Economy of Structure and Information #
@cite{heim-1991} @cite{hurford-1974} @cite{katzir-2007} @cite{katzir-singh-2015} @cite{magri-2009} @cite{magri-2011} @cite{spector-2014}
@cite{katzir-singh-2015}. Proceedings of Sinn und Bedeutung 19, pp. 322–339.
Two felicity conditions on assertions:
Question Condition (def 8): An assertion must address a good question — one not trivially settled by the common ground.
Answer Condition (def 15): An assertion must not be needlessly inferior to any alternative — where inferiority combines structural complexity with semantic strength.
These two conditions unify:
- @cite{magri-2009} oddness (# Some Italians come from a warm country)
- @cite{spector-2014} oddness (# All Italians...; # John has one wife)
- Hurford's constraint (# John visited France or Paris)
- Maximize Presupposition! (# A sun is shining)
- DE reversal of oddness patterns
Open problem: oddness under embedding (K&S §4) — the conditions are stated globally but oddness persists in embedded constituents.
Discourse scenario packaging meaning, complexity, context, and QUD.
meaning: interpretation of each utterance at each worldcomplexity: structural complexity; lower = simplercontext: common knowledge;context w = trueiff w is CK-compatiblequd: question under discussion (equivalence relation on worlds)utterances: speaker's available alternativesworlds: exhaustive world enumeration
- meaning : U → W → Bool
- complexity : U → ℕ
- context : W → Bool
- qud : QUD W
- utterances : List U
- worlds : List W
Instances For
Question Condition violation (K&S def 7–8): the QUD is trivially settled by CK — all context-compatible worlds give the same answer.
Equations
- s.badQuestion = s.cWorlds.all fun (w : W) => s.cWorlds.all fun (v : W) => s.qud.sameAnswer w v
Instances For
Semantic entailment over ALL worlds: ⟦u⟧ ⊆ ⟦v⟧. Uses all worlds (not just context), since the better-than relation compares general semantic strength (K&S def 16a).
Instances For
At-least-as-good-as (K&S def 16a): φ ≲ ψ iff φ is at most as complex as ψ AND semantically at least as strong (⟦φ⟧ ⊆ ⟦ψ⟧).
Equations
- s.atLeastAsGood u v = (decide (s.complexity u ≤ s.complexity v) && s.semEntails u v)
Instances For
Strictly better-than (K&S def 16b): φ ≺ ψ := φ ≲ ψ ∧ ¬(ψ ≲ φ).
Equations
- s.betterThan u v = (s.atLeastAsGood u v && !s.atLeastAsGood v u)
Instances For
Answer Condition violation (K&S def 15): u is needlessly inferior — there exists a strictly better alternative.
Equations
- s.needlesslyInferior u = s.utterances.any fun (v : U) => s.betterThan v u
Instances For
K&S Oddness: violates Question Condition or Answer Condition.
Equations
- s.isOdd u = (s.badQuestion || s.needlesslyInferior u)
Instances For
Contextual equivalence: same truth value at all CK-compatible worlds.
Instances For
Strengthened Answer Condition: also requires the dominating alternative
to be compatible with the context (true at some CK-world). This closes
a truth gap where needlesslyInferior could flag an utterance as odd
because a false-but-simpler alternative exists.
For all 5 K&S scenarios, this coincides with needlesslyInferior
(verified below).
Equations
- s.needlesslyInferiorStrict u = s.utterances.any fun (v : U) => s.betterThan v u && s.cWorlds.any (s.meaning v)
Instances For
Reusable semantic model: meaning, complexity, world enumeration, and
utterance alternatives. Factored out of Scenario so the same model
can be paired with different discourse contexts.
Instances For
Build a Scenario from composable pieces.
Equations
- Semantics.Questions.EconomyOddness.Scenario.mk' m d = { meaning := m.meaning, complexity := m.complexity, context := d.context, qud := d.qud, utterances := m.utterances, worlds := m.worlds }
Instances For
C-contradiction: incompatible with context. (@cite{spector-2014}, def 4a)
Equations
Instances For
C-equivalent to φ: same truth value in context. (@cite{spector-2014}, def 4c)
Equations
Instances For
No Trivial Alternatives violation (@cite{spector-2014}, def 5): ALL alternatives are trivial in C given φ.
Equations
- Semantics.Questions.EconomyOddness.allAlternativesTrivial worlds C φ alts = alts.all fun (φ' : W → Bool) => Semantics.Questions.EconomyOddness.isTrivialInC worlds C φ φ'
Instances For
K&S ex. (1)–(2): "# Some/All Italians come from a warm country"
CK: Italy is a warm country. Since all Italians come from Italy, the QUD "Do [some/all] Italians come from a warm country?" is trivially settled → Question Condition violation.
Both K&S and @cite{spector-2014} predict oddness here (§1.2).
- allWarm : ItalyWorld
- noneWarm : ItalyWorld
Instances For
Equations
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Instances For
Equations
- Semantics.Questions.EconomyOddness.instBEqItalyUtt.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)
Instances For
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.
Instances For
Contextual equivalence: some ↔ all (CK makes them interchangeable).
QUD trivially settled by CK → Question Condition violated.
K&S prediction: "some" is odd.
K&S prediction: "all" is odd.
Bridge: @cite{spector-2014} makes the same prediction (all alternatives trivial).
K&S ex. (14)/(17): "In this department, every professor gives the same grade to all of his students. Kim is a professor." (a) # This year, Kim assigned an A to some of his students — ODD (b) This year, Kim assigned an A to all of his students — OK
The QUD is good (we don't know Kim's grade). But "some" is needlessly weak: "all" is equally complex and semantically stronger.
Connects to Core.Scale: the ⟦all⟧ ⊆ ⟦some⟧ entailment that drives
the Answer Condition is the same relationship captured by
Core.Scale.Quantifiers.entails.all.some_ = true.
- allA : GradeWorld
- someNotAll : GradeWorld
- noneA : GradeWorld
Instances For
Equations
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Instances For
Equations
- Semantics.Questions.EconomyOddness.instBEqGradeUtt.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)
Instances For
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.
Instances For
QUD is NOT trivially settled (Kim might or might not have given A).
"all" entails "some" (semantically stronger in all worlds).
"some" does NOT entail "all" (true at someNotAll where "all" is false).
"all" ≺ "some" (equally complex, strictly stronger).
K&S prediction: "some" is odd (needlessly weak).
K&S prediction: "all" is fine (best available answer).
Contextual equivalence despite general non-equivalence.
K&S ex. (22): "# John visited France or Paris"
Since Paris ⊆ France, "France or Paris" ≡ "France" semantically. But the disjunction is structurally more complex (complexity 2 vs 1). So "France" ≺ "France or Paris" → disjunction is needlessly complex.
Connects to RSA/ScalarImplicatures/Hurford.lean which models Hurford's constraint via speaker rationality. K&S derive the same prediction from economy, without RSA machinery.
- parisOnly : VisitWorld
- franceNotParis : VisitWorld
- neither : VisitWorld
Instances For
Equations
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Instances For
Equations
- Semantics.Questions.EconomyOddness.instBEqVisitUtt.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Hurford alternatives: the disjunction vs. its simple equivalent. "Paris" is excluded from utterance alternatives because the Hurford comparison is between a complex form and its semantically equivalent simpler form, not between independently informative expressions.
Equations
- One or more equations did not get rendered due to their size.
Instances For
"France" and "France or Paris" are semantically equivalent (Paris ⊆ France, so the disjunction adds nothing).
"France" ≺ "France or Paris" (simpler + equally strong).
K&S prediction: "France or Paris" is odd (needlessly complex).
K&S prediction: "France" is fine (no better alternative available).
K&S ex. (18): "Every prof who assigned an A to [some/all] got a raise"
In DE restrictor, "some" picks out MORE professors → stronger universal. The entailment direction reverses: ⟦some⟧_DE ⊆ ⟦all⟧_DE. So "all" becomes needlessly weak (opposite of UE).
Connects to Semantics.Montague/Sentence/Entailment/Monotonicity.lean:
the reversal here is the same phenomenon as every_restr_DE.
Equations
- Semantics.Questions.EconomyOddness.instBEqDEWorld.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Instances For
Equations
- Semantics.Questions.EconomyOddness.instBEqDEUtt.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
DE scenario: "some" version is stronger (narrower truth set). In UE, ⟦all⟧ ⊆ ⟦some⟧; in DE, this reverses to ⟦some⟧_DE ⊆ ⟦all⟧_DE.
Equations
- One or more equations did not get rendered due to their size.
Instances For
DE reversal: "some" entails "all" (opposite of UE).
"all" does NOT entail "some" in DE.
DE: "some" ≺ "all" (reversed from UE where "all" ≺ "some").
K&S prediction in DE: "all" is odd (needlessly weak).
K&S prediction in DE: "some" is fine (the stronger answer).
Cross-scenario bridge: UE and DE have opposite entailment directions.
Cross-scenario bridge: UE and DE have opposite oddness verdicts.
K&S ex. (21): "# A sun is shining" vs. "The sun is shining"
CK: there is exactly one sun. Both utterances are contextually equivalent. But "the sun" presupposes uniqueness, making it semantically stronger (⟦the sun is shining⟧ ⊂ ⟦a sun is shining⟧). So "a sun" is needlessly weak → Maximize Presupposition falls out as Answer Condition.
Connects to Core/Presupposition.lean: the definite's stronger presupposition is what gives "the" its semantic advantage under K&S's ordering.
Equations
- Semantics.Questions.EconomyOddness.instBEqSunWorld.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Instances For
Equations
- Semantics.Questions.EconomyOddness.instBEqSunUtt.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)
Instances For
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.
Instances For
"the sun" entails "a sun" (presuppositional strength).
"a sun" does NOT entail "the sun" (true at manySuns where "the" fails).
"the sun" ≺ "a sun" (equally complex, strictly stronger).
K&S prediction: "a sun" is odd (= Maximize Presupposition!).
K&S prediction: "the sun" is fine.
Contextual equivalence: CK (unique sun) makes a/the equivalent.
The Question Condition and Answer Condition are independent:
- Magri: Question Condition violated, Answer Condition irrelevant
- Grade: Question Condition satisfied, Answer Condition violated This shows neither condition subsumes the other.
Verify that needlesslyInferiorStrict (context-aware) coincides with
needlesslyInferior on all 5 scenarios. This confirms the truth gap
doesn't affect the existing examples.
Demonstrate that SemanticModel + DiscourseContext can be composed
via Scenario.mk'. The DE scenario's semantic model is reused with a
different discourse context (all worlds CK-compatible).
DE scenario's semantic model, factored out for reuse.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Original DE discourse context (w2 ruled out by CK).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Alternative discourse context: all worlds CK-compatible.
Equations
- Semantics.Questions.EconomyOddness.deDiscourseOpen = { context := fun (x : Semantics.Questions.EconomyOddness.DEWorld) => true, qud := QUD.ofDecEq id }
Instances For
The composed scenario matches the original deScenario.
With open context (w2 now CK-compatible), "all" is still odd: "some" still entails "all" in the DE model, so the Answer Condition is unchanged. But the context change demonstrates reuse of the semantic model with a different discourse context.