@cite{von-fintel-1999} — Strawson entailment as a rescue for Fauconnier-Ladusaw #
von Fintel, K. (1999). NPI Licensing, Strawson Entailment, and Context Dependency. Journal of Semantics 16(2), 97–148.
The paper defends the Fauconnier-Ladusaw analysis of NPI licensing — that NPIs
are licensed in DE positions — against four "recalcitrant" arenas where NPIs
are licensed despite the host context not being classically DE: only, the
adversative attitude predicates (sorry, surprised, regret), superlatives,
and conditional antecedents. The key move is to weaken classical DE to
Strawson-DE (Definition 14): an inference f(q) ⊨ f(p) need only hold
under the additional assumption that the conclusion's presupposition is
satisfied. With this weakening, all four contexts come out DE and the
Fauconnier-Ladusaw schema goes through.
What this study file does #
It is not a re-derivation of Strawson-DE — that lives in
Theories/Semantics/Entailment/StrawsonEntailment.lean, which already
proves onlyFull_isStrawsonDE, sorryFull_isStrawsonDE, gladFull_isUE,
superlative_isStrawsonDE, conditional_antecedent_strawsonDE, and the
relevant non-DE lemmas. This file is the paper-citation index: each
theorem is named after the paper's example number(s) and discharged by
specializing the corresponding theorem from StrawsonEntailment.lean to
the example's lexical instance. NPI grammaticality judgments themselves
(starred or unstarred sentences) are not theorems — they are the empirical
motivation, recorded in section docstrings with the sentences quoted from
the paper.
Coverage #
- §2 only: ex. 10, 11, 18 — formalized.
- §2.2 since (Iatridou): exs. 20-22 — discussed; no
sinceFulloperator in the formal substrate yet. - §2.3 pseudo-anti-additivity (against Atlas): exs. 23-27 — discussed.
- §3 sorry/surprised/regret: exs. 28, 29, 30 — formalized via
sorryFull. - §3.1 ex. 31 (the prima-facie incoherence anchor for Kadmon-Landman) — discussed in the §3 section docstring.
- §3.3 glad asymmetry: ex. 52 — formalized via
gladFull_isUE. - §3.4 shifting contexts (exs. 60-65): discussed; requires a dynamic-context treatment not yet in the substrate.
- §3.4 Curveball #2 focus-
only(exs. 66-68): discussed (subsumed byonlyFull_isStrawsonDE). - §4.1 conditional antecedents: exs. 70-74 — formalized for the restrictor analysis (under which conditionals are DE); the Stalnaker-Lewis Strengthening-the-Antecedent failures are discussed.
- §4.2 superlatives: exs. 75, 76, 77 — partially formalized via
superlative_isStrawsonDE.
§2 — only #
The Fauconnier-Ladusaw puzzle: only John licenses NPIs in its scope
(ex. 10) yet the canonical DE inference fails (ex. 11) — extending the
inference to a narrower scope (kale ⊆ vegetables) is not classically
truth-preserving because the conclusion Only John ate kale carries a
presupposition (someone ate kale) that the premise does not guarantee.
Strawson-DE plugs the gap (ex. 18): with the conclusion's presupposition
added as an extra premise, the DE inference goes through, and Strawson-DE
is sufficient for NPI licensing.
(10) Only John ever ate any kale for breakfast. (p. 101) (11) Only John ate vegetables for breakfast. ⇏ Only John ate kale for breakfast. (p. 101) (18) Kale is a vegetable. John ate kale for breakfast. Only John ate vegetables for breakfast. ∴ Only John ate kale for breakfast. (p. 104)
Ex. 11 (p. 101): only is not classically downward entailing.
Witness: kale ⊆ vegetables but the inference fails because the
conclusion's existence presupposition (someone ate kale) is not
guaranteed by the premise.
Ex. 18 (p. 104): only is Strawson-DE. The definedness predicate
encodes the existence presupposition: there is some w' such that
the focused individual John (here · == .w0) holds at w' and the
scope predicate holds at w'.
Ex. 11 + 18: the central separation. only is Strawson-DE without
being classically DE — and Strawson-DE is enough to license NPIs
under von Fintel's revised Fauconnier-Ladusaw schema. This is the
paper's headline result for §2.
§2.2 since (Iatridou, p.c.) #
Von Fintel relays an example from Sabine Iatridou:
(20) It's been five years since I saw a bird of prey in this area. ⇏ It's been five years since I saw an eagle in this area. (p. 107) (21) It's been five years since I saw any bird of prey in this area. ✓ (22) (with the additional premise "Five years ago I saw an eagle") the inference of (20) is restored.
Same dialectical shape as only: since licenses NPIs but is not
classically DE; adding the temporal presupposition (the eagle-sighting)
makes the inference go through.
Ex. 22 (p. 107): since is Strawson-DE in its complement. The
definedness predicate is the temporal presupposition (a past
p-event existed).
§2.3 — pseudo-anti-additivity is useless for NPI licensing #
@cite{atlas-1996} suggests that only John is "pseudo-anti-additive"
(ex. 25, p. 109): it satisfies the half of anti-additivity in which
f(x) ∧ f(y) → f(x ∨ y). Von Fintel shows this is "useless for the
analysis of NPI licensing" (p. 110): pseudo-anti-additivity is too weak —
it is shared by many quantifiers that license NPIs (no student) and
many that do not (some student, every student, at least three students); see exs. 26 and 27. The negative argument doesn't admit a
single-theorem formalization — it is a survey of counterexamples — but
the upshot for the formal substrate is exactly what we already have:
Strawson-DE, not pseudo-AA, is the operative notion.
§3 — adversative attitude predicates #
Adversative factive verbs (sorry, regret, surprised, amazed)
license NPIs in their complement clauses despite the complement position
not being classically DE.
(28a) Sandy is amazed/surprised that Robin ever ate kale. (p. 111) (28b) Sandy is sorry/regrets that Robin bought any car. (p. 111) (29) Robin ate kale ⇒ Robin ate a green vegetable; but Sandy is amazed that Robin ate a green vegetable ⇏ Sandy is amazed that Robin ate kale. (p. 111) (30) Robin bought a Honda Civic ⇒ Robin bought a car; but Sandy is sorry that Robin bought a car ⇏ Sandy is sorry that Robin bought a Honda Civic. (p. 111)
The factivity presupposition (the complement holds at the evaluation world) blocks classical DE: the conclusion's narrower complement may not hold even when the premise does. Strawson-DE rescues the inference by adding factivity at the world of evaluation.
Ex. 31 — Kadmon-Landman's prima-facie coherence challenge #
(31) Sandy regrets that Robin bought a car, but Sandy does not regret that Robin bought a Honda Civic. (p. 112)
If regret were uniformly DE, (31) should be incoherent.
@cite{kadmon-landman-1993} defend monotonicity by appealing to a change
of perspective between the conjuncts; von Fintel's §3.1 reanalysis
treats this as a shift of the modal-base parameter rather than a failure
of the underlying operator's monotonicity. The Strawson-DE result for
sorry below holds on a constant perspective.
Ex. 30 (p. 111): sorry is not classically DE in its complement.
The factivity component is what blocks DE: doxastic factivity of
the narrower complement p ⊆ q may fail at the evaluation world
even when q's does.
Ex. 28b (p. 111) — the explanatory result: sorry is Strawson-DE.
The definedness predicate is doxastic factivity (dox w ⊆ p):
the agent at the evaluation world w believes p. Given doxastic
factivity of the conclusion's complement and p ⊆ q, the
inference sorry q ⊨ sorry p goes through.
This explains why "Sandy is sorry that Robin bought any car"
licenses any despite the complement position not being classically DE.
Ex. 30 + ex. 28b — the adversative analogue of the only-separation:
sorry is Strawson-DE without being classically DE.
§3.2 — want and the Asher/Heim non-monotonicity puzzle #
vF §3.2 (pp. 115-121) defends want as upward entailing under a
doxastic modal base (DOX*), in response to the Asher 1987 Concorde
example (eq. 46) and the Heim 1992 couch example (eq. 48). The
"non-monotonicity" of want collapses to a context shift in the
modal base, parallel to the §3.4 shifting-context analysis for
adversatives. Headline: want is monotonic relative to a constant
context.
want is upward entailing in its complement (vF §3.2 headline; eq. 45).
§3 footnote 8 — Asher's WDE as a sibling notion #
vF p. 112 (footnote 8) cites @cite{asher-1987}'s Weakened Downward
Entailment as a related but formally distinct notion: it has a
doxastic side condition (belief in the conclusion's complement) and
operates in the upward direction, in contrast to Strawson-DE's
presuppositional side condition in the downward direction. The
substrate's IsWDE predicate captures this; classical UE implies WDE
trivially (monotone_implies_WDE).
§3.3 — glad is upward entailing, hence does not license NPIs #
(52)
gladis UE: fromα is glad that pandp ⇒ q, inferα is glad that q(paraphrasing the analysis on p. 124).
The asymmetry between sorry (DE in the complement under Strawson-DE)
and glad (UE) is what predicts the asymmetry in NPI licensing:
"*Sandy is glad that Robin bought any car" is ungrammatical; the same
sentence with sorry is fine.
The same adversative/non-adversative asymmetry shows up in Hindi
(@cite{lahiri-1998} §4.5): aaScarya 'surprised' licenses koii bhii /
ek bhii; khuS 'glad' does not. See Phenomena/Polarity/Studies/Lahiri1998.lean
for the Hindi data (npi_adversative_surprise_ek,
npi_adversative_surprise_koii, npi_glad_bad). The two papers offer
different explanations — Lahiri posits a covert anti-additive operator
over the complement; von Fintel derives the asymmetry from the lexical
monotonicity of the attitude — but they make the same predictions on
the basic English/Hindi data.
Ex. 50 / K&L (p. 122): glad (K&L eq. 50 semantics) is upward
entailing. Predicts NPIs are not licensed in the complement of glad.
Ex. 52 (p. 124) — vF's preferred replacement: glad (vF eq. 52
semantics) is also upward entailing. Same NPI-licensing prediction,
different content (cf. vF's Honda Civic example, p. 124-125).
The §3 headline: sorry and glad agree on factivity but differ on
monotonicity in the complement, and this monotonicity asymmetry
directly tracks the NPI-licensing asymmetry. Holds for both the
K&L and the vF analyses of glad.
§3.4 — shifting contexts #
The coherent sequences
(60) Sandy is glad that Robin bought a car, but Sandy is sorry/not glad that Robin bought a Honda. (p. 129) (61) Sandy is sorry that Robin bought a car, but Sandy is glad/not sorry that Robin bought a Honda. (p. 129)
Von Fintel argues these do not threaten the monotonicity analysis: their
coherence depends on a shift in the modal-base parameter between the
conjuncts (an "implicit conditionalization"; see ex. 61 discussion).
Validity of monotonic inferences is checked against a constant context.
A formal treatment requires dynamic-context machinery not yet present in
StrawsonEntailment.lean.
§3.4 — Curveball #2: focus-sensitive only over a non-name (p. 133) #
(66) There only was any precipitation in [MEDFORD]_F. (67) (66) plus "There was rain in Medford" ⊢_S There only was rain in [MEDFORD]_F. (p. 133)
Focus-sensitive only over a place name (or any non-proper-name
associate) is also Strawson-DE in its prejacent. Von Fintel notes
(eq. 68 (a/b), p. 134) that this requires the option-(a) semantics for
propositional only (weakening the asserted claim by closure under
entailment of the prejacent), not the option-(b) semantics adopted in
@cite{von-fintel-1997}. The substrate's onlyFull already captures the
option-(a) reading via its assertion clause "no y ≠ x satisfies the
scope".
§4 — conditional antecedents and superlatives #
§4.1 — conditional antecedents #
(70a) If John subscribes to any newspaper, he is probably well informed. (p. 135) (70b) If he has ever told a lie, he must go to confession. (p. 135) (70c) If you had left any later, you would have missed the plane. (p. 135)
Conditional antecedents license NPIs (ex. 70). Whether they are DE depends on the conditional analysis adopted:
- Restrictor analysis (@cite{kratzer-1986}; eq. 72, p. 137):
the if-clause restricts the modal base of the consequent's modal
operator. With an idle ordering source, antecedent strengthening
only shrinks the domain, so the antecedent position is classically
DE. The substrate's
condNecessityformalizes precisely this idle-ordering case (the simpler subcase of eq. 72 withmax_gset to identity); the full Kratzer conditional with non-trivial ordering source is not monotone, which is exactly the §4 puzzle. - Stalnaker-Lewis non-monotonic analysis (@cite{stalnaker-1968}, @cite{lewis-1973}; ex. 73, p. 138): Strengthening the Antecedent fails, so the antecedent is not DE. Von Fintel §4.3 (p. 141) defends a dynamic monotonic semantics in @cite{von-fintel-2000} under which the apparent failures reduce to context shifts, parallel to §3.4.
We formalize the restrictor side. Stalnaker-Lewis non-monotonicity is
a property of a different operator (a similarity-based would) not yet
in the substrate.
Ex. 72 (p. 137), restrictor analysis with idle ordering source:
a Kratzer-style condNecessity is classically DE in its antecedent.
Domain restriction is monotone.
Note: this is the strict subcase of vF eq. 72 where max_g is
trivial. The full Kratzer/Stalnaker-Lewis conditional with a
non-trivial preference ordering is not monotone — see the
ex73_* theorems below for the counterexample built from the
real Kratzer apparatus in Conditionals/Restrictor.lean.
Restrictor-style conditional antecedents are a fortiori Strawson-DE
(since classical DE implies Strawson-DE via de_implies_strawsonDE).
§4.2 — superlatives #
(75) Emma is the tallest girl to ever win the dance contest. (p. 138) (76) Emma is the tallest girl in her class. ⇏ Emma is the tallest girl in her class to have learned the alphabet. (p. 139) (77) Emma has learned the alphabet. Emma is the tallest girl in her class. ∴ Emma is the tallest girl in her class to have learned the alphabet. (p. 139)
Adding a restriction to the comparison class can change the ranking, so ex. 76 is not classically DE. With the additional premise that Emma satisfies the new restriction (ex. 77's "Emma has learned the alphabet"), the inference is Strawson-valid.
The substrate's superlativeAssert and superlative_isStrawsonDE
encode this for the predicative use of the superlative (ex. 75 / 77).
The non-predicative case where the superlative restricts a definite
description (ex. 80 p. 140) does not have local Strawson-DE — this is
documented in the substrate's superlative section.
Ex. 77 (p. 139): the superlative is Strawson-DE in the restriction position. The definedness predicate encodes the presupposition that the designated subject α (Emma) satisfies the restriction (has learned the alphabet).
Hierarchy connection #
The paper's §1 establishes the standard DE / AA / AM hierarchy and von
Fintel's §2 (Strawson move) extends it with Strawson-DE as the weakest
licensing level. The substrate proves AM → AA → DE → Strawson-DE
(de_implies_strawsonDE); this study file just records that the four
recalcitrant operators (onlyFull, sorryFull, superlativeAssert,
condNecessity) all land at exactly Strawson-DE, while gladFull
sits outside the hierarchy entirely (UE).
The four "recalcitrant" Strawson-DE operators of @cite{von-fintel-1999}, each strict (Strawson-DE without classical DE) where applicable:
Cross-framework bridges #
These theorems make explicit the relationships between vF's Strawson-DE analysis and three sibling NPI theories already formalized in linglib. Per CLAUDE.md "chronological dependency" rule: this file may reference @cite{kadmon-landman-1993} (1993 < 1999), @cite{lahiri-1998} (1998 < 1999), and @cite{hoeksema-1983} (1983 < 1999), but not the reverse.
K&L ≡ vF on the basic adversative asymmetry. Both
@cite{kadmon-landman-1993} (K&L) and @cite{von-fintel-1999} (vF)
derive the sorry/glad asymmetry from formally identical Lean
theorems — K&L via "lexical entailment to want ¬A", vF via
Strawson-DE / UE of the attitude operator. The two prose
explanations are different; the formalizations agree by rfl.
This is exactly the kind of theoretical-incompatibility-collapse
that linglib's interconnection-density discipline aims to surface.
The empirical predictions are the same; choosing between the two
analyses requires looking at examples that distinguish their
semantic predictions, e.g. K&L's "settle for less" reading
(bridge_lahiri_glad_settle_overgeneration below).
gladFull_isUE overgenerates against Lahiri's "settle for less"
Hindi data. @cite{lahiri-1998} §4.5 (datum
npi_glad_settle) records that Hindi khuS + NPI koii bhii IS
grammatical on a "settle for less" reading. K&L flag the same in
English (file KadmonLandman1993.settle_glad_anybody,
settle_glad_tickets). The substrate's gladFull_isUE predicts
uniformly NO licensing — so the substrate undergenerates here, and
the "settle for less" reading would require either (a) a different
gladFull semantics with a perspective shift, or (b) a Strawson-DE
treatment of the rescued reading.
This theorem records the empirical refutation as a Lean-checkable incompatibility between the substrate and the data.
Hoeksema's S-comparative is anti-additive, hence trivially Strawson-DE.
@cite{hoeksema-1983} proves the S-comparative anti-additive
(Semantics.Degree.Comparative.sComparative_isAntiAdditive); the
inheritance chain AA → DE → Strawson-DE makes the bridge automatic.
This places the S-comparative in the same Strawson-DE class as
@cite{von-fintel-1999}'s recalcitrants, but with the additional
classical AA backing — meaning S-comparatives license strong NPIs
too (whereas vF's only only licenses weak NPIs).