Events #
This file defines the temporal trace of an event domain. Neo-Davidsonian event semantics, after
Davidson and Parsons, quantifies over events as individuals of their own sort: an event domain is
a type E, verbs denote predicates on it, and thematic roles relate its events to their
participants. The temporal trace τ sends each event to its run time, as in Krifka and
Champollion, and need not be injective, since two distinct events can go on at the same time.
Main definitions #
Event.TemporalTrace: the temporal trace of an event domain.
Implementation notes #
- Parthood is an order on the event domain from
Semantics/Mereology.lean. Laws relating it to the trace, such as Krifka's and Champollion's requirement that τ preserve sums, are hypotheses of the theorems that need them, and theories that never read a run time take no trace. - τ is total, where Champollion lets trace functions be partial. Krifka's richness, that every
time is the run time of some event, is the hypothesis
Function.Surjective τ. - Run times are
NonemptyInterval T, so the run time of a sum is the convex hull of the run times of its parts, where Krifka's times form a part structure with non-convex sums. - Bach divides eventualities into states and non-states; here the division is a property of
predicates (
Aspect.Dynamicity,Aspect.SortedProperty), not a field of events. - Intervals form an event domain with the identity trace, in which the studies build their satisfiability witnesses.
References #
The temporal trace of an event domain E, which sends each event to its run time.
- τ : E → NonemptyInterval T
The run time of an event.
Instances
@[instance_reducible]
instance
Event.instTemporalTraceNonemptyInterval
{T : Type u_1}
[LE T]
:
TemporalTrace (NonemptyInterval T) T
Each interval is an event whose run time is itself.
Equations
- Event.instTemporalTraceNonemptyInterval = { τ := id }
@[simp]