lean. This file instantiates that theory on a small FCWorld and verifies the paper's headline empirical prediction:

Exh^{IE+II}(◇(a ∨ b)) ⊨ ◇a ∧ ◇b

The contrast with simple disjunction (where the alternative set IS closed under conjunction and free choice does not arise) is captured via a parallel DisjWorld toy + simpleALT and simple_has_conjunction.

Why the cell-identification API matters #

In the paper, the move from "exhaustification of a disjunction" to "free choice" is enabled by a structural property of the alternative set: the pairwise conjunction of the disjunctive alternatives ◇a ∩ ◇b is NOT a member of Alt(◇(a∨b)) (paper eqn 13b p. 182). The cell of the partition induced by the alternatives is therefore consistent and identified by Exh^{IE+II}. The free choice proof is a one-line corollary of mem_II_of_cell_witness once a witness world for the cell is exhibited (the separatelyAB world, where each disjunct is individually possible but not jointly).

On the Exh^{IE+II} definition (paper §3, eqn 24-25 pp. 187-188) #

Exh^{IE+II} strengthens the prejacent with two operations:

  1. Innocent Exclusion (IE) — the intersection of maximal subsets of alternatives that can consistently be assigned false together with the prejacent. Members are negated.
  2. Innocent Inclusion (II) — the intersection of maximal subsets of alternatives that can consistently be assigned true together with the prejacent AND the IE negations from step 1. Members are asserted.

II is not "all non-IE alternatives" (a popular but incorrect gloss). The non-IE = II coincidence in the basic FC case is a derived fact (paper §3.3.3) once the cell is identified, not a definition. The substrate Theories/Semantics/Exhaustification/Operators/Basic.lean follows the paper's actual definitions.

Possible worlds for free choice: each represents a configuration of which disjuncts are individually or jointly permitted.

The separatelyAB world is the cell witness: each disjunct is individually permitted but they are not jointly permitted. This world distinguishes ◇a ∧ ◇b from ◇(a ∧ b) and is the cornerstone of @cite{bar-lev-fox-2020}'s free-choice derivation.

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

      The free-choice alternative set: {◇(a ∨ b), ◇a, ◇b, ◇(a ∧ b)}.

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

        @cite{bar-lev-fox-2020} eqn (13b) p. 182: the pairwise conjunction of the disjunctive alternatives is NOT closed by fcALT. The separatelyAB world satisfies permApermB but no element of fcALT (specifically not permAandB); witnesses at .onlyA/ .onlyB rule out the other three potential matches. This structural property — that Alt(◇(a∨b)) is not closed under pairwise conjunction — is what lets cell identification yield free choice.

        The cornerstone of the free-choice derivation is exhibiting a world that witnesses the cell of the partition induced by fcALT. Once this is in place, free choice follows as a one-line corollary of mem_II_of_cell_witness.

        The witness world is separatelyAB. Establishing the cell predicate at separatelyAB requires four facts about the IE structure of fcALT:

        Cell witness for the FC alternative set. The separatelyAB world verifies cell fcALT fcPrejacent — it satisfies the prejacent, falsifies every IE-excludable alternative (only permAandB), and verifies every non-excludable alternative (permAorB, permA, permB).

        Main free-choice theorem. Exh^{IE+II}(◇(a ∨ b)) ⊨ ◇a ∧ ◇b.

        Direct application of the substrate-level cell-witness factorization exhIEII_implies_cell_witnessed_alt (Operators/InnocentInclusion.lean) to each disjunct, using separatelyAB_in_cell as the cell witness. The substrate theorem encapsulates the abstract content of @cite{bar-lev-fox-2020} §3.3: any alternative true at a cell witness is entailed by exhIEII.

        Simple-disjunction worlds (no modal layer).

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

            Simple disjunction's alternative set: {a ∨ b, a, b, a ∧ b}.

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

              The structural contrast with FC: simple disjunction's alternative set is closed under conjunction (a ∧ b ∈ simpleALT). This is what blocks the cell from being consistent and prevents free choice from arising.

              SDA — the second contribution highlighted in the paper title #

              @cite{bar-lev-fox-2020} §7 derives simplification of disjunctive antecedents by applying Exh^{IE+II} to a counterfactual prejacent (p∨q)□→r over the 4-element alternative set obtained by disjunct-replacement (eqn 65 p. 205):

              Alt((p∨q)□→r) = {(p∨q)□→r, p□→r, q□→r, (p∧q)□→r}

              The maximal-false-assignment sets (eqn 66 p. 206) yield IE = {(p∧q)□→r}. Innocent Inclusion then asserts the three non-IE alternatives, giving Bar-Lev/Fox's verdict (eqn 67 p. 206):

              Exh^{IE+II}((p∨q)□→r) ⇔ (p□→r) ∧ (q□→r) ∧ ¬((p∧q)□→r)

              This is the central rival mechanism to @cite{santorio-2018}'s homogeneity-based SDA derivation: both predict the SDA inference (p∨q)□→r ⊨ (p□→r) ∧ (q□→r) via incompatible mechanisms (Bar-Lev/Fox: exhaustification-via-Innocent-Inclusion; Santorio: per-alternative DIST_π homogeneity over Katzir-generated truthmakers). The bar_lev_fox_sda_implies_santorio_sda_inference theorem at the end of this section makes this cross-mechanism agreement Lean-checkable.

              Worlds for the SDA toy model. actual is the evaluation world (no atomic prop holds); wp / wq / wpq are the relevant counterfactual alternatives where p (and r), q (and r), or both p and q (but not r) hold respectively.

              Designed so that, centered at actual, wp and wq are closer than wpq, making the three conditional alternatives evaluate as follows at actual: prejacent TRUE, p□→r TRUE, q□→r TRUE, (p∧q)□→r FALSE.

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

                  Similarity ordering: from actual, wp and wq are closer than wpq. From wp, wp itself is closer than wpq (so closest p-worlds from wp are just {wp}). Symmetric for wq. The minimal structure needed for the cell-witness analysis at actual.

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

                    The conditional p □→ r as a Set SDAWorld.

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

                      The conditional q □→ r as a Set SDAWorld.

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

                        The conditional (p∨q) □→ r (the prejacent) as a Set SDAWorld.

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

                          The conditional (p∧q) □→ r as a Set SDAWorld.

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

                            The SDA alternative set per eqn (65) p. 205.

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

                              Per-alternative verdicts at .actual #

                              Per-world verdicts at the MC-set witnesses .wp and .wq.

                              IE structure (paper eqn 66 p. 206) #

                              The maximal-false-assignment subsets of sdaALT (consistent with the prejacent) are {sdaPbox, sdaPandQbox} and {sdaQbox, sdaPandQbox}, yielding IE(prejacent, sdaALT) = {sdaPandQbox}. The proofs adapt §3's FC pattern (permA_not_ie / permB_not_ie / permAandB_is_ie) to conditional-typed alternatives.

                              sdaPandQbox is identically false on SDAWorld: the unique (p∧q)-world is .wpq, where r does not hold, so universalCounterfactual sim (p∧q) r u fails for every u.

                              sdaPbox is not innocently excludable: the MC-set {prejacent, sdaQboxᶜ, sdaPandQboxᶜ} does not contain sdaPboxᶜ.

                              sdaPandQbox IS innocently excludable: since sdaPandQbox is identically false on SDAWorld, sdaPandQboxᶜ holds at every world, so adjoining it to any MC-set keeps it consistent — maximality forces inclusion.

                              Cell witness for the SDA alternative set. The .actual world verifies cell sdaALT sdaPrejacent: it satisfies the prejacent plus the two non-IE conditional alternatives (sdaPbox, sdaQbox) and falsifies the unique IE alternative (sdaPandQbox). Once the IE-structure proofs above are complete, this follows from the per-alternative verdicts at .actual.

                              Bar-Lev/Fox SDA derivation (paper eqn 67 p. 206). Exh^{IE+II}((p∨q)□→r) entails (p□→r) ∧ (q□→r) ∧ ¬((p∧q)□→r). Three one-shot applications of the substrate-level cell-witness factorization theorems (Operators/InnocentInclusion.lean): exhIEII_implies_cell_witnessed_alt for the two non-IE conditionals (witnessed at .actual), and exhIEII_negates_excludable for the unique IE alternative.

                              The central debate Zani-Ciardelli-Sanfelici 2026 frames #

                              @cite{santorio-2018} derives SDA via per-alternative homogeneity (sdaEval = universal over per-disjunct counterfactuals). @cite{bar-lev-fox-2020} derives SDA via Innocent Inclusion (exhIEII over disjunct-replacement alternatives, asserting the non-IE conditional alternatives). The two mechanisms are RIVAL but they AGREE on the SDA inference proper:

                              (p∨q)□→r ⊨ (p□→r) ∧ (q□→r)

                              This agreement is the substrate for @cite{zani-ciardelli-sanfelici-2026}'s acquisition study, which contrasts the two frameworks' predicted developmental trajectories (Bar-Lev/Fox: AR→SDA; Santorio: DCR→SDA). The cross-mechanism agreement theorem below makes this Lean-checkable.

                              Bar-Lev/Fox's full verdict additionally entails ¬((p∧q)□→r) (the IE-driven negation of the conjunctive alternative); Santorio's sdaEval does not entail this. So Bar-Lev/Fox is STRICTLY STRONGER than Santorio on this scenario — they agree on SDA, diverge on the extra negative conjunct.

                              Cross-framework agreement on the SDA inference. @cite{bar-lev-fox-2020}'s Exh^{IE+II} verdict on the SDA prejacent entails @cite{santorio-2018}'s sdaEval verdict on the same scenario.

                              Direct application of the substrate-level multi-target cell-witness factorization exhIEII_implies_cell_witnessed_alts to the list of Santorio-style disjunct conditionals [sdaPbox, sdaQbox]. The factorization captures the abstract structural fact (any cell-witnessed alternatives are jointly entailed by Exh^{IE+II}); the bridge to sdaEval is mechanical via sdaEval_iff_forall. Santorio is silent on the negative conjunct ¬((p∧q)□→r) that Bar-Lev/Fox additionally derives — they coincide on the SDA inference proper, diverge on the negative component.

                              ◇∀x(Px ∨ Qx) ⊨ ◇∀xPx ∧ ◇∀xQx — second application of Exh^{IE+II} #

                              @cite{bar-lev-fox-2020} §6.1 derives universal free choice by applying Exh^{IE+II} to a counterfactual prejacent ◇∀x(Px ∨ Qx) over the 8-element alternative set obtained by replacing both the universal quantifier and the disjunction (eqn 55 p. 202):

                              Alt(◇∀x(P∨Q)) = {◇∀x(P∨Q), ◇∀xP, ◇∀xQ, ◇∀x(P∧Q), ◇∃x(P∨Q), ◇∃xP, ◇∃xQ, ◇∃x(P∧Q)}

                              The maximal-false-assignment subsets (eqn 56 p. 202) are three:

                              yielding IE = {◇∀x(P∧Q), ◇∃x(P∧Q)}. Innocent Inclusion then asserts the six non-IE alternatives, giving Bar-Lev/Fox's verdict (eqn 57 p. 202):

                              Exh^{IE+II}(◇∀x(P∨Q)) ⇔ ◇∀xP ∧ ◇∀xQ ∧ ¬◇∃x(P∧Q)

                              This is the second consumer of the substrate factorization theorems exhIEII_implies_cell_witnessed_alt and exhIEII_negates_excludable added in Theories/Semantics/Exhaustification/Operators/InnocentInclusion.lean. The cell witness wAllP_AllQ realizes simultaneous accessibility of the all-P and all-Q scenarios with no overlap-scenario; the main theorem follows in 3 substrate-application lines.

                              Worlds for the universal-FC toy model. Each world represents the set of accessible scenarios from the evaluation point — abstract enough to validate the 8-alternative IE structure of paper eqn 55 p. 202.

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

                                  The 8 alternative predicates per paper eqn (55) p. 202 #

                                  ◇∀x(P∧Q) — some accessible scenario has every individual reading both. None of our worlds witness an all-both scenario.

                                  Equations
                                  Instances For

                                    ◇∃x(P∧Q) — some accessible scenario has some individual reading both. None of our worlds witness an overlap scenario.

                                    Equations
                                    Instances For

                                      The 8-element alternative set per paper eqn (55) p. 202.

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

                                        Per-world verdicts at the cell witness wAllP_AllQ #

                                        Established by direct case analysis. The cell witness satisfies the prejacent + every non-IE alternative; the two IE alternatives (univFcAllBoth, univFcSomeBoth) are identically false on this toy model.

                                        Finset-side alternative set + bridge to Set side #

                                        For decidability of cell membership and IE-ness, work with Finset versions of the 8 alternatives. Bridge equations to the Set side let the general substrate theorems (Operators/InnocentInclusion.lean::mem_II_of_cell_witness, not_isInnocentlyExcludable_of_cell_witness) consume Finset-derived facts via decide.

                                        Finset version of univFcAllPorQ.

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

                                                The 8-element alternative set (Finset-typed) per paper eqn (55) p. 202.

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

                                                  Per-alt Set/Finset bridge equations #

                                                  ALT-level bridge: the Set-typed universalFcALT equals the asSetOfSets image of the Finset-typed universalFcALT_f. Lifts Finset-side membership facts to the Set-side substrate theorems.

                                                  Cell witness via Finset bridge #

                                                  Cell witness for the universal-FC alternative set (paper eqn 56 p. 202 cell). wAllP_AllQ is in the cell. Proof: decide- checkable on the Finset side (Operators/Decidable.lean::cellFinset); lifted to Set via mem_cellFinset_iff + universalFcALT_eq / universalFcPrejacent_eq. Replaces ~250 LOC of manual MC-set + IE-structure proofs from earlier versions.

                                                  Innocent excludability theorems (cell-witness corollaries) #

                                                  univFcAllBoth IS innocently excludable: identically false on UnivFCWorld, so its negation is trivially consistent with any MC-set; maximality forces inclusion.

                                                  Bar-Lev/Fox universal free choice (paper eqn 57 p. 202). Exh^{IE+II}(◇∀x(P∨Q)) ⊨ ◇∀xP ∧ ◇∀xQ ∧ ¬◇∃x(P∧Q).

                                                  Three one-shot applications of the substrate-level cell-witness factorization theorems: exhIEII_implies_cell_witnessed_alt for the two universal-distributive non-IE alternatives, and exhIEII_negates_excludable for the existential-conjunctive IE alternative. The substrate factorization (Theories/Semantics/ Exhaustification/Operators/InnocentInclusion.lean) is what makes this main theorem a 3-line consequence of the cell witness.