Update Semantics #
@cite{veltman-1996}
In Update Semantics:
- Information states are sets of worlds (not assignments)
- Sentences update states by eliminating incompatible worlds
- "Might φ" is a TEST: passes if some φ-worlds remain
- No discourse referents (simpler than DRT/DPL)
⟦φ⟧ : State → State where State = Set World
Update Semantics state: a set of possible worlds.
Unlike DPL/DRT, no assignment component - US focuses on propositional content.
Equations
- Semantics.Dynamic.UpdateSemantics.State W = Set W
Instances For
Update function: how a sentence modifies a state.
Equations
Instances For
Propositional update: eliminate worlds where φ fails.
⟦φ⟧(s) = { w ∈ s | φ(w) }
Equations
- Semantics.Dynamic.UpdateSemantics.Update.prop φ s = {w : W | w ∈ s ∧ φ w}
Instances For
Conjunction: sequential update.
⟦φ ∧ ψ⟧ = ⟦ψ⟧ ∘ ⟦φ⟧
Delegates to Core.CCP.seq.
Equations
- φ.conj ψ = Semantics.Dynamic.Core.CCP.seq φ ψ
Instances For
Negation: test and possibly fail.
⟦¬φ⟧(s) = s if ⟦φ⟧(s) = ∅, else ∅
Delegates to Core.CCP.neg.
Equations
Instances For
Epistemic "might": compatibility test.
⟦might φ⟧(s) = s if ⟦φ⟧(s) ≠ ∅, else ∅
Delegates to Core.CCP.might.
Equations
Instances For
Epistemic "must": universal test.
⟦must φ⟧(s) = s if ⟦φ⟧(s) = s, else ∅
Delegates to Core.CCP.must.
Equations
Instances For
Order matters for epistemic might.
"It's raining and it might not be raining" is contradictory: after learning rain, the might-not-rain test fails (no ¬rain worlds remain). But "it might not be raining and it's raining" can succeed: the might test passes on the initial state, then learning eliminates ¬rain worlds.
Requires Nontrivial W: for empty or singleton W, no state has both
p-worlds and ¬p-worlds, making the second conjunct unsatisfiable.
State s supports φ iff updating with φ doesn't change s.
s ⊨ φ iff ⟦φ⟧(s) = s
Equations
- Semantics.Dynamic.UpdateSemantics.supports s φ = (φ s = s)
Instances For
State s accepts φ iff updating with φ yields a non-empty state.
s accepts φ iff ⟦φ⟧(s) ≠ ∅
Equations
- Semantics.Dynamic.UpdateSemantics.accepts s φ = Set.Nonempty (φ s)
Instances For
Validity₁: updating the minimal state 0 with the premises in order yields a state that supports the conclusion.
ψ₁,...,ψₙ ⊩₁ φ iff 0[ψ₁]⋯[ψₙ] ⊨ φ
@cite{veltman-1996}, §1.2. This is the notion Veltman concentrates on: it captures the fact that default conclusions depend on exactly what information is available.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Validity₂: for every state σ, updating with the premises in order yields a state that supports the conclusion.
ψ₁,...,ψₙ ⊩₂ φ iff ∀σ, σ[ψ₁]⋯[ψₙ] ⊨ φ
Equations
- One or more equations did not get rendered due to their size.
Instances For
Validity₃: one cannot accept all premises without accepting the conclusion. Closest to the classical notion.
ψ₁,...,ψₙ ⊩₃ φ iff ∀σ, (σ ⊨ ψ₁ ∧ ... ∧ σ ⊨ ψₙ) → σ ⊨ φ
Equations
- One or more equations did not get rendered due to their size.
Instances For
Validity₂ implies validity₁: specializing σ = 0.
@cite{veltman-1996}, Proposition 1.3 (one direction, unconditional).
Validity₃ is monotonic: adding premises preserves validity.
@cite{veltman-1996}, §1.2: validity₃ is the only notion that is both left and right monotonic.
The designated undefined state: update failure.
@cite{yagi-2025} Definition 5: when a presupposition is not satisfied,
the update yields ∗. We model ∗ as none via Option (State W).
Equations
Instances For
Presuppositional update: update by φ_p is defined only when the
presupposition p is supported (i.e. s[p] = s).
@cite{yagi-2025} Definition 5: s[φ_p] = ∗ if s[p] ≠ s = s[φ] otherwise
@cite{heim-1982} @cite{beaver-2001} @cite{veltman-1996}
Equations
- Semantics.Dynamic.UpdateSemantics.PUpdate.presup p φ none = none
- Semantics.Dynamic.UpdateSemantics.PUpdate.presup p φ (some s) = if Semantics.Dynamic.UpdateSemantics.Update.prop p s = s then some (Semantics.Dynamic.UpdateSemantics.Update.prop φ s) else none
Instances For
Negation extended to PState: s[¬φ] = s/s[φ].
@cite{yagi-2025} Definition 4: s[¬φ] = s \ s[φ].
Equations
- Semantics.Dynamic.UpdateSemantics.PUpdate.neg φ none = none
- Semantics.Dynamic.UpdateSemantics.PUpdate.neg φ (some s) = some (s \ Semantics.Dynamic.UpdateSemantics.Update.prop φ s)
Instances For
Disjunction extended to PState: s[φ ∨ ψ] = s[φ] ∪ s[¬φ][ψ].
@cite{yagi-2025} Definition 4, @cite{heim-1982}. Extended with ∗ ∪ s = s ∪ ∗ = ∗.
Equations
- One or more equations did not get rendered due to their size.
- Semantics.Dynamic.UpdateSemantics.PUpdate.disj φ ψ none = none
Instances For
Presuppositional disjunction: s[φ_p ∨ ψ_q]. Apply presupposition checks to each disjunct.
This is the standard Heim/Beaver definition: s[φ_p ∨ ψ_q] = s[φ_p] ∪ s[¬φ_p][ψ_q]
Both presuppositional updates must be defined for the result to be defined: s[φ_p] requires s ⊨ p, and s[¬φ_p][ψ_q] requires s[¬φ_p] ⊨ q.
Equations
- One or more equations did not get rendered due to their size.
- Semantics.Dynamic.UpdateSemantics.PUpdate.disjPresup p φ q ψ none = none
Instances For
Flexible accommodation disjunction (dynamic version).
@cite{yagi-2025} (13) / @cite{geurts-2005} / @cite{aloni-2022}: s[φ ∨ ψ] = s[χ][φ] ∪ s[ω][ψ], where s[χ] ∪ s[ω] = s
The propositions χ and ω split the state s into two substates. By default χ = ω = ⊤ (both tautological), but when the default violates genuineness (@cite{zimmermann-2000}), the split becomes non-trivial: χ = ¬q and ω = ¬p for conflicting presuppositions.
Equations
- One or more equations did not get rendered due to their size.
- Semantics.Dynamic.UpdateSemantics.PUpdate.disjFlex χ φ_presup φ ω ψ_presup ψ _h_split none = none
Instances For
Presuppositional disjunction update is uninformative when both presuppositions are already supported: if s ⊨ p and s ⊨ q and the disjunction φ ∨ ψ is already true throughout s, the update returns s unchanged.
Note: this applies to non-conflicting presuppositions. When
p ∧ q = ⊥, the hypotheses hp and hq are jointly unsatisfiable
(unless s = ∅). For the conflicting case, see
update_yields_undefined in the Yagi2025 study, which shows the
update is undefined (∗) rather than uninformative.