Križ and Chemla (2015): finding truth-value gaps #
Križ and Chemla introduce two experimental methods for detecting truth-value gaps and use them to ask where the homogeneity gap of a plural definite survives embedding, under sentential negation, all, no and exactly 2. A display gives each of four cells, or boys, the trivalent value of the bare plural over its nine symbols, or presents, and a sentence is evaluated by its quantifier on the display resolved existentially or universally, the variants with some and with all in place of the definite. Gaps were found exactly where the two variants differ, which includes the GAP?? configuration of exactly 2 and excludes its GAP? configuration. Of the accounts the paper assesses, Spector's supervaluation over the two resolutions fits every finding; comparing the literal with the globally exhaustified meaning, after Magri, misses the gaps under negation, under no and in the GAP?? configuration; and the universal projection of a homogeneity presupposition predicts gaps where false was found.
Main results #
items_variants: every display of Table 13 realizes its condition, except the faulty A2 and B2 no items (faulty_items) and the GAP? item 5599 (misprinted_gapQQ).found_iff: Table 2 finds a gap exactly outside GAP?, apart from the discounted no tests.supervaluation_eq_one_iff_condition,supervaluation_eq_zero_iff_found: the supervaluation predicts every designed value and every finding.globalConstrual_misfit_iff: the implicature construal misfits exactly the downward-entailing GAP items and the GAP?? items;globalConstrual_wideScope: a wide-scope definite rescues it under plain negation.pointwise_ne_iff: supervaluating over per-boy resolutions departs exactly on GAP?.situations_variants: Table 12's parses and construals follow from its six situations.globalExh_iff_mem_strengthened: global exhaustification is Magri's double strengthening.
Implementation notes #
The stimuli and findings are the tables of Data/Experiments/KrizChemla2015. The E-neg tests reuse
the E-∅ displays with the negated sentence, whose designed values swap (negationItems), and
sentential negation is the outer negation notAll of the unembedded sentence. The weak and strong
variants are the some- and all-substituted readings, exchanged under a downward-entailing scope,
which is how §3's principle, stated for upward-entailing contexts, extends to negation and no.
The A2 and B2 no tests are discounted, as footnotes 10 and 14 do. The statistics are stored as
printed and not re-thresholded; the (si2) and (si4) construals are the supervaluation by Table
12's lit ≠ loc column.
TODO #
- The trivalent projection theory after George that §6.3 credits with matching the supervaluation is not implemented.
References #
Displays and their resolutions #
A display lists, for each boy, the trivalent value of he found his presents.
Equations
- KrizChemla2015.Display = List SignType
Instances For
A cell in which n of the nine objects are found or target-colored has the value of the
bare plural over the nine, true when all are, false when none are, and a gap otherwise.
Equations
- KrizChemla2015.cell n = Homogeneity.barePlural (fun (j m : ℕ) => j < m) (Finset.range 9) n
Instances For
A cell is true, designated at K3, when all nine objects are found.
A cell is non-false, designated at LP, when at least one object is found.
A number tree holds of a resolution according to how many boys it resolves out of and into the scope.
Equations
- KrizChemla2015.holds q c = q (List.count false c) (List.count true c)
Instances For
Reversing every boy's resolution evaluates the inner negation.
The uniform resolution of a display at a designation standard puts each boy in the scope when his value is designated.
Equations
- KrizChemla2015.resolve δ d = List.map (fun (v : SignType) => decide (δ.threshold ≤ v)) d
Instances For
Negating every cell resolves at the dual standard and reverses the result.
The resolutions of a display that choose a designation standard boy by boy.
Equations
- KrizChemla2015.resolutions d = (List.map (fun (v : SignType) => [decide (0 ≤ v), decide (v = 1)]) d).sections
Instances For
The K3 resolution is least and the LP resolution greatest among the resolutions.
A display without partial cells has a single resolution.
Moving boys into the scope preserves a scope-monotone quantifier.
The some- and all-substituted readings #
§3's guiding principle: a sentence with a definite plural has a gap in a situation where the variant with an existential in place of the definite is true while the variant with a universal is false.
A reading evaluates the quantifier over the uniform resolution at δ, which at LP
(non-false) is the existential resolution of the definite and at K3 (true) the universal one.
Equations
- KrizChemla2015.reading q d δ = KrizChemla2015.holds q (KrizChemla2015.resolve δ d)
Instances For
The universal quantifier holds of a resolution when every boy is in the scope.
The negative quantifier holds of a resolution when no boy is in the scope.
Negating the predicate in every cell evaluates the inner negation of the quantifier at the dual standard, so it exchanges the some- and all-substituted readings.
On a display without partial cells the two resolutions coincide.
Under a scope-monotone quantifier the universal variant entails the existential one.
Under a scope-antitone quantifier the existential variant entails the universal one.
Two-component verdicts #
The verdict from two meaning components is the supervaluation over the pair, clearly true when both hold, clearly false when neither does, and a truth-value gap when they conflict.
Equations
- KrizChemla2015.gapValue p q = SignType.supervaluation Finset.univ fun (b : Bool) => if b = true then p else q
Instances For
Two components that agree yield a bivalent verdict.
The approaches of §6 #
The globally double-exhaustified meaning, (30) and (39), is the conjunction of the some- and
all-substituted readings, [Mag14]'s double strengthening
(globalExh_iff_mem_strengthened). In a downward-entailing scope exhaustification is vacuous,
and the conjunction is then the literal meaning (globalExh_iff_of_scopeAntitone).
Equations
- KrizChemla2015.globalExh q d = (KrizChemla2015.someReading q d ∧ KrizChemla2015.allReading q d)
Instances For
Under a downward-entailing quantifier global exhaustification is vacuous.
The supervaluation of [Spe13b] (§6.2) makes the sentence true when it is true however the definite is resolved, existentially or universally, and false when it is false however it is resolved.
Equations
- KrizChemla2015.supervaluation q d = SignType.supervaluation Finset.univ (KrizChemla2015.reading q d)
Instances For
Equations
- KrizChemla2015.instDecidableGlobalExh q d = { decide := decide (KrizChemla2015.someReading q d) && decide (KrizChemla2015.allReading q d), reflects_decide := ⋯ }
The implicature construals (si1) and (si3) of §6.1.2, after [Mag09]'s oddness condition, put a gap where the literal and the globally exhaustified meaning conflict.
Equations
Instances For
Homogeneity as a presupposition projecting universally from the quantifier's scope, after
[Sch94], [Lob00] and [Gaj05] as assessed in §6.3, is the presupposition
that every cell is homogeneous weakly conjoined with the universal-force assertion. The
presupposition is the ∂ of [BK01], the join p ⊔ 0 with the gap.
Equations
- KrizChemla2015.universalPresupposition q d = (max (SignType.ofProp (∀ v ∈ d, v ≠ 0)) 0).meetWeak (SignType.ofProp (KrizChemla2015.allReading q d))
Instances For
Supervaluation over the resolutions that treat the partial cells one by one, a richer candidate set than the existential and universal readings, of the kind §6.2 says predicts a gap in the gap? condition; footnote 19 notes that quantified trivalent logics make the same prediction.
Equations
- KrizChemla2015.pointwise q d = SignType.supervaluation (KrizChemla2015.resolutions d).toFinset (KrizChemla2015.holds q)
Instances For
The supervaluation ranges over the two designation standards.
The supervaluation is true when both variants are.
The supervaluation is false when neither variant is true.
The supervaluation gaps when the variants differ.
A display without partial cells gets a bivalent verdict, whatever the quantifier.
The supervaluation does not see the scope of the definite relative to a negation inside the cells, since negating every cell gives the supervaluation of the inner negation.
The outer negation of the quantifier negates the supervaluation.
The global construal gaps where the literal meaning holds and the all-substituted one fails.
The global construals depart from supervaluation exactly where the literal meaning is false and the locally exhaustified meaning true.
In the scope of a scope-monotone quantifier such as every the implicature construals all align, since comparing the literal meaning with global exhaustification and with local exhaustification comes to the same thing (§6.1.3).
Without local exhaustification no gap can arise in the scope of a downward-entailing quantifier such as no, since exhaustification is vacuous there and the literal and globally exhaustified meanings never conflict.
Universal projection fails exactly on the displays with a partial cell.
Negation and the wide-scope parse (§6.1.3) #
Over one cell the unembedded sentence takes the cell's value.
The literal-against-global construal never gaps under plain negation, which is downward entailing.
On the parse (35) where the definite takes scope over negation, the literal-against-global construal of the shapes are not green is the supervaluation of its narrow-scope negation, so it gaps exactly on the mixed cell. No such parse is available under no, whose definite contains a variable the quantifier binds.
Global exhaustification is double strengthening #
The displays of n boys on which the some-substituted reading holds, the weak pole.
Equations
- KrizChemla2015.weakPole q n = {w : Fin n → SignType | KrizChemla2015.someReading q (List.ofFn w)}
Instances For
The displays of n boys on which the all-substituted reading holds, the strong pole.
Equations
- KrizChemla2015.strongPole q n = {w : Fin n → SignType | KrizChemla2015.allReading q (List.ofFn w)}
Instances For
Exhaustifying the sentence twice over its some- and all-substituted alternatives, by
[Mag14]'s double strengthening, yields globalExh, (30) and (39), provided either the strong
pole is weaker, as in a downward-entailing scope, or the two poles are compatible.
Per-boy resolutions #
The two-candidate supervaluation ranges over the uniform resolutions.
Richer candidates can only add gaps, since whatever the quantifier the per-boy supervaluation is at most as informative as the two-candidate one.
Under a scope-monotone quantifier the per-boy resolutions add no gap, the uniform ones being their extremes.
The outer negation of the quantifier negates the per-boy supervaluation.
Under a scope-antitone quantifier, likewise.
The experiments #
The number tree of each environment; sentential negation is the outer negation of the unembedded sentence over its one cell.
Equations
- KrizChemla2015.Embedding.unembedded.tree = Quantifier.NumberTree.all
- KrizChemla2015.Embedding.negation.tree = Quantifier.NumberTree.notAll
- KrizChemla2015.Embedding.all.tree = Quantifier.NumberTree.all
- KrizChemla2015.Embedding.no.tree = Quantifier.NumberTree.no
- KrizChemla2015.Embedding.exactly.tree = Quantifier.NumberTree.cardinal {KrizChemla2015.numeral}
Instances For
Equations
- KrizChemla2015.instDecidableRelNatTree KrizChemla2015.Embedding.unembedded = KrizChemla2015.instDecidableRelNatTree._aux_1
- KrizChemla2015.instDecidableRelNatTree KrizChemla2015.Embedding.all = KrizChemla2015.instDecidableRelNatTree._aux_3
- KrizChemla2015.instDecidableRelNatTree KrizChemla2015.Embedding.negation = KrizChemla2015.instDecidableRelNatTree._aux_5
- KrizChemla2015.instDecidableRelNatTree KrizChemla2015.Embedding.no = KrizChemla2015.instDecidableRelNatTree._aux_7
- KrizChemla2015.instDecidableRelNatTree KrizChemla2015.Embedding.exactly = KrizChemla2015.instDecidableRelNatTree._aux_9
Negation and no are downward entailing; the other environments are not.
Equations
- KrizChemla2015.instDecidableScopeAntitoneTree KrizChemla2015.Embedding.unembedded = isFalse KrizChemla2015.instDecidableScopeAntitoneTree._proof_1
- KrizChemla2015.instDecidableScopeAntitoneTree KrizChemla2015.Embedding.all = isFalse KrizChemla2015.instDecidableScopeAntitoneTree._proof_2
- KrizChemla2015.instDecidableScopeAntitoneTree KrizChemla2015.Embedding.negation = isTrue Quantifier.NumberTree.scopeAntitone_notAll
- KrizChemla2015.instDecidableScopeAntitoneTree KrizChemla2015.Embedding.no = isTrue Quantifier.NumberTree.scopeAntitone_no
- KrizChemla2015.instDecidableScopeAntitoneTree KrizChemla2015.Embedding.exactly = isFalse ⋯
The display of an item, each cell's trivalent value.
Equations
- i.display = List.map KrizChemla2015.cell i.cells
Instances For
The designed value of the negated sentence swaps TRUE and FALSE.
Equations
Instances For
The E-neg items are the E-∅ displays judged with the negated sentence.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The items judged in the experiments, Table 13's and the E-neg ones.
Instances For
The weak variant of the sentence, which the strong one entails, is the some-substituted reading, or under a downward-entailing scope the all-substituted one.
Equations
- KrizChemla2015.weak e d = if e.tree.ScopeAntitone then KrizChemla2015.allReading e.tree d else KrizChemla2015.someReading e.tree d
Instances For
The strong variant of the sentence.
Equations
- KrizChemla2015.strong e d = if e.tree.ScopeAntitone then KrizChemla2015.someReading e.tree d else KrizChemla2015.allReading e.tree d
Instances For
Equations
- KrizChemla2015.instDecidableWeak e d = id inferInstance
Equations
- KrizChemla2015.instDecidableStrong e d = id inferInstance
The GAP? item 5599 of Table 13, whose two full cells §3.3.2's description of the GAP? items excludes.
Equations
- KrizChemla2015.misprinted = { embedding := KrizChemla2015.Embedding.exactly, condition := KrizChemla2015.Condition.gapQ, cells := [5, 5, 9, 9], status := KrizChemla2015.ItemStatus.used }
Instances For
The item realizes the GAP?? pattern, its all-variant true and its some-variant false.
Every tested item realizes its condition by §3's principle, read with the entailment direction of its scope, the weak variant holding in the TRUE and GAP conditions and the strong one in the TRUE and GAP?? conditions.
A test is discounted when it is one of the A2 and B2 tests of no, whose sentences lacked negative inversion (fn 10) and were plausibly read with an unbound definite (fn 14).
Equations
Instances For
Equations
- One or more equations did not get rendered due to their size.
Table 2 found a gap in every GAP and GAP?? test and in no GAP? test, except in the discounted tests.
Table 2 tests only the gap conditions.
The supervaluation is true exactly on the variants' agreement in truth.
The supervaluation gaps exactly where the variants differ.
The supervaluation is true exactly on the items designed true.
The supervaluation gaps on an item exactly when its environment and condition were found to gap, outside the discounted tests.
The literal-against-global construal misfits exactly the GAP items under a downward-entailing scope, sentential negation and no, and the GAP?? items, where the all-variant is true and the some-variant false (§6.1.3).
Universal projection predicts a presupposition failure on a FALSE item of all, where the sentence was judged false as soon as one cell has no target symbols, the argument from (42).
The per-boy supervaluation departs from the two-candidate one on exactly the GAP? items.
The at-least reading of exactly 2 that §3.4 discusses.
Equations
- KrizChemla2015.atLeastTwo = Quantifier.NumberTree.cardinal {b : ℕ | 2 ≤ b}
Instances For
Equations
- KrizChemla2015.instDecidableRelNatAtLeastTwo x✝ b = { decide := Nat.ble 2 b, reflects_decide := ⋯ }
On the at-least reading the GAP? items are gap items.
On the at-least reading the FALSE items of exactly with three or more full cells are true, the items with elevated true responses in Fig. 10.
The display of a situation of Table 12, Bill's, Mary's and Sue's cells.
Equations
- s.display = List.map KrizChemla2015.Share.value [s.bill, s.mary, s.sue]
Instances For
Table 12 follows from its situations. The literal and locally exhaustified parses hold as its corresponding conditions require, the (si2) and (si4) construals, the supervaluation, gap exactly where a gap was found, and the (si1) and (si3) construals miss the GAP?? situation.