@cite{rett-2020}: Antonymy + Aspectual Coercion #
@cite{rett-2020} @cite{rett-2026} @cite{jin-koenig-2021} @cite{krifka-2010b}before and after are antonyms on converse scales, with strong defaults @cite{de-swart-1998} @cite{dolling-2014} (before-start, after-finish). Non-default readings require aspectual coercion: INCHOAT (GLB, atelic → onset) or COMPLET (LUB, telic → telos), which incur processing cost.
Rett's formal analysis (eqs. 22a-b):
- ⟦A before B⟧ = ∃t ∈ A [t ≺ MAX(B_≺)]
- ⟦A after B⟧ = ∃t ∈ A [t ≻ MAX(B_≻)]
Both theories use ∃ over the main clause A: "some time in A bears the relation to (some characterization) B." They differ in how B's reference point is selected (all of B vs MAX of B).
Level #
Level 2 (interval sets): operates on SentDenotation directly, using
maxOnScale from Core.Scale to select the informative bound.
Bridges #
INCHOATextracts the same point asCoSType.inception(onset of a state)COMPLETextracts the dual: the finish point of a telic eventstativeDenotationhas the subinterval property (connects to Krifka CUM)- Both theories agree on unambiguous cases (stative before, telic after)
Ambidirectionality #
before is truth-conditionally insensitive to negation of its argument (ambidirectional), which is why it licenses expletive negation cross- linguistically. after and while are not ambidirectional.
Rett's before (eq. 22a): ∃t ∈ times(A) [t ≺ MAX(times(B)_≺)]. Some time in A precedes the maximal (on the ≺ scale) time of B.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Rett's after (eq. 22b): ∃t ∈ times(A) [t ≻ MAX(times(B)_≻)]. Some time in A succeeds the maximal (on the ≻ scale) time of B.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Inchoative coercion / GLB (@cite{rett-2020}, eq. 19; @cite{de-swart-1998}; @cite{dolling-2014}). Maps a process (atelic) denotation to a singleton containing its onset point. GLB(T) = ιt[t ∈ T ∧ ∀t' ∈ T, t ≤ t']
Linguistically: "Amy was surprised" → "the start of Amy being surprised". Cross-linguistically realized as inchoative morphology (Russian -sja, Tagalog PFV.NEUT).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Completive coercion / LUB (@cite{rett-2020}, eq. 21; @cite{dolling-2014}). Maps a culmination (telic) denotation to a singleton containing its telos. LUB(T) = ιt[t ∈ T ∧ ∀t' ∈ T, t ≥ t']
Linguistically: "Jane climbed the mountain" → "the moment Jane reached the top". Cross-linguistically realized as completive morphology (Tagalog AIA).
Equations
- One or more equations did not get rendered due to their size.
Instances For
INCHOAT extracts the start point of a stative denotation.
COMPLET extracts the finish point of an accomplishment denotation.
Both theories predict "before-start" for statives.
When B is stative (subinterval-closed), both theories reduce to "some time in A precedes all times in B":
- Anscombe directly: ∃ t ∈ A, ∀ t' ∈ B, t < t'
- Rett: ∃ t ∈ A, t < MAX(B_≺). For statives, MAX on ≺ picks B.start (the GLB), and t < B.start ↔ t < all times in B.
Rett's analysis implies Anscombe's for telic "after".
The converse does NOT hold: Anscombe.after only requires some point of B to precede some point of A (∃ t' ∈ B, t' < t), while Rett requires A to follow B's finish (t > MAX₍>₎(B) = i_B.finish). These differ when A overlaps B without extending past B's endpoint.
Rett's before implies Anscombe's before in general.
From Rett: t < m where m = min(timeTrace B). Since m < all other points in timeTrace B (by maxOnScale), t < every point in timeTrace B. This gives Anscombe's ∀-quantified conclusion.
Rett's after implies Anscombe's after in general.
Immediate: m ∈ maxOnScale(timeTrace B) implies m ∈ timeTrace B, and t > m gives the existential witness for Anscombe.after.
Expletive negation and ambidirectionality #
@cite{rett-2026} shows that before is ambidirectional: negating B in "A before B" doesn't change truth conditions. This is why before-clauses license expletive negation cross-linguistically (@cite{jin-koenig-2021}: 50 of 74 EN-attesting languages, from a 722-language survey).
The mechanism: for B = [s, f], both B and its pre-event complement (−∞, s] share s as their "most informative closed bound" on the < scale. The before construction relates A only to this bound, so negating B is truth-conditionally vacuous.
After is NOT ambidirectional: negating B shifts the relevant bound. While requires total temporal overlap; ¬B fails when A overlaps B.
Before truth conditions depend only on MAX₍<₎ of B's time trace.
When B's time trace is a closed interval [s, f], Rett.before reduces to "∃ t ∈ A, t < s".
COMPLET on a stative denotation extracts the finish point.
The pre-event complement of an event interval [s, f].
Equations
- Semantics.Tense.TemporalConnectives.preEventDenotation bot i hbot = Semantics.Tense.TemporalConnectives.stativeDenotation { start := bot, finish := i.start, valid := hbot }
Instances For
The time trace of a stative denotation is the closed interval [start, finish].
MAX₍<₎ of a stative denotation's time trace is {start}.
The time trace of COMPLET(preEventDenotation bot i) is the degenerate
interval {i.start}.
MAX₍<₎ of the COMPLET of a pre-event denotation is {start}.
Before is truth-conditionally insensitive to event polarity.
Both select the same boundary point s through different mechanisms: the original uses the default before-start reading (MAX₍<₎), while the negated version requires COMPLET coercion to extract the end of the pre-event interval.
After is NOT ambidirectional: negating B changes truth conditions because MAX₍>₎(B) ≠ MAX₍>₎(¬B).
While is not ambidirectional.
Temporal Connective Truth-Condition Examples #
@cite{heinamaki-1974} @cite{rett-2020} @cite{karttunen-1974}
Concrete scenarios verifying that the Anscombe, Rett, and Karttunen formalizations produce correct truth-value judgments.
Scenarios 1–6 verify @cite{rett-2020} Table 1 for before/after. Scenarios 7–10 verify @cite{heinamaki-1974} Chs. 6, 8, 9 for since, by, till.
Scenarios (ℕ time points) #
| # | ME | EE | Connective | Result |
|---|---|---|---|---|
| 1 | point(1) | stative [5,10] | before | True |
| 2 | point(12) | stative [5,10] | after | True |
| 3 | point(1) | accomplishment [3,8] | before | True |
| 4 | point(12) | accomplishment [3,8] | after | True |
| 5 | point(7) | stative [5,10] | before | False |
| 6 | point(7) | stative [5,10] | after | False |
| 7 | stative[5,10] | point(5) | since | True |
| 8 | point(1) | point(3) | by | True |
| 9 | point(3) | point(3) | by | True |
| 9' | point(3) | point(3) | before | False |
| 10 | stative[5,10] | point(5) | till | True |
ME: "John left" — punctual event at time 1 (early).
Instances For
ME: "John left" — punctual event at time 12 (late).
Instances For
ME: punctual event at time 7 (inside the stative EE).
Instances For
EE: "she was president" — stative, running [5, 10].
Equations
- Phenomena.TemporalConnectives.Examples.ee_stative = { start := 5, finish := 10, valid := Phenomena.TemporalConnectives.Examples.ee_stative._proof_2 }
Instances For
EE: "she climbed the mountain" — accomplishment, [3, 8].
Equations
- Phenomena.TemporalConnectives.Examples.ee_accomplishment = { start := 3, finish := 8, valid := Phenomena.TemporalConnectives.Examples.ee_accomplishment._proof_2 }
Instances For
Equations
Instances For
"John left₁ before she was president₅₋₁₀" — True under Anscombe. Witness: t = 1 ∈ ME, and 1 < all t' ∈ [5, 10].
"John left₁ before she was president₅₋₁₀" — True under Rett. Witness: t = 1, m = 5 (the GLB of [5, 10] on the ≺ scale).
"John left₁₂ after she was president₅₋₁₀" — True under Anscombe. Witness: t = 12, t' = 5 (any point in EE suffices).
"John left₁₂ after she was president₅₋₁₀" — True under Rett. Witness: t = 12, m = 10 (the LUB of [5, 10] on the ≻ scale).
"John met Mary₁ before she climbed the mountain₃₋₈" — True under Anscombe. Witness: t = 1, and 1 < all t' in [3, 8].
"John met Mary₁ before she climbed the mountain₃₋₈" — True under Rett. Witness: t = 1, m = 3 (the GLB of [3, 8] on the ≺ scale).
"John met Mary₁₂ after she climbed the mountain₃₋₈" — True under Anscombe. Witness: t = 12, t' = 3.
"John met Mary₁₂ after she climbed the mountain₃₋₈" — True under Rett. Witness: t = 12, m = 8 (the LUB of [3, 8] on the ≻ scale).
"John left₇ before she was president₅₋₁₀" — False under Anscombe. Any witness t from ME (t=7) fails: t'=7 ∈ EE and ¬(7 < 7).
"John left₇ after she was president₅₋₁₀" — False under Rett. The max on the ≻ scale is 10, and ¬(7 > 10).
Both theories agree on scenario 1 (ME before stative EE).
Both theories agree on scenario 2 (ME after stative EE).
Both theories agree on scenario 3 (ME before accomplishment EE).
Both theories agree on scenario 4 (ME after accomplishment EE).
INCHOAT of "she was president₅₋₁₀" = {point(5)}. Verifies that inchoative coercion extracts the onset.
COMPLET of "she climbed the mountain₃₋₈" = {point(8)}. Verifies that completive coercion extracts the telos.
ME: stative [5, 10] — "He has been happy."
Equations
Instances For
EE: punctual event at time 5 — "she arrived."
Instances For
ME: punctual event at time 3 — "arrived at 3pm" (for by coincidence case).
Instances For
Equations
Instances For
EE: punctual event at time 3 — "3pm" (deadline).
Instances For
Equations
Instances For
"He has been happy₅₋₁₀ since she arrived₅" — True under Karttunen.since.
Witness: t = 5 ∈ B, and ∀t' ∈ A (i.e., 5 ≤ t' ≤ 10), 5 ≤ t'.
Since is veridical for its complement in scenario: she arrived.
"He arrived₁ by 3pm₃" — True under Karttunen.by_.
Witness: t = 1 ∈ A, and ∀t' ∈ B (t' = 3), 1 ≤ 3.
"He arrived₃ by 3pm₃" — True under Karttunen.by_.
Witness: t = 3 ∈ A, 3 ≤ 3 (coincidence allowed).
"He arrived₃ before 3pm₃" — FALSE under Anscombe.before.
Need 3 < 3, which fails. Shows by ⊋ before.
By but not before on the same scenario: demonstrates the strict
weakening from before_implies_by.
"He slept₅₋₁₀ till she arrived₅" — True under Karttunen.till.
Witness: t = 5 ∈ both time traces (overlap).
Till agrees with until on the same scenario (definitional).
Not...until scenarios #
Karttunen's identity: punctual until = ¬before (eq. 33). We verify this on concrete time points.
| # | ME | EE | Construction | Result |
|---|---|---|---|---|
| 11 | point(3) | point(5) | not...until | True |
| 12 | point(3) | point(5) | presup + assert | when |
| 13 | point(7) | point(5) | not...until | False |
ME: "The princess woke up" — punctual event at time 3 (early).
Instances For
EE: "The prince kissed her" — punctual event at time 5.
Instances For
Equations
Instances For
Scenario 11: "The princess woke up₃ before the prince kissed her₅" — TRUE. 3 < 5. This is the base before that gets negated in punctual until.
Scenario 11': "The princess didn't wake up₃ until the prince kissed her₅" — FALSE. NOT(BEFORE) = NOT(wake₃ before kiss₅) = ¬(3 < 5) = False. The princess DID wake up before the kiss, so "not until" is false.
ME: "The princess woke up" — punctual event at time 5 (AT the kiss).
Instances For
Equations
Instances For
Scenario 12: "The princess didn't wake up₅ until the prince kissed her₅" — TRUE. NOT(BEFORE) = NOT(wake₅ before kiss₅) = ¬(5 < 5) = True. She woke up at exactly the time of the kiss.
Scenario 12': Presupposition (wake₅ before kiss₅ ∨ wake₅ when kiss₅) is satisfied: the waking happens at the kiss time (when), so the left disjunct is false but the right is true.
Scenario 12'': Presupposition + assertion → when (disjunctive syllogism). This is Karttunen's key result: "not until" + presupposition derives "when".
ME: "The princess woke up" — punctual event at time 7 (AFTER the kiss).
Instances For
Equations
Instances For
Scenario 13: "The princess didn't wake up₇ until the prince kissed her₅" — TRUE. NOT(BEFORE) = NOT(wake₇ before kiss₅) = ¬(7 < 5) = True. She woke up after the kiss, so she didn't wake up before it.
Scenario 13': wake₇ is NOT when kiss₅ (no overlap at any time point).
Scenario 13'': The presupposition (before ∨ when) is NOT satisfied, so not...until is vacuously true but pragmatically infelicitous. This models why "She didn't wake up until the prince kissed her" is odd when she woke up AFTER the kiss — the presupposition fails.