Partition as Inquiry — Setoid → Question embedding #
@cite{ciardelli-groenendijk-roelofsen-2018} @cite{theiler-etal-2018} @cite{puncochar-2019}
The faithful embedding of partition-based inquiry (Setoid W — what
POSWQ.inquiry carries) into the more general inquisitive-content
representation (Question W — downward-closed nonempty
families of information states).
Architectural placement #
This file implements the embedding direction prescribed in the
"Architectural note: Setoid vs. Question" section of
POSWQ.lean (added 0.229.922): we keep inquiry : Setoid W as the
field of POSWQ (partitions are the right shape for the propositional-
discourse use cases that currently exist), and provide a one-way
embedding Setoid → Question. The embedding goes this way
and not the other because every Setoid-based partition can be expressed
as an Question (each equivalence class becomes a maximal
proposition / alternative), but the reverse fails — mention-some,
intermediate-exhaustive, and conditional question alternatives are
non-disjoint or non-exhaustive and so are not representable as the
classes of any equivalence relation (see @cite{theiler-etal-2018}).
This mirrors mathlib's Set.toFinset / Filter.principal /
UniformSpace.toTopologicalSpace pattern: the "less general" structure
sits inside the "more general" one via a faithful but not surjective
forgetful map.
What this gives us #
fromSetoid r : Question W— the inquisitive content whose alternatives are the equivalence classes ofr. Concretely,props = {q | q = ∅ ∨ ∃ c ∈ r.classes, q ⊆ c}: a non-empty information stateqresolves the issue iff it is contained in some equivalence class (i.e., everything inqisr-equivalent).info_fromSetoid— informative content isSet.univ. Setoid-based inquiry is non-informative: it raises an issue but supplies no information. (This matches the standard partition-semantics view.)isInquisitive_fromSetoid_of_two_classes— ifrhas at least two distinct equivalence classes, the resulting content is inquisitive (inQuestion's sense:info ∉ props). The trivial partition (one class) yields a declarative.
The Question associated with a setoid r on W: a
proposition q resolves the issue iff q = ∅ or q is contained
in some r-equivalence class (i.e., everything in q agrees on the
r-question). The maximal such propositions are the equivalence
classes themselves.
Equations
- Core.Question.fromSetoid r = { props := {q : Set W | q = ∅ ∨ ∃ c ∈ r.classes, q ⊆ c}, contains_empty := ⋯, downward_closed := ⋯ }
Instances For
The informative content of a partition-based inquiry is the whole universe: every world is in some equivalence class, so every world appears in some resolving proposition. Setoid-derived inquisitive contents are non-informative — they raise issues but provide no information.
If a setoid has two distinct equivalence classes, the resulting
inquisitive content is inquisitive: Set.univ is not contained in
any single equivalence class (it would have to coincide with both),
so info ∉ props.
Setoid–Question bridge: refinement = entailment #
The fromSetoid embedding is monotone in mathlib's lattice order on
Setoid (where r₁ ≤ r₂ iff r₁ is finer: r₁.Rel x y → r₂.Rel x y)
and the entailment order on Question (P ≤ Q iff P.props ⊆ Q.props).
The bridge is moreover a lattice embedding: ≤ is reflected back.
This is the Set-based form of Groenendijk–Stokhof's "partition refinement
= question entailment", formulated as a one-liner over Setoid and
Core.Question with no worlds : List W quantifier and no isPartition
precondition.
The Setoid → Question embedding reflects the order: a setoid is
finer than another iff its inquisitive content is entailed by the
other's. The forward direction is monotonicity; the backward
direction uses the fact that r-equivalent pairs are resolved by
the two-element state {x, y}, which any cell-membership theorem
can recover.
Monotonicity of the fromSetoid embedding (the easy direction of
fromSetoid_le_iff).
Equivalence classes are alternatives #
Conversely, every alternative of fromSetoid r is either the empty
state or an equivalence class of r. The empty case can only occur
when W is empty (otherwise some class is in props and strictly
contains ∅, displacing it from maximality).
Each equivalence class of r is a maximal element (alternative) of
fromSetoid r. The proof uses (i) classes are nonempty (so they
aren't displaced by ∅), and (ii) classes are pairwise disjoint or
equal — so a class can only be properly contained in itself.
POSWQ bridge #
Lift the partition-based inquiry component of a POSWQ to its full
Question. This makes every existing POSWQ-using study
automatically a consumer of the inquisitive-content API: info,
alt, isInquisitive, the lattice operations, and the
mention-some/IE-question forcing arguments all become available
without rewriting the underlying state.
The inquisitive content embedded in a POSWQ via its inquiry
partition. Always non-informative (info = univ); inquisitive
iff the partition has more than one cell.