Križ and Spector (2021): Interpreting Plural Predication #
Križ and Spector account for homogeneity and non-maximality as follows. A sentence with a plural
definite is underspecified between a family of candidate interpretations: the definite is read as an
existential quantifier over a convex set of pluralities generated by some parts of its denotation
x together with x itself (Cand, the paper's (21)). The sentence is true when every candidate
is true, false when every candidate is false, and undefined otherwise (TrueOnAll, FalseOnAll).
The candidate interpretations are the library's readings of a plural predication
(image_Cand_eq_readings), and on a distributive predicate the resulting value is Križ's trivalent
plural (value_reading_distributive). Non-maximality comes from pragmatics: only the candidates
strongly relevant to the current issue count (TrueOnRelevant), so that under an issue that lumps
reading eight books with reading all of them, Mary read the books conveys that she read at least
eight (trueOnRelevant_reading) and its negation that she read fewer than three
(trueOnRelevant_not_reading).
The compositional half of the paper interprets predicates relative to a homogeneity
parameter HParam assigning a candidate domain to every argument index and plurality, and
treats all as universal quantification over the parameters that differ from the current
one only at its indices (all, the paper's (71)). This removes homogeneity and
non-maximality on the associated argument while leaving other arguments alone
(all_pred, all_pred₂), makes the girls all didn't dance mean that no girl danced
(all_not_pred), renders floating all vacuous under a quantifier that already
neutralizes the parameter (all_all), collapses coordinated predicates to the strong
reading (all_and), and does not reach a pronoun bound by its associate
(all_binding). The set of candidate interpretations is the same whether negation scopes
above or below the distributivity operator (interpretations_not_dist, the paper's
footnote proof by disjunctive normal form).
The comparison with Križ's earlier account is the paper's own (§4): Križ's Addressing condition has
strong relevance as its special case on bivalent propositions
(addressesIssue_iff_stronglyRelevant), and on the exam issue of the paper's (36) Križ's
account communicates that Mary passed while no candidate is strongly relevant on the
present account (communicatedContent_solvedMath, not_felicitous_solvedMath). The
appendix's candidates with superpluralities (CandUp) yield the upward-homogeneity falsity
condition and coincide with Križ's generalised homogeneity (valueUp_eq_generalisedTruthValue).
Implementation notes #
Pluralities are Finset Atom ordered by inclusion, so parts are nonempty subsets, the
convex closure of (21) is a union of Finset.Icc intervals, and candidate domains are
finite. A candidate interpretation is presented by its domain; reading D P is the
existential quantifier over D applied to the predicate P on pluralities. Sentence
frames take the domain as an argument (TrueOnAll x φ), which covers negation and
non-monotonic embeddings such as the paper's (22). Strong relevance is the substrate's
Homogeneity.isStronglyRelevantProp, constancy on the cells of the issue, the paper's
gloss of (25). The reading-list example (26)–(28) is stated for any thresholds
0 < lo ≤ hi ≤ |books| in place of the paper's 3, 8, and 10.
TODO #
- The paper's §2.2 critique of Malamud's decision-theoretic account and its §6 open problems (functional non-maximality, overt questions) are not formalized.
- The appendix's revision of all in terms of maximal individual witnesses, (117)–(119),
is not formalized;
CandUpcovers only the candidate side of the appendix.
References #
Candidate denotations (§3.1) #
The nonempty mereological parts of a plurality.
Equations
- KrizSpector2021.parts x = Finset.Ioc ⊥ x
Instances For
The convex closure of a set of pluralities collects everything in between two of its members, the paper's (21).
Equations
- KrizSpector2021.conv A = A.biUnion fun (a : Finset Atom) => A.biUnion fun (b : Finset Atom) => Finset.Icc a b
Instances For
The candidate denotations of x, the paper's (21), are the existential quantifiers over
the
convex closure of a set of parts of x together with x, presented by their domains.
Equations
- KrizSpector2021.Cand x = Finset.image (fun (S : Finset (Finset Atom)) => KrizSpector2021.conv (insert x S)) (KrizSpector2021.parts x).powerset
Instances For
Every candidate is entailed by the universal one, since x lies in every domain.
The universal candidate, from the empty set of parts.
The existential candidate, from all the parts.
The two candidates of [Kri96], the universal and the existential reading (§2.1).
Equations
- KrizSpector2021.twoCand x = {{x}, KrizSpector2021.parts x}
Instances For
Truth on all readings (§2.1, §3.1) #
The reading of the predicate P on pluralities under the candidate domain D holds when
some member of D satisfies P.
Equations
- KrizSpector2021.reading D P w = ∃ z ∈ D, P z w
Instances For
The candidate interpretations, (21), are the library's readings of the plural predication,
each candidate domain being the trace on the parts of x of an upper set containing x.
Equations
- KrizSpector2021.instDecidableReading D P w = { decide := decide (∃ (a : Finset Atom) (_ : a ∈ D), (fun (z : Finset Atom) => P z w) a), reflects_decide := ⋯ }
A sentence frame φ sends the candidate domain chosen for the plural definite to the
resulting candidate interpretation. The sentence is true when every candidate is.
Equations
- KrizSpector2021.TrueOnAll x φ w = ∀ D ∈ KrizSpector2021.Cand x, φ D w
Instances For
The sentence is false when every candidate is.
Equations
- KrizSpector2021.FalseOnAll x φ w = ∀ D ∈ KrizSpector2021.Cand x, ¬φ D w
Instances For
Equations
- One or more equations did not get rendered due to their size.
Equations
- One or more equations did not get rendered due to their size.
The trivalent value of the sentence is true, false, or undefined on its candidates.
Equations
- KrizSpector2021.value x φ w = if KrizSpector2021.TrueOnAll x φ w then 1 else if KrizSpector2021.FalseOnAll x φ w then -1 else 0
Instances For
Negating the sentence negates its candidates — falsity on all readings of φ is truth on
all readings of ¬ φ, the paper's (61)–(62).
Plural predication is true on all readings iff the predicate holds of the whole plurality.
Plural predication is false on all readings iff the predicate fails of every part.
On a distributive predicate, the meta-language distributivity operator of the paper's (63a), the value on the candidates is [Kri16]'s trivalent plural, true when every atom satisfies the predicate and false when none does.
Non-monotonic contexts (§3.1) #
The paper's (22): Mary read all of the books, Sue read half of them and Peter the other half. On the two candidates of [Kri96], Exactly two of the students read the books is false; the candidate generated by the half Peter read, the paper's (23), is true while the universal candidate is false, so on the enriched candidate set the sentence is neither true nor false.
Equations
- KrizSpector2021.instDecidableEqStudent x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- KrizSpector2021.instDecidableEqBook x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
The books.
Equations
Instances For
The scenario records which books each student read.
Equations
- KrizSpector2021.read₀ KrizSpector2021.Student.mary = {KrizSpector2021.Book.a, KrizSpector2021.Book.b, KrizSpector2021.Book.c, KrizSpector2021.Book.d}
- KrizSpector2021.read₀ KrizSpector2021.Student.sue = {KrizSpector2021.Book.c, KrizSpector2021.Book.d}
- KrizSpector2021.read₀ KrizSpector2021.Student.peter = {KrizSpector2021.Book.a, KrizSpector2021.Book.b}
Instances For
Exactly two of the students read the books under the candidate domain D, at a world
w recording which books each student read.
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.
Both of Krifka's candidates are false in the scenario.
The universal candidate is false, since only Mary read all the books.
The candidate generated by Peter's half is true, since exactly two students read a plurality containing both of those books, the paper's (23).
Strong relevance and non-maximality (§3.2) #
The sentence is judged true when every strongly relevant candidate is true.
Equations
- KrizSpector2021.TrueOnRelevant q x φ w = ∀ D ∈ KrizSpector2021.Cand x, Homogeneity.isStronglyRelevantProp q (φ D) → φ D w
Instances For
The sentence is judged false when every strongly relevant candidate is false.
Equations
- KrizSpector2021.FalseOnRelevant q x φ w = ∀ D ∈ KrizSpector2021.Cand x, Homogeneity.isStronglyRelevantProp q (φ D) → ¬φ D w
Instances For
Felicity requires some strongly relevant candidate (the paper's fn. 17).
Equations
- KrizSpector2021.Felicitous q x φ = ∃ D ∈ KrizSpector2021.Cand x, Homogeneity.isStronglyRelevantProp q (φ D)
Instances For
Without a strongly relevant candidate the sentence is judged both true and false.
Truth on all readings at an issue-equivalent world gives truth on the relevant ones.
Under a monotone-decreasing predicate the reading generated by a set of parts is the disjunction over those parts.
The reading list (§3.2) #
The paper's (26)–(28): reading hi or more of the books is as good as reading all of them
and reading fewer than lo is as bad as reading none. A world records which books Mary
read.
Mary read the plurality z at w when every book in z is among those she read.
Equations
- KrizSpector2021.readP z w = (z ⊆ w)
Instances For
The issue (26) asks whether Mary read at least lo of the books, and whether at least
hi.
Equations
- KrizSpector2021.issue lo hi = Setoid.ker fun (w : Finset Book) => (decide (lo ≤ w.card), decide (hi ≤ w.card))
Instances For
The candidate generated by the n-element parts reads as Mary read n or more.
Mary read the books, judged on its strongly relevant candidates, conveys that she read
hi or more of them, the paper's (27).
Mary didn't read the books conveys that she read fewer than lo of them, the paper's
(28), since the negated candidates that are strongly relevant are the negations of the relevant
ones and their conjunction is the bottom cell.
Comparison with Križ (2016) (§4) #
Križ's Addressing condition, the paper's (34a), has strong relevance as its special case on bivalent propositions.
The paper's (36): to pass, Mary must solve all of the math problems, or some of them and a physics problem, and the issue is whether she passed. On Križ's account Mary solved the math problems addresses the issue and communicates that she passed; on the present account no candidate is strongly relevant, so the sentence cannot address this issue at all.
A world records the math problems Mary solved, and whether she solved a physics problem.
Equations
- KrizSpector2021.ExamWorld Math = (Finset Math × Bool)
Instances For
Passing requires all of the math problems, or some of them and a physics problem.
Equations
- KrizSpector2021.Pass w = (w.1 = Finset.univ ∨ w.1.Nonempty ∧ w.2 = true)
Instances For
Equations
- KrizSpector2021.instDecidablePass w = { decide := decide (w.1 = Finset.univ) || decide (w.1.Nonempty ∧ w.2 = true), reflects_decide := ⋯ }
The issue asks whether Mary passed.
Equations
- KrizSpector2021.passFail = Setoid.ker fun (w : KrizSpector2021.ExamWorld Math) => decide (KrizSpector2021.Pass w)
Instances For
Mary solved the math problems as Križ's trivalent sentence.
Equations
- KrizSpector2021.solvedMath = Homogeneity.barePlural (fun (m : Math) (w : KrizSpector2021.ExamWorld Math) => m ∈ w.1) Finset.univ
Instances For
Križ's sentence addresses the pass/fail issue, since its true worlds pass and its false worlds fail.
On Križ's account the sentence communicates that Mary passed.
On the present account no candidate interpretation is strongly relevant to the pass/fail issue, since the universal candidate splits the passing worlds and any other candidate splits the failing ones.
Homogeneity parameters and all (§5.3) #
A homogeneity parameter assigns a candidate domain to every argument index and plurality (§5.3.1).
Equations
- KrizSpector2021.HParam Atom = (ℕ → Finset Atom → Finset (Finset Atom))
Instances For
Two parameters agree except at the indices in I, the paper's (70).
Equations
- KrizSpector2021.HParam.EquivExcept I H H' = ∀ k ∉ I, ∀ (x : Finset Atom), H k x = H' k x
Instances For
A parameter is admissible when each of its values is a candidate for its plurality.
Equations
- H.Admissible = ∀ (i : ℕ) (x : Finset Atom), H i x ∈ KrizSpector2021.Cand x
Instances For
A one-place predicate with argument index i, the paper's (53a).
Equations
- KrizSpector2021.pred H i P x w = KrizSpector2021.reading (H i x) P w
Instances For
A two-place predicate with argument indices i and j, the paper's (53b).
Equations
- KrizSpector2021.pred₂ H i j R a b w = ∃ z ∈ H i a, ∃ y ∈ H j b, R z y w
Instances For
The distributivity operator with argument index i, the paper's (63b).
Equations
- KrizSpector2021.dist H i Q x w = KrizSpector2021.pred H i (fun (z : Finset Atom) (w : W) => ∀ a ∈ z, Q a w) x w
Instances For
all with indices I, the paper's (71), is true relative to every admissible parameter
that agrees with the current one outside I.
Equations
- KrizSpector2021.all I φ H x w = ∀ (H' : KrizSpector2021.HParam Atom), H'.Admissible → KrizSpector2021.HParam.EquivExcept I H H' → φ H' x w
Instances For
The candidate interpretations of a sentence collect one proposition per admissible parameter.
Equations
- KrizSpector2021.interpretations φ = {p : W → Prop | ∃ (H : KrizSpector2021.HParam Atom), H.Admissible ∧ p = φ H}
Instances For
all is idempotent, so floating all is vacuous under a quantifier that already neutralizes the parameter, the paper's (92) and (97).
all removes homogeneity and non-maximality on its argument, the sentence holding iff the predicate holds of the whole plurality, whatever the parameter, the paper's (73)–(74).
With all above negation, the paper's (78), the girls all didn't dance holds iff no part of the girls danced.
all is selective, the paper's (82), so with all on the subject of a two-place predicate the object still ranges over its candidate.
With all over coordinated predicates, the paper's (104), the professors all smiled and left holds iff both predicates hold of the whole plurality.
all does not reach a pronoun bound by its associate, the paper's (108), so the chemists all met a linguist they like reads the subject universally while the pronoun keeps its candidate.
The negation of a distributive candidate is a distributive candidate for the negated predicate, with the domain generated by the transversals of the original parts, after the paper's footnote proof through disjunctive normal form.
Negation and distributivity commute in the set of candidate interpretations, the paper's
(69), so scoping negation below or above dist yields the same propositions.
Upward homogeneity (§3.3, Appendix) #
The appendix's candidates (113) are generated by parts of x together with pluralities
containing x; the sentence is then false only when no plurality overlapping x satisfies
the predicate, which is [Kri16]'s generalised homogeneity.
The pluralities containing x.
Equations
- KrizSpector2021.sups x = Finset.Icc x ⊤
Instances For
The candidate domains of the appendix, (113), are convex closures of some parts of x
and
some pluralities containing x, restricted to those containing x itself.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The paper's candidates are the appendix's with x as the only superplurality.
Every member of an appendix candidate overlaps x.
The weakest appendix candidate admits every plurality overlapping x.
Falsity on all appendix readings is the upward-homogeneity condition of the paper's
(114b), that no plurality overlapping x satisfies the predicate.
The trivalent value on the appendix's candidates.
Equations
- KrizSpector2021.valueUp x P w = if ∀ D ∈ KrizSpector2021.CandUp x, KrizSpector2021.reading D P w then 1 else if ∀ D ∈ KrizSpector2021.CandUp x, ¬KrizSpector2021.reading D P w then -1 else 0
Instances For
The appendix's candidates give [Kri16]'s generalised homogeneity over all pluralities.