Entailment Profiles and the Argument Selection Principle #
@cite{dowty-1991} @cite{davis-koenig-2000} @cite{grimm-2011} @cite{levin-2019} @cite{beavers-2010}
Entailment profiles encode the lexical entailments a verb imposes on each of its arguments, reflecting the modern consensus on proto-roles (@cite{levin-2019}). Proto-Agent and Proto-Patient are cluster concepts: each is a set of entailments, and an argument's "degree of agenthood/patienthood" is determined by which entailments it satisfies.
Entailment structure #
The 10 entailments are not fully independent (@cite{levin-2019} §2.1):
- P-Agent pairing: three P-Agent entailments are paired with P-Patient entailments in asymmetric relations (causation↔changeOfState, movement↔stationary, independentExistence↔dependentExistence).
- P-Agent dependency: volition presupposes sentience.
- P-Agent priority: causation outranks other P-Agent entailments for subject selection (@cite{davis-koenig-2000}).
- P-Patient implicational structure: the affectedness-related entailments (changeOfState, incrementalTheme, causallyAffected) form an implicational hierarchy (@cite{beavers-2010}).
These dependencies are exploited by the modern ASP (lattice comparison, §3–4)
and formalized algebraically in AgentivityLattice.lean:
- P-Agent →
AgentivityNode(4 privative features on a lattice; @cite{grimm-2011}) - P-Patient →
PersistenceLevel(4 persistence dimensions; @cite{grimm-2011}), andAffectednessDegree(implicational hierarchy of truth-conditional strength; @cite{beavers-2010})
Argument Selection Principle (lattice-based) #
The ASP uses lattice comparison (@cite{grimm-2011}): argument A outranks argument B for subjecthood iff A's P-Agent feature set dominates (is a superset of) B's. When P-Agent features are incomparable, P-Patient features break the tie. This naturally handles priority because feature-set inclusion respects the entailment dependencies.
For unaccusativity, the priority-based approach checks whether the sole argument has core agentive features (volition/causation) rather than flat-counting.
@cite{dowty-1991}'s original flat-counting ASP is preserved in
Phenomena/ArgumentStructure/Studies/Dowty1991.lean for comparison.
The 10 entailments defining Proto-Agent and Proto-Patient (@cite{dowty-1991} pp.572–573).
Proto-Agent entailments (a)–(e):
- volition, sentience, causation, movement, independent existence
Proto-Patient entailments (a)–(e):
- change of state, incremental theme, causally affected, stationary, dependent existence
- volition : Bool
(a) volitional involvement in the event
- sentience : Bool
(b) sentience/perception
- causation : Bool
(c) causing an event or change of state in another participant
- movement : Bool
(d) movement (relative to another participant)
- independentExistence : Bool
(e) exists independently of the event named by the verb
- changeOfState : Bool
(a) undergoes change of state
- incrementalTheme : Bool
(b) incremental theme (Krifka's SINC)
- causallyAffected : Bool
(c) causally affected by another participant
- stationary : Bool
(d) stationary relative to another participant
- dependentExistence : Bool
(e) does not exist independently of the event
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
Count of Proto-Agent entailments. Informational — the modern ASP uses lattice comparison, not counting. Retained for bridge theorems and backward compatibility with study files.
Equations
- p.pAgentScore = p.volition.toNat + p.sentience.toNat + p.causation.toNat + p.movement.toNat + p.independentExistence.toNat
Instances For
Count of Proto-Patient entailments.
Equations
- p.pPatientScore = p.changeOfState.toNat + p.incrementalTheme.toNat + p.causallyAffected.toNat + p.stationary.toNat + p.dependentExistence.toNat
Instances For
P-Agent feature dominance: p has every P-Agent feature that q has.
This is the subset ordering on P-Agent feature sets.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Semantics.ArgumentStructure.EntailmentProfile.instDecidablePAgentDominates p q = id inferInstance
P-Patient feature dominance: p has every P-Patient feature that q has.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Semantics.ArgumentStructure.EntailmentProfile.instDecidablePPatientDominates p q = id inferInstance
Strict P-Agent dominance: p dominates q but not vice versa.
Means p has a strict superset of q's P-Agent features.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Semantics.ArgumentStructure.EntailmentProfile.instDecidablePAgentStrictlyDominates p q = id inferInstance
Strict P-Patient dominance.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Semantics.ArgumentStructure.EntailmentProfile.instDecidablePPatientStrictlyDominates p q = id inferInstance
Modern ASP: subj outranks obj for subjecthood iff:
(1) subj strictly dominates obj on P-Agent features
(subj's P-Agent set ⊃ obj's P-Agent set), OR
(2) Neither strictly dominates the other on P-Agent, AND obj
strictly dominates subj on P-Patient (obj is a "better"
object, hence subj wins by exclusion).
This replaces @cite{dowty-1991}'s flat counting with lattice comparison (@cite{grimm-2011}, @cite{davis-koenig-2000}). Priority of causation is captured structurally: {causation, IE} ⊃ {IE} but {causation, IE} ⊥ {sentience, IE} (incomparable).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Semantics.ArgumentStructure.EntailmentProfile.instDecidableOutranksForSubject subj obj = id inferInstance
Corollary 1 (@cite{dowty-1991} p.579): Neither argument outranks the other → alternation is possible (buy/sell, like/please).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Semantics.ArgumentStructure.EntailmentProfile.instDecidableAllowsAlternation arg1 arg2 = id inferInstance
Predicts unaccusativity for intransitive verbs.
An intransitive subject is unaccusative iff it lacks core agentive
features (volition AND causation) and has at least one P-Patient
feature. This matches @cite{dowty-1991}'s Table 1 cell 4 and fixes
the arrive anomaly that flat counting gets wrong.
@cite{davis-koenig-2000}: causation and volition are the priority P-Agent entailments. Their absence signals non-agentivity.
Equations
- Semantics.ArgumentStructure.EntailmentProfile.PredictsUnaccusative p = (p.volition = false ∧ p.causation = false ∧ p.pPatientScore > 0)
Instances For
Equations
- Semantics.ArgumentStructure.EntailmentProfile.instDecidablePredictsUnaccusative p = id inferInstance
Predicts unergative for intransitive verbs.
An intransitive subject is unergative iff it has at least one core agentive feature (volition or causation) and no P-Patient features.
Equations
- Semantics.ArgumentStructure.EntailmentProfile.PredictsUnergative p = ((p.volition = true ∨ p.causation = true) ∧ p.pPatientScore = 0)
Instances For
Equations
- Semantics.ArgumentStructure.EntailmentProfile.instDecidablePredictsUnergative p = id inferInstance
Intra-profile well-formedness: volition presupposes sentience. Only sentient entities can be volitional (@cite{dowty-1991} p.607, @cite{levin-2019} §2.1).
Equations
- p.WellFormedInternal = (p.volition = true → p.sentience = true)
Instances For
Equations
- Semantics.ArgumentStructure.EntailmentProfile.instDecidableWellFormedInternal p = id inferInstance
Inter-argument well-formedness (@cite{davis-koenig-2000}, @cite{primus-1999}): three P-Agent entailments are paired with P-Patient entailments in asymmetric relations.
- causation (subject) → changeOfState (object)
- movement (subject) → stationary (object)
- independentExistence (subject) → dependentExistence (object)
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Semantics.ArgumentStructure.EntailmentProfile.instDecidableWellFormedPair subj obj = id inferInstance
Equations
- Semantics.ArgumentStructure.EntailmentProfile.instDecidableEqDoTestSource x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- One or more equations did not get rendered due to their size.
Instances For
The do-test passes (semantically) iff at least one of {volition, causation, movement} holds.
Equations
- Semantics.ArgumentStructure.EntailmentProfile.PassesDoTestFromProfile p = (p.volition = true ∨ p.causation = true ∨ p.movement = true)
Instances For
Equations
- Semantics.ArgumentStructure.EntailmentProfile.instDecidablePassesDoTestFromProfile p = id inferInstance
An effector (@cite{van-valin-wilkins-1996}) is a self-energetic force bearer: movement + independent existence. Realized as external argument.
Equations
- Semantics.ArgumentStructure.EntailmentProfile.IsEffector p = (p.movement = true ∧ p.independentExistence = true)
Instances For
Equations
- Semantics.ArgumentStructure.EntailmentProfile.instDecidableIsEffector p = id inferInstance
A force recipient is causally affected or stationary. Realized as internal argument.
Equations
- Semantics.ArgumentStructure.EntailmentProfile.IsForceRecipient p = (p.causallyAffected = true ∨ p.stationary = true)
Instances For
Equations
- Semantics.ArgumentStructure.EntailmentProfile.instDecidableIsForceRecipient p = id inferInstance
Effectors have at least 2 P-Agent entailments (movement + IE).
"kick" subject: V+S+C+M+IE (prototypical agent, 5 P-Ag)
Equations
- One or more equations did not get rendered due to their size.
Instances For
"kick" object: CoS+CA+St (3 P-Pat)
Equations
- One or more equations did not get rendered due to their size.
Instances For
"build" subject: V+S+C+M+IE (5 P-Ag)
Equations
- One or more equations did not get rendered due to their size.
Instances For
"build" object: CoS+IT+CA+DE (4 P-Pat, incremental theme, dependent existence)
Equations
- One or more equations did not get rendered due to their size.
Instances For
"see" subject: S+IE (experiencer, 2 P-Ag)
Equations
- One or more equations did not get rendered due to their size.
Instances For
"run" subject: V+S+M+IE (4 P-Ag, unergative)
Equations
- One or more equations did not get rendered due to their size.
Instances For
"arrive" subject: M+IE+CoS (2 P-Ag, 1 P-Pat — unaccusative)
Equations
- One or more equations did not get rendered due to their size.
Instances For
"die" subject: CoS+CA+DE (0 P-Ag, 3 P-Pat — unaccusative)
Equations
- One or more equations did not get rendered due to their size.
Instances For
"eat" subject: V+S+C+M+IE (5 P-Ag)
Equations
- One or more equations did not get rendered due to their size.
Instances For
"eat" object: CoS+IT+CA (3 P-Pat, incremental theme)
Equations
- One or more equations did not get rendered due to their size.
Instances For
"buy" subject: V+S+C+IE (4 P-Ag)
Equations
- One or more equations did not get rendered due to their size.
Instances For
"sell" subject: V+S+C+IE (same as buy — 4 P-Ag)
Equations
- One or more equations did not get rendered due to their size.
Instances For
"sweep" basic-sense subject: M+IE (movement + independent existence). Underspecified for volition — agentivity is pragmatically resolved.
Equations
- One or more equations did not get rendered due to their size.
Instances For
"sweep" broom-sense subject: V+S+C+M+IE (obligatory agent, 5 P-Ag). Instrument lexicalization forces volition, sentience, causation.
Equations
- One or more equations did not get rendered due to their size.