Wang & Davidson (2026): Presupposition Filtering in Disjunction #
@cite{wang-davidson-2026}
Yiqian Wang and Kathryn Davidson, "Presupposition filtering in disjunction: The role of exclusive interpretation." Proceedings of Sinn und Bedeutung 30.
Summary #
The paper asks whether exclusive interpretation of disjunction (via scalar implicature / exhaustification) affects presupposition projection. The answer, both theoretically and empirically, is:
Most theories predict yes, but the experiment finds no.
Theoretical Contribution (§3) #
The paper surveys combinations of:
- Exhaustification theories: bivalent EXH, trivalent EXH¹, EXH²
- Projection theories: Strong Kleene, George 2008, classic dynamic semantics
These divide into two classes:
- Type A: increasing exclusivity reduces filtering (bivalent EXH + SK/George/dynamic; EXH² + any projection)
- Type B: exclusivity has no effect on filtering (EXH¹ + any projection)
The core mechanism: under Strong Kleene, inclusive disjunction
(Truth3.join) can return .true even when one disjunct is
undefined — so a true first disjunct "filters" the second's
presupposition failure. Exclusive disjunction (Truth3.xor)
cannot: it returns .indet whenever either input is .indet.
Feed-forward assumption (§5) #
The Type A prediction for bivalent EXH depends on a feed-forward assumption: the strengthened (exclusive) truth conditions must be computed early enough to be visible to the projection computation. Without this, bivalent EXH + SK would not predict Type A because the projection computation would see only the original inclusive truth conditions. This assumption is architecturally non-trivial — see @cite{wang-davidson-2026} §5.
Empirical Contribution (§4) #
Mandarin experiment using huozhe ('or'), manipulating exclusivity via environmental monotonicity (UE → more exclusive, DE → less). Two presupposition triggers: jie 'quit' and zhidao 'know'.
Key result: PREDICATE × MONOTONICITY interaction is not significant (p = .30, BF₁₀ = 0.52). No evidence that exclusivity modulates filtering. This challenges Type A theories and is consistent with Type B (EXH¹).
Secondary finding: filtering (a)symmetry depends on trigger. jie shows asymmetric filtering (PREDICATE × ORDER: p = .01); zhidao shows symmetric filtering (uniform, p = .99 for PREDICATE).
Inclusive vs exclusive disjunction under Strong Kleene #
The fundamental asymmetry: inclusive disjunction can "see past" an undefined disjunct when the other is true. Exclusive cannot.
This single fact drives the Type A prediction for bivalent EXH + SK:
since InnocentExclusion.disj_exh_eq_exor shows Exh strengthens ∨ to ⊻,
and Truth3.xor_indet_iff shows ⊻ propagates undefinedness
unconditionally, exhaustification eliminates filtering.
Inclusive disjunction allows filtering: a true first disjunct absorbs the second's presupposition failure.
Exclusive disjunction does not filter: even when one disjunct is true, an undefined partner makes the result undefined.
The filtering contrast is symmetric: both join and xor are
commutative, so the direction doesn't matter for SK.
This is what @cite{kalomoiros-schwarz-2024} call "symmetric projection" — filtering is equally (un)available in both directions.
PrProp exclusive disjunction #
PrProp.xor requires both presuppositions to hold — it never
filters presupposition failure from either disjunct. This mirrors
the SK XOR truth table (Figure 2 in the paper).
Note: SK inclusive filtering (Truth3.join .true .indet = .true)
is an emergent property of the SK truth table, not a PrProp
connective. The contrast is between Truth3.join (filters) and
Truth3.xor (does not filter), verified in §1 above.
PrProp.xor does not filter: when q's presupposition fails,
the result is always undefined regardless of p.
Theory classification from §3.3: theories are grouped by their prediction about the effect of exclusivity on presupposition filtering across disjunction.
- typeA : TheoryClass
Increasing exclusivity reduces filtering.
- typeB : TheoryClass
Exclusivity has no effect on filtering.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Instances For
Exhaustification strategy: bivalent (@cite{fox-2007}) or trivalent (@cite{spector-sudo-2017}).
- bivalent : ExhStrategy
- exh1 : ExhStrategy
- exh2 : ExhStrategy
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Instances For
Semantic presupposition projection theory.
- strongKleene : ProjectionTheory
- george2008 : ProjectionTheory
- dynamicSemantics : ProjectionTheory
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Classify a combination of exhaustification + projection theory into Type A or Type B.
Type A (exclusivity reduces filtering):
- bivalent EXH + any of the three projection theories
- EXH² + any projection theory
Type B (no effect of exclusivity):
- EXH¹ + any projection theory
Equations
- One or more equations did not get rendered due to their size.
Instances For
EXH¹ is always Type B regardless of projection theory.
Bivalent EXH is always Type A regardless of projection theory.
EXH² is always Type A regardless of projection theory.
Bivalent EXH strengthens inclusive to exclusive #
The bridge from bivalent EXH to the SK prediction:
InnocentExclusion.disj_exh_eq_exor: Exh(Alt)(p∨q) = p ⊕ q- The exclusive truth conditions, when lifted to Truth3 via SK,
yield
Truth3.xor— which propagates#unconditionally - Therefore: bivalent EXH + SK → no filtering (Type A prediction)
This chain depends on the feed-forward assumption (§5): the strengthened exclusive truth conditions must be visible to the projection computation. Without it, projection would see only the original inclusive conditions and filtering would be unaffected.
Fox 2007 already proves that bivalent EXH strengthens inclusive to exclusive disjunction (reexported for context).
The classical exclusive disjunction (Bool XOR) agrees with Strong Kleene XOR on defined inputs.
Trivalent exhaustification and presupposition inheritance #
The bathroom disjunction verification from Trivalent.lean shows:
- EXH¹ preserves filtering at the critical world (
pOnly) - EXH² destroys filtering at the critical world (
pOnly)
This drives the Type A/B split. The mechanism:
- EXH¹ uses weak negation (
~# = true), so the conjunction alternative's undefinedness atpOnlyis harmlessly "negated" - EXH² uses strong negation (
~# = #), so the conjunction alternative's undefinedness propagates upward as a new presupposition requirement
Re-export: EXH¹ preserves filtering (Type B).
Re-export: EXH² destroys filtering (Type A).
Monotonicity environment, used as between-subjects factor. UE = unembedded disjunction (more exclusive readings); DE = disjunction in conditional antecedent (fewer exclusive readings).
- UE : Monotonicity
- DE : Monotonicity
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Instances For
Predicate type: whether test sentence contains presupposition trigger.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Order of trigger in disjunction.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Instances For
Presupposition triggers used in the experiment.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Instances For
Experimental finding summary.
Instances For
Norming task validation: the monotonicity manipulation successfully modulates exclusivity (Fisher's exact test, p = .011). UE: 23.3% exclusive responses; DE: 0% exclusive responses.
Equations
- Phenomena.Presupposition.Studies.WangDavidson2026.normingValidation = { description := "Norming task: exclusive responses UE 23.3% vs DE 0%, p = .011", significant := true }
Instances For
The critical null result: PREDICATE × MONOTONICITY is not significant. p = .30 (frequentist), BF₁₀ = 0.52 (Bayesian). No evidence that exclusivity modulates filtering.
Equations
- Phenomena.Presupposition.Studies.WangDavidson2026.mainResult = { description := "PREDICATE × MONOTONICITY interaction: p = .30, BF₁₀ = 0.52", significant := false }
Instances For
Control validation: the paradigm detects presuppositional definedness costs. CONTEXT manipulation is significant (p < .001).
Equations
- Phenomena.Presupposition.Studies.WangDavidson2026.controlValidation = { description := "CONTEXT (EI vs S) main effect: p < .001", significant := true }
Instances For
jie shows asymmetric filtering: PREDICATE × ORDER is significant (β = −1.81, SE = 0.72, p = .01). R-to-L filtering is weaker than L-to-R.
Equations
- Phenomena.Presupposition.Studies.WangDavidson2026.jieAsymmetry = { description := "jie PREDICATE × ORDER: β = −1.81, SE = 0.72, p = .01", significant := true }
Instances For
zhidao shows symmetric filtering: PREDICATE × ORDER is not significant (p = .40), and PREDICATE main effect p = .99 (uniform filtering).
Equations
- Phenomena.Presupposition.Studies.WangDavidson2026.zhidaoSymmetry = { description := "zhidao PREDICATE × ORDER: p = .40; PREDICATE: p = .99", significant := false }
Instances For
The norming task confirms the manipulation is effective.
The null result is consistent with Type B theories (EXH¹).
The null result challenges all Type A theories: three bivalent EXH + projection combinations and EXH² + any.
End-to-end argumentation chain: Fox 2007 computes exclusive truth conditions → SK propagates undefinedness → Type A predicted → experiment finds no effect → challenges bivalent EXH + SK.
This links InnocentExclusion.disj_exh_eq_exor, Truth3.xor_indet_iff,
the Type A classification, and the null experimental result.