Karttunen 1971: Implicative Verbs [Kar71] #
Implicative Verbs. Language 47(2): 340–358.
Core Contribution #
Complement-taking predicates that take infinitival complements divide into implicative and non-implicative classes based on complement entailment:
- Implicative (manage, remember, bother): "managed to VP" → VP; "didn't manage to VP" → ¬VP. Two-way entailment.
- Non-implicative (hope, want, intend): no complement entailment.
- Negative implicatives (fail, forget): entail ¬complement.
- Sufficient-only (force, cause): affirmative → complement, but negation ↛ ¬complement.
- Necessary-only (be able, possible): negation → ¬complement, but affirmative ↛ complement.
Historical Context #
Karttunen's 2×2 classification (necessary × sufficient) was the original
descriptive taxonomy. The modern consensus ([Nad23a]) derives
the entailment patterns from causal structure rather than from
presuppositional schemas. The theory layer (Causation/Implicative.lean)
implements the modern causal analysis; this study file preserves
Karttunen's original classification. The bridges to the modern types
live with the papers that draw them: Studies/Nadathur2023.lean
(classification conversion) and Studies/NadathurLauer2020.lean
(entailment cells vs causal mechanism).
Key differences from the modern analysis:
- Karttunen treats the entailment as arising from a presupposition that v(S) is a nec/suf condition for S. Nadathur derives it from causal sufficiency in a structural equation model.
- Karttunen's classification ignores aspect. Nadathur shows that be able is aspect-governed (perfective triggers entailment, imperfective doesn't).
- Karttunen groups force/cause/make together (ex. 56) as having the same entailment pattern (sufficient-only). Nadathur & Lauer show they have the same entailment pattern but different truth conditions: make/force assert causal sufficiency, while cause asserts causal necessity. This difference surfaces in overdetermination scenarios.
Karttunen's schemas: - (37) nec + suf: PRESUP v(S) is nec+suf for S. PROP v(S). - (41) nec + suf (neg): same but for ¬S. - (54) nec only: PRESUP v(S) is nec for S. PROP v(S). - (59) suf only: PRESUP v(S) is suf for S. PROP v(S). - neither: no complement entailment.
Karttunen's descriptive classification of complement-entailing predicates as a 2×2: necessary × sufficient × polarity.
This is the historical taxonomy from the 1971 paper. The modern
causal analysis uses ImplicativeClass (which adds aspectGoverned)
and Causative (which distinguishes causal mechanisms).
- isSufficient : Bool
v(S) is sufficient for S: affirmative entails complement.
- isNecessary : Bool
v(S) is necessary for S: negation entails ¬complement.
- polarity : Features.Implicative
Positive (manage: entails S) vs negative (fail: entails ¬S).
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
Equations
Equations
- Karttunen1971.KarttunenClass.manage = { isSufficient := true, isNecessary := true, polarity := Features.Implicative.positive }
Instances For
Equations
- Karttunen1971.KarttunenClass.fail = { isSufficient := true, isNecessary := true, polarity := Features.Implicative.negative }
Instances For
Sufficient only: "forced X to VP" → VP; "didn't force" ↛ ¬VP.
Equations
- Karttunen1971.KarttunenClass.force = { isSufficient := true, isNecessary := false, polarity := Features.Implicative.positive }
Instances For
Sufficient only, negative: "prevented X from VP-ing" → ¬VP.
Equations
- Karttunen1971.KarttunenClass.prevent = { isSufficient := true, isNecessary := false, polarity := Features.Implicative.negative }
Instances For
Necessary only: "wasn't able to VP" → ¬VP; "was able" ↛ VP.
Equations
- Karttunen1971.KarttunenClass.beAble = { isSufficient := false, isNecessary := true, polarity := Features.Implicative.positive }
Instances For
Neither: hope, want, intend.
Equations
- Karttunen1971.KarttunenClass.hope = { isSufficient := false, isNecessary := false, polarity := Features.Implicative.positive }
Instances For
Equations
- k.isTwoWay = (k.isNecessary && k.isSufficient)
Instances For
Equations
- k.hasEntailment = (k.isNecessary || k.isSufficient)
Instances For
manage and fail differ only in polarity.
force and prevent differ only in polarity.
Verify that Fragment verb entries carry the correct annotations, matching Karttunen's inventory (ex. 2, p.341).
happen is a raising verb, not subject-control. "It happened to rain" is grammatical — the matrix subject receives no theta role from happen. Karttunen (§9) describes happen's presupposition as chance-dependence, but does not discuss its syntactic control type.
dare and bother have both presupposition (occasion verbs) AND implicative entailment: "John dared to speak" presupposes risk AND entails "John spoke." These are compatible per Karttunen §9.
The key diagnostic: factives preserve complement presupposition under negation; implicatives reverse the complement entailment.
"John didn't realize he had no money" — still presupposes "he had no money."
"John didn't manage to solve it" — entails "he didn't solve it."
The grounding chain: KarttunenClass → Implicative → causal semantics → complement entailment.
For sufficient-positive classes, the chain is:
`KarttunenClass.manage.polarity` = `.positive`
→ `Implicative.positive.toSemantics` = `manageSem`
→ `manage_entails_complement`: manageSem sc → complement true
These theorems derive the entailment from the classification,
not just re-export the theory-layer theorem.
sufficient_positive_class_entails / sufficient_negative_class_entails
/ manage_class_entails / fail_class_entails — these grounded
KarttunenClass.polarity.toSemantics chain theorems were over the
legacy ImplicativeScenario/normalDevelopment substrate; deleted
in Phase D-H. The polymorphic V2 analog is direct: for any
BoolSEM V and a positive (resp. negative) KarttunenClass,
Implicative.toSemantics M .positive bg p xP c xC IS
Implicative.manageSem M bg p xP c xC IS
SEM.causallySufficient M bg p xP c xC (rfl-chain).
Double negation cancellation is a signature property of implicative verbs. Karttunen's examples:
- (13) "John didn't remember not to lock his door" → "John locked his door."
- (40a) "John didn't forget to lock his door" → (40d) "John locked his door."
The current causal semantics models the *positive* direction
(manageSem → complement true) and the *negative* direction
(failSem → complement false) separately. Full double negation
— where matrix negation and complement negation interact to yield
a positive entailment — would require compositional negation over
the causal model, which is not yet formalized.
What we CAN verify: the two directions (positive entailment, negative
entailment) are separately grounded, and two-way KarttunenClasses
predict both directions.
Two-way classes predict entailment in BOTH directions: the positive polarity grounds the affirmative direction, the negative polarity grounds the negation direction.
All four cells of the 2×2 are populated by Fragment entries.
be able is NOT a lexical implicative — it has no implicative.
The actuality entailment is aspect-governed ([Nad23a]):
perfective "was able to VP" → VP; imperfective "was able to VP" ↛ VP.
Karttunen (§11) notes these verbs are ambiguous between implicative
and non-implicative readings.
be able takes infinitival complement with subject control. "He was able to leave" — the subject has the ability (theta role).
Tension with Noonan's Reality Status #
[noonan-2007] classifies achievement CTPs (*manage*, *fail*) as
IRREALIS because they take infinitival complements. But Karttunen shows
these verbs ENTAIL complement truth — semantically realis. Complement
*form* (irrealis) and complement *entailment* (realis) diverge for
implicative verbs.