Napoli & Nespor (1976): Negatives in Comparatives @cite{napoli-nespor-1976} #
Language 52(4), 811–838.
The Italian word non appears in some comparative clauses without truth-conditional effect: Maria è più intelligente di quanto non sia Carlo 'Maria is more intelligent than Carlo (is)'. The construction had previously been treated either as pleonastic (Antinucci & Puglielli 1971) or as evidence that all than-clauses are underlyingly negative (Seuren 1969). @cite{napoli-nespor-1976} reject both: non₂ is real negation, but its distribution is governed by a discourse condition, not by the syntax of comparison.
Core Claim #
Non₂ is licensed iff the speaker presupposes that their assertion
contradicts a prior belief (their own or attributed to the addressee),
the move is assertive, the matrix is unnegated, and the construction
does not require precise knowledge. The four conditions are formalized
once in Pragmatics.Bias.BiasLicensingProfile and consumed here; this
study file is the first historical attestation of the predicate, applied
to Italian comparatives.
What's Foundational vs. Historical #
@cite{napoli-nespor-1976} bundles two contributions:
- Foundational (preserved): the licensing predicate, the Dario/Paolo context paradigm, the morphosyntactic diagnostics for underlying negation (subjunctive co-occurrence, pur-NPI licensing, neanche- conjunction, [-specific] indefinites, complementizer alternation), and the cross-construction generalization to indirect questions and cross-linguistic parallels (French ne, German modal particles).
- Historical (not preserved): the Generative Semantics deep-structure apparatus — abstract higher S₂ dominating S₃ with non₂ base-generated in S₃ and optionally deleted by transformation. The covert-operator insight is preserved by modern verum / use-conditional / commitment- update accounts (@cite{romero-han-2004}, @cite{repp-2013}, @cite{gutzmann-2015}, @cite{farkas-bruce-2010}); the deletion-rule mechanism specifically is not.
Connection to Existing linglib Studies #
- @cite{rett-2026} classifies Italian comparative EN as low/optional with
weak-NPI rejection. The N&N data refines this: optionality is
contextually conditioned (a
BiasLicensingProfile, not aBool), and weak NPIs are licensed when the profile licenses (modulo independent precision-blockers like affatto). SeeRett2026.italianComparativeand the patches in §6 below. - @cite{tsiakmakis-2025} classifies Italian comparative non as NEG₁ (apparent EN, standard negation masked). N&N's analysis is the original NEG₁-style account: non₂ is real negation in underlying structure, masked at surface by deletion or by an abstract operator.
- @cite{greco-2020} lists comparativeClauses among Italian weak-EN environments. N&N's pur-licensing data motivates this classification.
Four contexts varying the speaker's epistemic state #
@cite{napoli-nespor-1976} pp. 812–813 construct four dialogues showing that non₂'s acceptability tracks the speaker's presupposition that their assertion contradicts a prior belief — neither too uncertain (no presupposition to contradict) nor too explicit (use simple negation instead).
Context (6): Dario gives no opinion of Maria or Carlo; Paolo asserts Maria > Carlo. No prior belief to contradict ⇒ non₂ infelicitous.
Equations
Instances For
Context (7): Dario implies Carlo > Maria at chess; Paolo asserts Maria more intelligent. Inferred contradiction ⇒ non₂ may appear.
Instances For
Context (8): Dario explicitly says Maria is dumb; Paolo disagrees. The contradiction is too explicit; speaker should use simple negation, not non₂. Modeled here as failing the imprecise/inferred condition.
Instances For
Context (9): Dario's complaint implies he expects Maria can't help; Paolo asserts Maria smart enough to ask. Inferred contradiction ⇒ non₂ used.
Instances For
Predictions: non₂ infelicitous in (6), licensed in (7), blocked
in (8), licensed in (9). Each follows by decide from the licensing
predicate in Pragmatics.Bias.
Five distributional environments, each isolating one axis #
@cite{napoli-nespor-1976} demonstrates each licensing condition by exhibiting an environment where exactly that condition fails.
§2.1 ex. 10b: È più intelligente di quanto non sia Carlo? 'Is she more intelligent than Carlo is?' — questioned comparative blocks non₂. Speaker cannot simultaneously question and contradict.
Equations
Instances For
§2.2 ex. 18a: Maria non è più intelligente di quanto non sia Carlo 'Maria is not more intelligent than Carlo is' — matrix-negated blocks non₂.
Equations
Instances For
§2.3 ex. 22b: *Maria è tanto intelligente quanto non sia Carlo 'Maria is as intelligent as Carlo is' — equality (tanto…quanto) requires precise knowledge incompatible with the inferred-belief presupposition.
Equations
Instances For
§3.22 ex. 32b, 33b: *Maria è {molto più / due metri più} intelligente di quanto tu non creda — explicit precision modifiers fail the imprecise condition.
Equations
Instances For
§2.3 ex. 26b: Maria è meno intelligente di quanto tu non creda 'Maria is less intelligent than you think' — meno (less) allows non₂. This is N&N's sanity check against Seuren 1969: if non₂ were just "negation triggered by than-clause", it should appear equally with meno; in fact it does, but only under the same contextual conditions as with più.
Equations
Instances For
Predictions derived from the licensing profile #
@cite{napoli-nespor-1976} §3 marshals six independent diagnostics for
the presence of underlying negation in the comparative clause. Each
diagnostic targets a different surface correlate (mood morphology, NPI
surface form, indefinite specificity, conjunction admissibility,
complementizer choice, predicative clitic) and so returns a value of a
different type — flattening them all to Bool := p.licenses would
collapse six empirically distinct probes into one fact. The agreement
theorem licensed_diagnostics_agree collects the six predictions on the
licensed profile rather than asserting they are tautologically equal.
Specificity restriction on embedded indefinites. Without non₂ an indefinite in the than-clause is [±specific]; with non₂ it is restricted to [−specific] under the scope of the underlying negation. @cite{napoli-nespor-1976} §3.11 ex. 44–45.
- unrestricted : SpecificityProfile
- nonspecificOnly : SpecificityProfile
Instances For
Equations
- Phenomena.Negation.Studies.NapoliNespor1976.instDecidableEqSpecificityProfile 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
Complementizer choice in Italian comparatives. Di quanto is the standard than-complementizer; che surfaces when the abstract higher S of @cite{napoli-nespor-1976} §3.24 is present, which co-varies with non₂.
- che : ComplementizerChoice
- diQuanto : ComplementizerChoice
Instances For
Equations
- Phenomena.Negation.Studies.NapoliNespor1976.instDecidableEqComplementizerChoice 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
Predicate-clitic lo presence. §3.25 treats the clitic as genuinely
binary (present iff the abstract higher S licenses non₂), but the
two outcomes carry distinct surface morphology — Bool would lose
that contrast at the type level.
- present : CliticPresence
- absent : CliticPresence
Instances For
Equations
- Phenomena.Negation.Studies.NapoliNespor1976.instDecidableEqCliticPresence 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
§3.21: grammatical mood of the than-clause. Subjunctive surfaces
iff non₂ is underlyingly present (modulo lexical mood control by
credere etc., abstracted away here). Returning Core.Mood.GramMood
rather than Bool connects this prediction to the mood semantics in
Theories/Semantics/Mood/.
Equations
Instances For
§3.11 ex. 46–48: under bias-licensed negation the comparative admits
a weak NPI on its surface; the canonical witness in the Italian
Fragment is pur. Returning Option String carries the predicted
surface form (or none when blocked) rather than collapsing to a
coarse Bool. The form string is derived from the Fragment entry to
keep the connection true by construction.
Equations
- Phenomena.Negation.Studies.NapoliNespor1976.predictsWeakNPISurface p = if p.licenses then some Fragments.Italian.PolarityItems.pur.form else none
Instances For
§3.11 ex. 44–45: specificity restriction on embedded indefinites. Bias-licensed comparatives restrict the indefinite to [−specific]; blocked profiles leave it unrestricted.
Equations
- One or more equations did not get rendered due to their size.
Instances For
§3.11 ex. 50–52: neanche-conjunction is grammatical iff the first
conjunct is underlyingly negated. The Fragment registers neanche
as requiring .negation among its licensing contexts — the prediction
is derived from that registry fact in conjunction with the bias
profile, so changing the neanche lexical entry would break this
prediction by construction rather than requiring a separate update
here.
Equations
- One or more equations did not get rendered due to their size.
Instances For
§3.24: the complementizer che (vs. di quanto) is licensed iff
the abstract higher S is present, which co-varies with non₂.
Returning ComplementizerChoice exposes both alternants rather than
encoding "che present" as a Bool.
Equations
- One or more equations did not get rendered due to their size.
Instances For
§3.25: clitic lo substituting for the predicate adjective is
licensed iff the comparative has the abstract S structure (non₂
present). Returns CliticPresence rather than Bool to keep the
surface alternants distinguishable at the type level.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The six diagnostics agree on the licensed profile: each surface
correlate takes its bias-marked value. The conjunction below is the
theoretical unification @cite{napoli-nespor-1976} §3 argues for —
one licensing predicate explains all six surface diagnostics. The
neanche-conjunction conjunct is non-trivial: it factors through
the Italian Fragment's neanche.licensingContexts and needs decide
on the registry data.
Dual: on a blocked profile every diagnostic returns its non-bias-marked
value. Together with licensed_diagnostics_agree this witnesses that
the six diagnostics are genuinely predictive (their values vary
with the bias profile) rather than constant functions.
Why some weak NPIs are licensed and others blocked #
@cite{napoli-nespor-1976} §3.11 ex. 46–48 show that pur (weak NPI) is licensed in non₂-comparatives. §3.22 fn (i) shows that affatto (also a weak NPI) is blocked. The contrast is not about NPI licensing per se: affatto requires precise knowledge of the listener's belief, which fails N&N's imprecise condition. The block is semantic, not licensing- theoretic.
The weakNPIs field of Rett2026.ENDatum (refactored from Bool to a
WeakNPILicensing sum type for exactly this reason) classifies
italianComparative as .selective [pur.form]: it licenses some weak
NPIs (pur) and blocks others (affatto) for orthogonal reasons.
The two NPIs are registered in the Italian Fragment as
Fragments.Italian.PolarityItems.pur and
Fragments.Italian.PolarityItems.affatto. The contrast is already
witnessed at the lexical layer (the structural theorems
pur_licensed_in_comparative and affatto_not_licensed_in_comparative
in the Fragment), so this study file just consumes those facts and
shows them follow from the bias profile.
A weak NPI from the Italian Fragment is licensed in a bias-conditioned
negation environment iff (a) its registry lists .comparativeS (the
clausal-comparative slot — N&N's non₂ originates in subjunctive
clausal complements, not surface NP-comparatives) among its licensing
contexts AND (b) the bias profile licenses. The first conjunct rules
out affatto (registry-blocked); the second rules out unlicensed
contexts.
.comparativeS rather than .comparativeNP because surface NP-
comparatives are not NPI environments (@cite{hoeksema-1983} §3.6).
Equations
- Phenomena.Negation.Studies.NapoliNespor1976.predictsWeakNPI p npi = (p.licenses ∧ npi.licensingContexts.contains Features.LicensingContext.comparativeS = true)
Instances For
Pur is licensed in any context where non₂ is licensed.
Combines pur_licensed_in_comparative from the Italian Fragment with
the bias profile's licensing predicate.
Affatto is never licensed in non₂-comparatives — the block is
registered in the lexical entry itself (affatto.licensingContexts
excludes .comparativeS per @cite{napoli-nespor-1976} §3.22 fn 6). The
bias profile's licensing status is irrelevant: the Fragment's
distributional fact alone settles the case.
This makes the N&N contrast structural — derived from the lexical
licensingContexts field — rather than stipulated by an inline
inductive in this study file.
Non₂ outside comparatives: indirect questions #
@cite{napoli-nespor-1976} §4 ex. 88–91 show that the same non₂ appears in indirect questions where the speaker presupposes the negated proposition is contrary to expectation: Chissà se non vale la pena di comprarlo 'Who knows if it's not worth buying it' (suggests the speaker expects it to be worth buying).
The licensing profile is the same. This is the foundational insight that makes N&N's contribution a general predicate rather than a construction-specific rule.
Chissà se non… — indirect question with biased negation. Same licensing profile as the basic comparative.
Instances For
The licensing profile is invariant under construction type: the same profile that licenses comparative non₂ also licenses indirect- question non₂. This is why N&N's analysis generalizes; it is also the prediction we want for parallel phenomena (French ne in biased contexts, etc.).
Refining Rett2026.italianComparative.isOptional #
Rett2026.italianComparative.isOptional := true is technically correct
but coarse: optionality is contextually conditioned on a bias profile,
not free. The function below derives isOptional from a bias profile,
restoring the connection that the Bool field flattens.
Derived optionality: a construction is "optional" iff its licensing
profile is satisfied by some context but not all. For the
BiasLicensingProfile paradigm this is automatically true (the
licensed/blocked profiles witness both polarities).
Bridge to @cite{rett-2026}: the italianComparative datum's
isOptional field reflects exactly this contextual conditioning.