Bar-Lev (2021): An Implicature Account of Homogeneity and Non-Maximality #
Bar-Lev derives homogeneity as an implicature. A plural definite means existential
pluralization, that some atom of the plurality in a domain has the property, and its alternatives
replace the domain by its subsets; each alternative is one of the readings of the plural
predication (supClosure_subset_readings). Exhaustifying over them gives the universal
reading, while under negation exhaustification is vacuous. Pruning the alternatives to subdomains
of a fixed size gives the threshold readings of non-maximality, which are readings of the plural
predication as well (readings_subset_readings); the question the context raises selects among
them, and a world is gappy when it verifies the reading one question selects and falsifies
another's.
Main results #
exhIEII_eq_universal,exhIEII_compl_biUnion: the maximality implicature, and its absence under negation.exhIEII_pruned_eq_atLeast,isReading_polar_atLeast: pruning yields the threshold readings, and a polar question on a threshold selects it.gappy_iff_barePlural_eq_zero: the gappy worlds are those where Križ's trivalent plural is undefined;neg_barePlural_eq_zero_and_notMem_exhIEII_compl: the accounts part under negation.gappy_iff_krizChemla_gap,krizChemla_negation_residue: on Križ and Chemla's unembedded displays the account predicts their gaps for the positive sentence and misses those under negation.disjunction_not_both,plural_both,reading_66,reading_80,reading_81: the paper's examples.
Implementation notes #
readings are the threshold readings, obtained by pruning to a fixed subdomain size; the
paper's asymmetric conjunctions of alternatives are omitted, so IsReading agrees with the
paper's selection on the partitions whose only decided readings are thresholds, which
includes the polar questions of (66), (80) and (81). The maximality constraint on prunings
is not formalized. Gappy ranges over all partitions and is context-free; that a context
fixing the partition removes the gap is carried by reading_80 and reading_81.
TODO #
- The negative sentence is modelled only at the LF with negation over
∃-PL; the LF with∃-PLover negation and the cover-restricted∃-PLof §5.3, the paper's source of the marginal negative non-maximal readings and of Križ and Chemla's negative gaps, are not formalized. - The paper's full set of readings, and the non-distributive extension of §8.
References #
Existential pluralization and its subdomain alternatives #
The existential pluralization ∃-PL_D of P over x is the disjunction of the atoms of x in
the domain D. A subdomain D' ⊆ D leaves the atoms of x ∩ D', so the subdomain alternatives
are the closure of the atoms under disjunction, supClosure (atom P '' ↑(x ∩ D)).
Equations
The subdomain alternatives are readings of the distributive plural predication, since the sub-disjunctions of the atoms lie in the lattice they generate.
Every atom laughs alone at some world, and all laugh together at some world.
- single (a : Atom) : a ∈ x ∩ D → ∃ (w : W), ∀ b ∈ x ∩ D, P b w ↔ b = a
Each atom of
x ∩ DsatisfiesPalone at some world. - all : ∃ (w : W), ∀ a ∈ x ∩ D, P a w
All atoms of
x ∩ DsatisfyPtogether at some world.
Instances For
Homogeneity #
Exhaustifying ∃-PL over its subdomain alternatives gives the universal reading, the
maximality implicature.
With negation over ∃-PL every negated subdomain alternative is entailed by the negated
prejacent, so exhaustification over any of them is vacuous and no pruning yields a non-maximal
reading.
The existential basic meaning does not entail that a given kid laughed, (6a).
Simple disjunction, whose alternatives include the conjunction, is strengthened the other way, the conjunction being denied.
Non-maximality by pruning #
The readings of the kids laughed are the maximal one and the pruned ones.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The reading a partition selects is the strongest reading it decides.
Equations
- BarLev2021.IsReading D x P Q q = (q ∈ BarLev2021.readings D x P ∧ Q.Decides q ∧ ∀ r ∈ BarLev2021.readings D x P, Q.Decides r → q ⊆ r)
Instances For
How many atoms of the plurality satisfy P at w.
Equations
- BarLev2021.count D x P w = {x ∈ x ∩ D | P x w}.card
Instances For
"At least k of the kids laughed".
Equations
- BarLev2021.atLeast D x P k = {w : W | k ≤ BarLev2021.count D x P w}
Instances For
Exhaustifying over the size-m alternatives yields "at least n + 1 - m of the kids
laughed".
The readings are the thresholds from one to the whole plurality.
The pruned readings are readings of the plural predication too, the threshold k reading the
predicate over the parts of x with at least k atoms in the domain.
Pruning only weakens, the maximal reading entailing every pruned one.
Relevance and gappiness #
Given the polar question whether at least k of the kids laughed, the kids laughed means
that at least k did, provided every number of laughers is realized at some world.
A world is gappy when it is true on the reading some partition selects and false on another's.
Equations
- BarLev2021.Gappy D x P w = ((∃ (Q : Setoid W) (q : Set W), BarLev2021.IsReading D x P Q q ∧ w ∈ q) ∧ ∃ (Q : Setoid W) (q : Set W), BarLev2021.IsReading D x P Q q ∧ w ∉ q)
Instances For
The worlds judged neither true nor false are those where some but not all of the kids laughed.
A plain positive definite plural is gappy exactly where [Kri16]'s trivalent plural predication is undefined, where the accounts agree.
At a gappy world the trivalent negation is still undefined, while the negative sentence with
negation over ∃-PL, exhaustified over any of its alternatives, is plainly false, where the
accounts part.
The model whose worlds record who laughed #
In the model whose worlds are the sets of atoms satisfying the predicate, a satisfies it at
the world w when a ∈ w.
Equations
- BarLev2021.holds a w = (a ∈ w)
Instances For
Every atom satisfies the predicate alone at some world and all do at some world.
Every number of laughers up to the size of the plurality is realized.
The paper's examples #
Equations
- BarLev2021.instDecidableEqKid x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- BarLev2021.instFintypeKid = { elems := { val := ↑BarLev2021.Kid.enumList, nodup := BarLev2021.Kid.enumList_nodup }, complete := BarLev2021.instFintypeKid._proof_1 }
Kelly and Jane, the kids of (41).
Equations
Instances For
Kelly laughed.
Instances For
Jane laughed.
Instances For
Kelly or Jane laughed, (41a), whose alternatives include the conjunction, is strengthened to "not both".
The ten books on the reading list.
Equations
- BarLev2021.Book = Fin 10
Instances For
In the context of (80), where reading any five of the ten books passes the test, Mary read the books means that she read at least five of them; the paper offers the prediction that a fixed partition removes the gap tentatively.
Križ and Chemla's unembedded displays #
An unembedded display of [KC15] shows nine symbols, some number of which have the target color, so its worlds are the sets of target-colored symbols.
On Križ and Chemla's unembedded displays the account makes the positive sentence gappy exactly at the GAP displays, and every experiment that tested them found a gap there (Table 2).
Under negation the account, with negation over ∃-PL, makes every GAP display plainly false
whatever is pruned, while Table 2 found a gap there in Experiments A1 and B1, the residue the
paper attributes to the LF with ∃-PL over negation.
The paper's examples report gappiness for positive sentences and reduced gappiness for the
negative (82b), the asymmetry neg_barePlural_eq_zero_and_notMem_exhIEII_compl idealizes.