Docs / Studies / Studies.Romoli2013b

A Scalar Implicature-Based Approach to Neg-Raising

Romoli, J. (2013). Linguistics and Philosophy, 36, 291–353

Romoli derives neg-raising as a scalar implicature. A neg-raising predicate such as think is a plain universal over its modal base, and its lexical alternative is the excluded middle, where a predicate without neg-raising such as be certain has its existential dual. Exhaustifying a negated neg-raising predicate against the negated excluded middle gives the predicate over the negated complement, the truth set that Gajewski's presupposition gives, while the same sentences are plainly false, not undefined, where the subject has no opinion. Since the excluded middle is entailed wherever the predicate is not under a downward entailing operator, it does not project from questions, conditionals or modals as a presupposition would, and in the restrictor of a universal exhaustification yields an inference of its own. With the enriched meanings as the direct implicatures that the licensing of strict NPIs reads, the library's licensing theory admits the strict NPI of each example where Romoli accepts it, the restrictors of quantifiers aside.

Implementation notes

  • Romoli's exhaustification denies every alternative that the prejacent does not entail. Which alternatives those are is fixed by the theory, not by the model: the families of alternatives are indexed in order of logical strength, and the exhaustified prejacent holds where its index is the greatest true one, IsGreatest. In an upward entailing environment the excluded middle is the weaker alternative, which Exhaustification.exh shows in any model.
  • Romoli adopts the proposal of Gajewski and of Chierchia that a strict NPI needs an environment that stays downward entailing once implicatures and presuppositions are added. The library's PolarityItem.Licenser.scalar pairs an operator with its meaning enriched by its direct implicature, and Licenser.Carries reads anti-additivity off the enriched meaning, so the two negated predicates are scalar licensers that share an operator and differ in the enrichment.

TODO

  • Suspension by relevance to the question under discussion, the negated universal, and the truth predicates it is true that, whose presupposition is to disrupt licensing in two of the examples, are not formalized.
  • Stacked predicates are exhaustified here without the condition that exhaustification leave the presupposition of its prejacent untouched, which is what blocks the lowered reading of want over believe. It needs want with von Fintel's presupposition and exhaustification of presupposing prejacents.
  • The restrictors of every and no are to fail to license because of their existential presupposition. The library's restrictor contexts carry none, and the example with in weeks awaits a lexical entry.

References

The basic case and negation (§5.1)

Where a neg-raising predicate stands in an upward entailing environment f, the excluded middle alternative is entailed and exhaustification adds nothing: unembedded (88), and under a possibility modal (96).

theorem exh_of_monotoneL79–87
{W : Type u_1} {V : Type u_2} {f : Set W → Set V} (hf : Monotone f) (R : SetRel W W) (p : Set W) :
Exhaustification.exh {f (R.core p), f (NegRaising.excludedMiddle R p)} (f (R.core p)) = f (R.core p)
L113–114
L138–142

Non-upward-entailing environments (§5.2.2)

In the restrictor of a universal, exhaustification adds that someone who believes the complement false is outside the scope ((100)–(102)), an inference that the presuppositional account does not make.

theorem isGreatest_everyWhoL151–162
{W : Type u_1} {p : Set W} {w : W} {E : Type u_2} {B : E → SetRel W W} {Q : E → Set W} :
IsGreatest {i : Fin (Nat.succ 0).succ | w ∈ ![everyWho (fun (x : E) => (B x).core p) Q, everyWho (fun (x : E) => NegRaising.excludedMiddle (B x) p) Q] i} 0 ↔ w ∈ everyWho (fun (x : E) => (B x).core p) Q ∧ ∃ (x : E), w ∈ (B x).core pᶜ ∧ w ∉ Q x

Built from

Negative quantifiers (§7.2)

L172–175
L178–189

Stacked predicates (§7.3.1)

Exhaustified against its three alternatives, a negated neg-raising predicate over another has the negation below both, whichever predicate is on top ((172)–(175)). The scalar account alone therefore predicts no asymmetry between believe over want and want over believe.

theorem isGreatest_stackL231–249
{W : Type u_1} {w : W} (R₁ R₂ : SetRel W W) (p : Set W) :
IsGreatest {i : Fin 2 × Fin 2 | w ∈ stackAlternatives R₁ R₂ p i} (0, 0) ↔ w ∈ R₁.dom ∩ R₁.core (R₂.dom ∩ R₂.core pᶜ)

Built from

Strict NPIs (§6)

Each environment is a licenser whose enriched meaning is the exhaustified one. Negated think and negated be certain share the operator of a negated universal and differ in the enrichment, by exhNot_eq and isGreatest_compl_core_preimage.

L257–268
L313–318

The paper's judgments

The licensing theory admits the strict NPI of each example in its environment exactly when the paper judges the example acceptable, the restrictor of a universal aside: negated think is anti-additive once enriched, and negated be certain is no longer downward entailing ((112), (123)–(127)).

theorem judgments_predictedL365–372

Built from , ,