Psych Verb Causal Links #
@cite{kim-2024} @cite{allen-1983} @cite{bach-1986} Formal integration of @cite{kim-2024}'s maintenance relation with existing @cite{lewis-1973} infrastructure: temporal intervals, event sorts, and counterfactual semantics.
Kim's core claim: stative Class II psych verbs involve a maintenance causal relation, not eventive causation. The two flavors differ along three dimensions:
| Property | Eventive | Maintenance |
|---|---|---|
| Cause sort | event (external percept) | state (mental representation) |
| Effect | BECOME + state (transition) | state only (no transition) |
| Temporal | cause precedes effect | cause and effect contemporaneous |
| Counterfactual | effect persists after cause ceases | effect ceases with cause |
The first three properties are formalized using existing Linglib types:
EventSort, Interval.precedes/.overlaps.
The fourth uses universalCounterfactual from Counterfactual.lean.
Key results #
- Maintenance is temporally symmetric (states overlap); eventive is asymmetric
- Temporal precedence and overlap are mutually exclusive (the two flavors can't hold simultaneously for the same pair of eventualities)
CausalSource.toLinkgrounds the two-constructor enum in event structure
A causal link between two eventualities in psych verb semantics.
Bundles event-structural and temporal properties that distinguish eventive causation (percept → state change) from maintenance causation (representation → state persistence).
- causeSort : Events.EventSort
Ontological sort of the causing eventuality
- effectSort : Events.EventSort
Ontological sort of the caused eventuality
- involvesTransition : Bool
Does the effect involve a transition (BECOME in @cite{rappaport-hovav-levin-1998})? Eventive: [CAUSE [BECOME [STATE]]] — yes. Maintenance: [CAUSE [STATE]] — no.
- temporalConstraint : Core.Time.Interval Time → Core.Time.Interval Time → Prop
Temporal constraint on the runtimes of cause and effect
Instances For
Eventive causation: an external percept/event CAUSES a change of mental state. The cause temporally precedes the effect.
Event structure: [[percept ACT] CAUSE [BECOME [experiencer STATE]]] Example: "The noise frightened John" — the noise event happens, THEN John enters the frightened state.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Maintenance causation: a mental representation MAINTAINS a psychological state. Cause and effect are temporally contemporaneous.
Event structure: [[representation STATE] CAUSE [experiencer STATE]] Example: "The problem concerns John" — the problem's mental representation and John's concern state coexist; if the representation ceased, the concern would cease.
@cite{kim-2024} identifies three defining properties: (a) Relates two eventualities (both states) (b) Temporal contemporaneity (τ(cause) overlaps τ(effect)) (c) Counterfactual dependence (effect ceases when cause ceases)
Equations
- One or more equations did not get rendered due to their size.
Instances For
Ground CausalSource (a two-constructor enum) in the richer
PsychCausalLink structure. External source = eventive causation;
internal source = maintenance causation.
Equations
- Semantics.Causation.PsychLink.CausalSource.toLink Time Semantics.Causation.Psych.CausalSource.external = Semantics.Causation.PsychLink.eventiveLink Time
- Semantics.Causation.PsychLink.CausalSource.toLink Time Semantics.Causation.Psych.CausalSource.internal = Semantics.Causation.PsychLink.maintenanceLink Time
Instances For
Maintenance is temporally symmetric: if cause overlaps effect,
effect overlaps cause. Delegates to Interval.overlaps_symm.
Eventive causation is temporally irreflexive: no eventuality
can precede itself. Delegates to Interval.precedes_irrefl.
Precedence and overlap are mutually exclusive: if cause precedes
effect, they cannot overlap. This is the structural basis for the
eventive/stative dichotomy — the two temporal configurations are
incompatible for any given pair of eventualities.
Delegates to Interval.precedes_not_overlaps.
Maintenance relates two states (@cite{kim-2024} property (a)).
Eventive causation relates two dynamic eventualities.
Maintenance involves no transition (no BECOME).
Eventive causation involves a transition (BECOME).
The two causal flavors assign opposite values on every dimension.
Counterfactual dependence: in the closest worlds where the cause doesn't hold, the effect doesn't hold either.
¬cause □→ ¬effect
This characterizes maintenance causation (@cite{kim-2024} property (c)): "The problem concerns John" — in the closest worlds where John no longer has the mental representation, the concern ceases.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Semantics.Causation.PsychLink.counterfactuallyDependent_decidable sim causeProp effectProp w = Semantics.Causation.PsychLink.counterfactuallyDependent_decidable._aux_1 sim causeProp effectProp w
Counterfactual persistence: in the closest worlds where the cause doesn't hold, the effect STILL holds.
¬cause □→ effect
This characterizes eventive causation: "The noise frightened John" — even if the noise hadn't occurred (in the closest worlds), the frightened state, once established by BECOME, persists independently.
Equations
- Semantics.Causation.PsychLink.counterfactuallyPersistent sim causeProp effectProp w = Semantics.Conditionals.Counterfactual.universalCounterfactual sim (fun (w : W) => ¬causeProp w) effectProp w
Instances For
Equations
- Semantics.Causation.PsychLink.counterfactuallyPersistent_decidable sim causeProp effectProp w = Semantics.Causation.PsychLink.counterfactuallyPersistent_decidable._aux_1 sim causeProp effectProp w
Counterfactual dependence and persistence are mutually exclusive when the set of closest ¬cause worlds is non-empty.
If all closest ¬cause worlds satisfy ¬effect (dependence), then at least one closest world falsifies effect, so persistence fails.
Proved by extracting a witness from the non-empty closest-world
list and showing it satisfies ¬ effectProp w (from dependence)
and effectProp w (from persistence), a contradiction.
The three defining properties of maintenance causation from @cite{kim-2024}, formalized using existing infrastructure:
(a) Relates two eventualities — both are states (EventSort.state)
(b) Temporal contemporaneity — Interval.overlaps
(c) No transition — effect is a persisting state, not a change
Property (c) is formalized structurally (no BECOME) rather than via counterfactuals, because the counterfactual characterization ("effect ceases when cause ceases") requires a concrete world space and similarity ordering. The structural version — no BECOME means no independent grounding for the effect — is the deeper explanation for WHY maintenance-caused states are counterfactually dependent.