Cobreros, Égré, Ripley and van Rooij 2012: Tolerant, classical, strict #
A T-model is a classical model whose predicates each carry an indifference relation ∼_P, reflexive
and symmetric but possibly non-transitive (Definition 4), the vague predicate's "looks the same with
respect to P". Besides classical truth a T-model supports two further notions defined by
simultaneous induction (Definition 9): Pa is strictly true when every ∼_P-neighbour of a is
classically P and tolerantly true when some neighbour is, negation swapping the two, so that the
strict extension sharpens and the tolerant extension coarsens the classical one (Lemma 1) and the
tolerance principle Pa ∧ a I_P b → Pb is tolerantly valid (Fact 2) up to two similarity steps
(Fact 3). Borderline cases are the tolerantly-but-not-strictly P individuals (Definition 10), the
ones tolerantly both P and ¬P. On the similarity-free vocabulary tolerant validity is classical
validity while nothing is strictly valid (Theorems 1 and 2), so strict validity is not Fine's
supervaluationist validity, which is classical; and tolerant and strict consequence are Priest's LP
and Kleene's strong K3 (Theorem 3), via the model correspondences of Lemmas 4 and 5. Mixing the
three notions gives nine consequence relations ⊨ᵐⁿ (Definition 17), ordered by Lemma 7, dualised
by Lemma 6, collapsing to four on the restricted vocabulary (Lemmas 8 and 9) but distinct once
similarity statements are expressible (§3.4), where the deduction theorem singles out the self-dual
st, cc and ts (Lemma 10). A relation is reflexive iff its premise standard is at most its
conclusion standard, and closed under cut when it is at least (§3.3.2); ct, sc and st fail
even simple transitivity (§3.4.1). The paper's choice is st, strict premises to tolerant
conclusions: it validates tolerance and modus ponens and is non-transitive, so it validates every
step of a sorites while invalidating the chain (§3.6, Version 1), and it makes the tolerance-premise
sorites valid but unsound, its tolerance premise not being strictly true (Version 2). The five-men
model of §4.2.3 gives the paper's account of Alxatib and Pelletier's data: the median man is
borderline however he is classified.
Main definitions #
TModel,TModel.Realize,Mode: T-models and their three notions of truth, the modes ordered by strengthstrict < classical < tolerant.TModel.Borderline: Definition 10.Valid,Unsat,Consequence: Definitions 11, 12 and 17, the last multiple-conclusionConsequence.MixedConsequence.IsRestricted,TModel.identity,allBorderline,toMV,ofMV: the restricted vocabulary and the models of Lemmas 2, 3, 4 and 5.series: the sorites series of §3.4.1 and §3.6, whose instances are the four-element model of p. 354 and the five-men model of §4.2.3.
Main results #
TModel.realize_mono(Lemma 1),isDual(Remark 1),tolerance_valid(Fact 2),st_two_step(Fact 3).TModel.Borderline.exists_ne: a model with a borderline case has two (footnote 4);TModel.eq_tolerant_of_realize_conj_neg: a contradiction can only be tolerantly true.valid_tolerant_iff,unsat_strict_iff(Theorem 1);not_valid_strict,exists_realize_tolerant(Theorem 2);consequence_tolerant_iff_lp,consequence_strict_iff_k3(Theorem 3).consequence_iff_dual(Lemma 6),Consequence.mono(Lemma 7),Consequence.restricted_iff(Lemma 8),not_consequence_tolerant_strict(Lemma 9),Consequence.iff_valid_imp,deductionTheorem_iff,Consequence.iff_valid_imp_of_restricted(Lemma 10).not_cc_one_step,not_valid_classical_tolerance,not_sc_two_step,not_ct_two_step: the §3.4 distinctness ofcc,sc,ctandst, on the four-element model.stdRefl_consequence_iff,Consequence.cut,isTrans_consequence_iff: reflexivity and transitivity of the nine relations (§§3.3.2, 3.4.1).not_st_version1,st_version2,not_series_realize_strict_tolerance: the sorites of §3.6.alxatibPelletier_median_borderline: the five-men model of §4.2.3.
Implementation notes #
The language is the paper's propositional fragment over a set of constants C: negation and
conjunction over predications Pa and similarity statements a I_P b, with disjunction and
the conditional defined classically. The universally quantified principles appear as their
instances, which the paper notes changes nothing (§3.6). Identity, which the paper adds in
§2.3.1, is not represented. Consequence quantifies over T-models of every domain in the
universe of C (TModels), as the paper's does; the atomic strict and tolerant clauses are
ModalLogic.Box and ModalLogic.Diamond along ∼_P, so Lemma 1's atomic case is the T axiom.
References #
- P. Cobreros, P. Égré, D. Ripley and R. van Rooij, Tolerant, Classical, Strict (2012)
- S. Alxatib and F. J. Pelletier, The Psychology of Vagueness: Borderline Cases and Contradictions (2011)
- K. Fine, Vagueness, truth and logic (1975)
- S. C. Kleene, Introduction to Metamathematics (1952)
- G. Priest, The Logic of Paradox (1979)
T-models #
A T-model (Definition 4) is a classical model of the constants C and predicates Pred over
the domain D, with for each predicate a reflexive and symmetric, possibly non-transitive,
indifference relation ∼_P.
- const : C → D
The interpretation of the constants.
- interp : Pred → D → Prop
The classical extension of each predicate.
- sim : Pred → SetRel D D
The indifference relation
∼_Pof each predicate. - sim_ktb (P : Pred) : ModalLogic.IsKTBFrame (self.sim P)
Instances For
Equations
- CobrerosEtAl2012.instDecidableEqMode x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- CobrerosEtAl2012.instFintypeMode = { elems := { val := ↑CobrerosEtAl2012.Mode.enumList, nodup := CobrerosEtAl2012.Mode.enumList_nodup }, complete := CobrerosEtAl2012.instFintypeMode._proof_1 }
The duality d of Definition 20, with d(t) = s, d(s) = t and d(c) = c.
Equations
Instances For
Equations
Instances For
Equations
The propositional fragment of the language (Definition 2).
Equations
- CobrerosEtAl2012.Formula Pred C = Trivalent.Formula (CobrerosEtAl2012.Atom Pred C)
Instances For
The instance Pa ∧ a I_P b → Pb of the tolerance principle (1).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Truth in a mode (Definition 9, with Definition 5 for the classical mode). A predication is
strictly true when P holds throughout the ∼_P-neighbourhood of the individual, □, and
tolerantly true when it holds somewhere in it, ◇; negation swaps to the dual mode; and
similarity statements are classical in every mode (Remark 2).
Equations
- M.Realize x✝ (Trivalent.Formula.atom (CobrerosEtAl2012.Atom.sim P a b)) = ((M.const a, M.const b) ∈ M.sim P)
- M.Realize CobrerosEtAl2012.Mode.classical (Trivalent.Formula.atom (CobrerosEtAl2012.Atom.pred P a)) = M.interp P (M.const a)
- M.Realize CobrerosEtAl2012.Mode.strict (Trivalent.Formula.atom (CobrerosEtAl2012.Atom.pred P a)) = ModalLogic.Box (M.sim P) (M.interp P) (M.const a)
- M.Realize CobrerosEtAl2012.Mode.tolerant (Trivalent.Formula.atom (CobrerosEtAl2012.Atom.pred P a)) = ModalLogic.Diamond (M.sim P) (M.interp P) (M.const a)
- M.Realize x✝ φ.neg = ¬M.Realize x✝.dual φ
- M.Realize x✝ (φ.conj ψ) = (M.Realize x✝ φ ∧ M.Realize x✝ ψ)
- M.Realize x✝ (φ.disj ψ) = (M.Realize x✝ φ ∨ M.Realize x✝ ψ)
- M.Realize x✝ (φ.cond ψ) = (M.Realize x✝.dual φ → M.Realize x✝ ψ)
Instances For
The conjunction ⋀(φ, Γ) of a nonempty list.
The disjunction ⋁(φ, Δ) of a nonempty list.
Equations
- One or more equations did not get rendered due to their size.
- M.decidableRealize x✝ (Trivalent.Formula.atom (CobrerosEtAl2012.Atom.sim P a b)) = CobrerosEtAl2012.TModel.decidableRealize._aux_1 M x✝ P a b
- M.decidableRealize CobrerosEtAl2012.Mode.classical (Trivalent.Formula.atom (CobrerosEtAl2012.Atom.pred P a)) = CobrerosEtAl2012.TModel.decidableRealize._aux_3 M P a
- M.decidableRealize x✝ φ.neg = instDecidableNot
- M.decidableRealize x✝ (φ.conj ψ) = instDecidableAnd
- M.decidableRealize x✝ (φ.disj ψ) = instDecidableOr
- M.decidableRealize x✝ (φ.cond ψ) = { decide := if h : M.Realize x✝.dual φ then decide ((fun (a : M.Realize x✝.dual φ) => M.Realize x✝ ψ) h) else true, reflects_decide := ⋯ }
A contradiction is true only tolerantly (§4.1).
Excluded middle fails only strictly (§2.4).
Borderline cases #
d is a borderline case of P iff it is tolerantly but not strictly P (Definition 10).
Equations
- M.Borderline P d = (ModalLogic.Diamond (M.sim P) (M.interp P) d ∧ ¬ModalLogic.Box (M.sim P) (M.interp P) d)
Instances For
Equations
- One or more equations did not get rendered due to their size.
A borderline case is one indifferent both from a P and from a non-P individual (p. 365).
A borderline case is tolerantly both P and ¬P (Definition 10).
A borderline case is neither strictly P nor strictly ¬P (Definition 10).
Excluded middle fails strictly at a borderline case (§2.4).
By symmetry, a model with a borderline case has at least two (footnote 4).
Consequence #
The T-models of the language over every domain, which validity and consequence quantify over (Definitions 11 and 17).
Equations
- CobrerosEtAl2012.TModels Pred C = ((D : Type ?u.2) × CobrerosEtAl2012.TModel Pred C D)
Instances For
φ is n-valid iff every T-model n-satisfies it (Definition 11).
Equations
- CobrerosEtAl2012.Valid n φ = ∀ (M : CobrerosEtAl2012.TModels Pred C), M.Realize n φ
Instances For
φ is n-unsatisfiable iff no T-model n-satisfies it (Definition 12).
Equations
- CobrerosEtAl2012.Unsat n φ = ∀ (M : CobrerosEtAl2012.TModels Pred C), ¬M.Realize n φ
Instances For
Γ ⊨ᵐⁿ Δ iff every T-model that m-satisfies all of Γ n-satisfies some member of Δ
(Definition 17).
Equations
- CobrerosEtAl2012.Consequence m n Γ Δ = Consequence.MixedConsequence (fun (x : CobrerosEtAl2012.TModels Pred C) => x.Realize m) (fun (x : CobrerosEtAl2012.TModels Pred C) => x.Realize n) Γ Δ
Instances For
mn-validity is n-validity (Definition 18).
mn-unsatisfiability is m-unsatisfiability (Definition 18).
φ is n-unsatisfiable iff ¬φ is d(n)-valid (Remark 1, for unsatisfiability).
Consequence grows as premises are held to a stronger standard and conclusions to a weaker one (Lemma 7).
Negation swaps tolerant and strict truth and fixes classical truth, so each notion of truth
is dual to its image under d (Remark 1, Definition 20).
⊨ᵐⁿ is the dual of ⊨^{d(n)d(m)}, so Γ ⊨ᵐⁿ Δ iff ¬Δ ⊨^{d(n)d(m)} ¬Γ (Lemma 6).
st, cc and ts, the relations with m = d(n), are self-dual (§3.1).
In sequent form, the deduction theorem for self-dual relations moves a premise into the
antecedent of a conclusion when m = d(n).
Premises are read conjunctively and conclusions disjunctively, so φ, Γ ⊨ᵐⁿ ψ, Δ iff
⋀(φ, Γ) ⊨ᵐⁿ ⋁(ψ, Δ).
When m = d(n), Γ ⊨ᵐⁿ Δ iff ⊨ᵐⁿ ⋀Γ → ⋁Δ, for the nonempty finite sides the language
can conjoin and disjoin (Lemma 10, right to left; footnote 13).
Cut holds when the premise standard is at least the conclusion standard, so tc and cs
are transitive (§3.3.2, footnote 11).
Transitivity in the simple form of footnote 14, when n ≤ m (§3.3.2).
Tolerance #
The tolerance principle is tolerantly valid (Fact 2).
{Pa, a I_P b} ⊨ˢᶜ Pb (§3.4).
{Pa, a I_P b} ⊨ᶜᵗ Pb (§3.4).
Each step of a sorites is st-valid (§3.6).
Tolerance holds up to two similarity steps (Fact 3).
The restricted vocabulary #
A formula is in the restricted vocabulary (§2) when it contains no similarity statements.
Equations
- CobrerosEtAl2012.IsRestricted (Trivalent.Formula.atom (CobrerosEtAl2012.Atom.pred a a_1)) = True
- CobrerosEtAl2012.IsRestricted (Trivalent.Formula.atom (CobrerosEtAl2012.Atom.sim a a_1 a_2)) = False
- CobrerosEtAl2012.IsRestricted φ.neg = CobrerosEtAl2012.IsRestricted φ
- CobrerosEtAl2012.IsRestricted (φ.conj ψ) = (CobrerosEtAl2012.IsRestricted φ ∧ CobrerosEtAl2012.IsRestricted ψ)
- CobrerosEtAl2012.IsRestricted (φ.disj ψ) = (CobrerosEtAl2012.IsRestricted φ ∧ CobrerosEtAl2012.IsRestricted ψ)
- CobrerosEtAl2012.IsRestricted (φ.cond ψ) = (CobrerosEtAl2012.IsRestricted φ ∧ CobrerosEtAl2012.IsRestricted ψ)
Instances For
The identity model agrees classically with M on the restricted vocabulary.
On the restricted vocabulary, tolerant validity is classical validity (Theorem 1).
On the restricted vocabulary, strict unsatisfiability is classical unsatisfiability (Theorem 1).
The model of Lemma 3 has an individual for each constant plus one more, all indifferent, the extra one alone falling outside every predicate, so that every predication is borderline.
Equations
- CobrerosEtAl2012.allBorderline Pred C = { const := some, interp := fun (x : Pred) (d : Option C) => d.isSome = true, sim := fun (x : Pred) => Set.univ, sim_ktb := ⋯ }
Instances For
In allBorderline nothing restricted is strictly true and everything restricted is
tolerantly true (Lemma 3).
No restricted formula is strictly valid (Theorem 2).
Every restricted formula is tolerantly satisfiable (Theorem 2).
On the restricted vocabulary st-consequence is classical consequence, the identity model
of each classical countermodel being an st-countermodel (Lemma 8).
On the restricted vocabulary cc, sc, ct and st coincide (Lemma 8).
On the restricted vocabulary ts-consequence is empty (Lemma 9).
LP and K3 #
In the MV-model of Lemma 4, a predication is 1 when strictly true, -1 when not
tolerantly true and 0 when borderline, and similarity statements are two-valued.
Equations
- CobrerosEtAl2012.toMV M (CobrerosEtAl2012.Atom.pred P a) = if ModalLogic.Box (M.sim P) (M.interp P) (M.const a) then 1 else if ModalLogic.Diamond (M.sim P) (M.interp P) (M.const a) then 0 else -1
- CobrerosEtAl2012.toMV M (CobrerosEtAl2012.Atom.sim P a b) = if (M.const a, M.const b) ∈ M.sim P then 1 else -1
Instances For
The T-model of Lemma 5 lives on the doubled domain, where (a, false) is a and
(a, true) its double a', indifferent from each other and from nothing else; a is P when
v makes Pa true and a' when v makes it non-false.
Equations
- One or more equations did not get rendered due to their size.
Instances For
On the restricted vocabulary, LP-truth in v is tolerant truth and K3-truth strict truth in
ofMV v (Lemma 5).
On the restricted vocabulary, tolerant consequence is LP-consequence, by the model correspondences of Lemmas 4 and 5 (Theorem 3).
On the restricted vocabulary, strict consequence is K3-consequence (Theorem 3).
The deduction theorem #
In this model on the constants, a alone is P and everything is indifferent, so tolerance
from a to another constant fails classically.
Equations
- CobrerosEtAl2012.cutAt a = { const := id, interp := fun (x : Pred) (d : C) => d = a, sim := fun (x : Pred) => Set.univ, sim_ktb := ⋯ }
Instances For
The deduction theorem holds for ⊨ᵐⁿ iff m = d(n), so for st, cc and ts only
(Lemma 10). Validity depends only on n, so the theorem makes ⊨ᵐⁿ and ⊨^{d(n)n} agree;
against the conclusion ¬ a I_P a, which nothing satisfies, they are m- and
d(n)-unsatisfiability, which a contradiction and the negated tolerance principle separate.
On the restricted vocabulary the deduction theorem also holds for sc and ct, which there
coincide with cc and st (§3.4).
Sorites series #
The sorites series of §3.4.1 and §3.6 has n individuals in a line, each indifferent from its
neighbours, the first i classically P. series 4 2 is the four-element model of p. 354,
a, b, c, d being 0, 1, 2, 3 and the classical extension {a, b}.
Equations
- CobrerosEtAl2012.series n i = { const := id, interp := fun (x : Unit) (x_1 : Fin n) => ↑x_1 < i, sim := fun (x : Unit) => {p : Fin n × Fin n | ↑p.1 ≤ ↑p.2 + 1 ∧ ↑p.2 ≤ ↑p.1 + 1}, sim_ktb := ⋯ }
Instances For
Equations
- CobrerosEtAl2012.instDecidableMemProdFinSetRelSimUnitSeriesMk n i P x y = { decide := decide (↑x ≤ ↑y + 1) && decide (↑y ≤ ↑x + 1), reflects_decide := ⋯ }
Equations
- CobrerosEtAl2012.instDecidablePredFinInterpUnitSeries n i P x = { decide := (↑x).succ.ble i, reflects_decide := ⋯ }
In a series with a cut the borderline cases are the two individuals either side of it.
In the four-element model of p. 354 the strict, classical and tolerant extensions of P
are {a}, {a, b} and {a, b, c}, so b and c are borderline.
b is tolerantly both P and ¬P in the four-element model (p. 356).
{Pa, a I_P b} ⊨ Pb is not cc-valid (§3.4), as b, c of the four-element model show.
Tolerance is not classically valid, hence neither cc- nor sc-valid (§3.4).
{Pa, a I_P b, b I_P c} ⊨ Pc is not sc-valid (§3.4), as a, b, c of the four-element
model show.
{Pa, a I_P b, b I_P c} ⊨ Pc is not ct-valid (§3.4), as b, c, d of the four-element
model show.
When n < m, some T-model makes Pa m-true but not n-true, so the three notions of truth
come apart on every predication (§3.3.2).
⊨ᵐⁿ is reflexive iff m ≤ n (§3.3.2), so φ ⊨ᵗᵗ φ and φ ⊨ˢˢ φ but Pa ⊭ᵗᶜ Pa and
Pa ⊭ᶜˢ Pa.
ct is not transitive (§3.4.1), since along b, c, d of the four-element model,
Pb ∧ b I_P c ∧ c I_P d ⊨ᶜᵗ Pc ∧ c I_P d and Pc ∧ c I_P d ⊨ᶜᵗ Pd but
Pb ∧ b I_P c ∧ c I_P d ⊭ᶜᵗ Pd.
sc is not transitive (§3.4.1), since along a, b, c of the four-element model,
Pa ∧ a I_P b ∧ b I_P c ⊨ˢᶜ Pb ∧ b I_P c and Pb ∧ b I_P c ⊨ˢᶜ Pc but
Pa ∧ a I_P b ∧ b I_P c ⊭ˢᶜ Pc.
st is not transitive (§3.4.1). By Fact 3 a counterexample takes three similarity steps,
along the whole four-element model, where
Pa ∧ a I_P b ∧ b I_P c ∧ c I_P d ⊨ˢᵗ Pb ∧ b I_P c ∧ c I_P d and
Pb ∧ b I_P c ∧ c I_P d ⊨ˢᵗ Pd but Pa ∧ a I_P b ∧ b I_P c ∧ c I_P d ⊭ˢᵗ Pd.
On the sorites language the transitive relations among the nine are exactly those whose
conclusion standard is at most their premise standard, namely tc, cs and ts besides the
unmixed three (§3.3.2), but not ct, sc or st (§3.4.1).
The similarity premises a₁ I_P a₂, …, aₙ I_P aₙ₊₁ of the sorites (§3.6).
Equations
- CobrerosEtAl2012.steps n = List.map (fun (k : Fin n) => Trivalent.Formula.atom (CobrerosEtAl2012.Atom.sim () k.castSucc k.succ)) (List.finRange n)
Instances For
The tolerance premises Pa₁ ∧ a₁ I_P a₂ → Pa₂, … of the sorites, Version 2 (§3.6).
Equations
- CobrerosEtAl2012.tolerances n = List.map (fun (k : Fin n) => CobrerosEtAl2012.tolerance () k.castSucc k.succ) (List.finRange n)
Instances For
Version 1 of the sorites is st-invalid for every series of at least four members, the
four-element model being a countermodel, although each of its steps is st-valid
(st_one_step). Below four members it is valid by Facts 2 and 3.
Version 2 of the sorites, with tolerance as a premise, is st-valid, since it is classically
valid and st extends classical consequence.
Version 2 is unsound, since across the cut of any series the tolerance premise is not strictly true (§3.6).
[AP11]'s five men are ranked by height (§4.2.3), #1 < #4 < #2 < #5 < #3
as ranks 0 to 4, with #1 and #4 tall and #5 and #3 not. Whichever way the median
man #2, rank 2, is classified, series 5 2 or series 5 3, he is borderline, tolerantly
tall and tolerantly not tall.