Alternative semantics: focused element evokes alternatives
- ordinary : α
The ordinary semantic value
- alternatives : List α
The focus alternatives
Instances For
Covert EVEN (@cite{crnic-2014}, building on @cite{lahiri-1998}) #
@cite{rooth-1992}
EVEN has two semantic contributions:
- Scalar presupposition: The focused element is least likely among alternatives
- Additive presupposition: At least one alternative is true (for "also" reading)
- Assertion: The prejacent is true
For NPI licensing, only the scalar presupposition matters.
Likelihood ordering over propositions (context-dependent).
likelihood a b holds when a is less likely (more surprising) than b.
Equations
- Semantics.FocusParticles.LikelihoodOrder World = ((World → Bool) → (World → Bool) → Prop)
Instances For
Traditional EVEN semantics
- prejacent : World → Bool
The prejacent proposition
- alternatives : List (World → Bool)
Focus alternatives
- likelihood : LikelihoodOrder World
Likelihood ordering
Instances For
EVEN asserts the prejacent
Instances For
EVEN presupposes prejacent is least likely.
This is @cite{karttunen-peters-1979}'s universal threshold: the prejacent
must be less likely than ALL alternatives. @cite{francescotti-1995} argues
this is too strong — see EvenThreshold.most for the revised condition.
Equations
- even.presupposition = ∀ (alt : World → Bool), alt ∈ even.alternatives → even.likelihood even.prejacent alt
Instances For
Full EVEN meaning: defined and true
Instances For
NPI Licensing Mechanism #
The key insight: In DE contexts, wide-domain NPIs make the prejacent LESS likely, satisfying EVEN's presupposition.
"John didn't see anyone" = EVEN [John didn't see anyone] = Presupposes: For all x, P(John didn't see x) ≥ P(John didn't see anyone) = "Not seeing anyone" is less likely than "not seeing some particular person" = Presupposition SATISFIED (negation creates DE context)
"*John saw anyone" = EVEN [John saw anyone] = Presupposes: For all x, P(John saw x) ≥ P(John saw anyone) = "Seeing anyone" is MORE likely than seeing some particular person = Presupposition VIOLATED
NPI licensing condition: EVEN presupposition must be satisfiable.
Uses ContextPolarity from Core.NaturalLogic.
Equations
- Semantics.FocusParticles.npiLicensed Core.NaturalLogic.ContextPolarity.downward npiDomain regularDomain _hWider = True
- Semantics.FocusParticles.npiLicensed Core.NaturalLogic.ContextPolarity.upward npiDomain regularDomain _hWider = False
- Semantics.FocusParticles.npiLicensed Core.NaturalLogic.ContextPolarity.nonMonotonic npiDomain regularDomain _hWider = False
Instances For
NPI licensed in DE contexts
NPI unlicensed in UE contexts
NPI unlicensed in non-monotonic contexts
Overt "only" #
"Only" is the overt counterpart of EXH:
- Presupposes: The prejacent is true
- Asserts: No stronger alternative is true
"Only John came" = Presupposes: John came = Asserts: No one other than John came
This is equivalent to EXH with the prejacent as a presupposition.
Traditional "only" semantics
- prejacent : World → Bool
The prejacent (the focused element's contribution)
- alternatives : List (World → Bool)
The alternatives (what focus evokes)
Instances For
"only" presupposes the prejacent
Equations
- only.presupposition = only.prejacent
Instances For
"only" asserts no alternative is true. The alternatives list excludes the prejacent (Roothian focus alternatives minus the focused element's contribution).
Equations
- only.assertion w = only.alternatives.all fun (alt : World → Bool) => !alt w
Instances For
Full "only" meaning
Instances For
A likelihood ordering is MONOTONE w.r.t. entailment when stronger propositions (true in fewer worlds) are less likely.
If p entails q (i.e., p is true only at worlds where q is true),
then lessLikely p q (p is at least as unlikely as q).
This is the bridge between Theories/Semantics/Entailment/ and
focus particle semantics — the connection that @cite{lahiri-1998}
relies on to derive NPI licensing from the cardinality scale.
Equations
- Semantics.FocusParticles.LikelihoodMonotone lessLikely = ∀ (p q : W → Bool), (∀ (w : W), p w = true → q w = true) → lessLikely p q
Instances For
Focus Particle Comparison #
| Particle | Presupposition | Assertion | Polarity Effect |
|---|---|---|---|
| EVEN | Least likely | Prejacent | Licenses NPIs (DE) |
| EXH | None | Prejacent ∧ ¬alternatives | Scalar implicatures (UE) |
| only | Prejacent | ¬alternatives | Explicit exhaustivity |
Key Observations #
EVEN and EXH are duals:
- EVEN: active in DE contexts (licenses NPIs)
- EXH: active in UE contexts (generates SIs)
Only is overt EXH:
- Same semantic effect as covert EXH
- But with prejacent as presupposition, not assertion
Threshold variants for the EVEN scalar presupposition. The theoretical dispute concerns how many alternatives the prejacent must exceed in unlikelihood:
- @cite{bennett-1982}: at least one (too weak)
- @cite{karttunen-peters-1979}: all (too strong)
- @cite{francescotti-1995}: most (correct)
- existential : EvenThreshold
S* more surprising than at least one neighbor
- universal : EvenThreshold
S* more surprising than all neighbors
- most : EvenThreshold
S* more surprising than most neighbors
Instances For
Equations
- Semantics.FocusParticles.instDecidableEqEvenThreshold x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Count of alternatives that the prejacent exceeds under a decidable ordering.
Equations
- Semantics.FocusParticles.countExceeded prejacent alternatives moreSurprising = (List.filter (moreSurprising prejacent) alternatives).length
Instances For
Generalized EVEN presupposition parameterized by threshold.
moreSurprising a b returns true when a is more surprising
(less likely) than b.
Equations
- One or more equations did not get rendered due to their size.