[Dow91] Thematic Proto-Roles and Argument Selection

Study file connecting the proto-role theory (Semantics/ArgumentStructure/EntailmentProfile.lean and the Levin-class role-list map, Semantics/ArgumentStructure/LevinClass.lean) to argument selection phenomena. The paper's explicit per-argument entailment attributions are typed data rows in Data/ProtoRoles/Dowty1991.json (generated module Data.ProtoRoles.Dowty1991), checked against the class templates in the final section — including the divergences the templates do NOT encode (build-object stationary, break-object incremental themehood, the eat-object dependent-existence tension).

Dowty's original flat-counting ASP

[Dow91]'s Argument Selection Principle uses flat counting: the argument with the greatest number of Proto-Agent entailments is subject. The library's default ASP uses lattice comparison ([Gri11], [DK00b]), which handles priority and fixes anomalies like arrive. This study file preserves Dowty's original counting-based predictions to document where they succeed and where they diverge from the modern approach.

Key predictions formalized

• §9.1: Partially symmetric interactive predicates — volition as asymmetric P-Agent
• §9.2: Psych verb doublets — inchoative change-of-state breaks ties
• §9.3: Three verb classes (spray/load, break, hit) — CoS symmetry predicts alternation
• §12: Agentivity × telicity → unaccusativity (Table 1)

def Dowty1991.flatSelectsSubject (p : ArgumentStructure.EntailmentProfile) : Bool

[Dow91]'s original single-argument ASP (flat counting): an argument selects for subjecthood iff its P-Agent count exceeds its P-Patient count. Superseded by lattice-based OutranksForSubject in EntailmentProfile.lean.

Equations

• Dowty1991.flatSelectsSubject p = decide (p.pAgentScore > p.pPatientScore)

def Dowty1991.flatSelectsObject (p : ArgumentStructure.EntailmentProfile) : Bool

Equations

• Dowty1991.flatSelectsObject p = decide (p.pPatientScore > p.pAgentScore)

def Dowty1991.flatOutranksForSubject (arg1 arg2 : ArgumentStructure.EntailmentProfile) : Bool

[Dow91]'s between-argument comparison (flat counting): arg1 outranks arg2 for subjecthood iff arg1 has strictly more P-Agent entailments, OR they tie on P-Agent but arg2 has strictly more P-Patient entailments.

Equations

• Dowty1991.flatOutranksForSubject arg1 arg2 = (decide (arg1.pAgentScore > arg2.pAgentScore) || arg1.pAgentScore == arg2.pAgentScore && decide (arg2.pPatientScore > arg1.pPatientScore))

def Dowty1991.flatAllowsAlternation (arg1 arg2 : ArgumentStructure.EntailmentProfile) : Bool

Corollary 1 (flat counting): neither argument outranks the other → alternation.

Equations

• Dowty1991.flatAllowsAlternation arg1 arg2 = (!Dowty1991.flatOutranksForSubject arg1 arg2 && !Dowty1991.flatOutranksForSubject arg2 arg1)

def Dowty1991.flatPredictsUnaccusative (p : ArgumentStructure.EntailmentProfile) : Bool

[Dow91] Corollary 2 (flat counting): unaccusative iff pPatient > pAgent.

Equations

• Dowty1991.flatPredictsUnaccusative p = decide (p.pPatientScore > p.pAgentScore)

def Dowty1991.flatPredictsUnergative (p : ArgumentStructure.EntailmentProfile) : Bool

Complement: unergative iff pAgent > pPatient.

Equations

• Dowty1991.flatPredictsUnergative p = decide (p.pAgentScore > p.pPatientScore)

Verbs like kiss, embrace, marry denote actions requiring volitional involvement of two parties, but only the SUBJECT is entailed to be volitional. This single asymmetric P-Agent entailment predicts the transitive argument configuration.

def Dowty1991.kissSubjectProfile : ArgumentStructure.EntailmentProfile

"kiss" subject: V+M+IE — volitional, in motion, independently existing.

Equations

• Dowty1991.kissSubjectProfile = { volition := true, movement := true, independentExistence := true }

def Dowty1991.kissObjectProfile : ArgumentStructure.EntailmentProfile

"kiss" object: M+IE only — same minus volition.

Equations

• Dowty1991.kissObjectProfile = { movement := true, independentExistence := true }

theorem Dowty1991.kiss_subject_outranks : ArgumentStructure.OutranksForSubject kissSubjectProfile kissObjectProfile

The subject outranks the object (lattice: {V,M,IE} ⊃ {M,IE}).

theorem Dowty1991.kiss_subject_outranks_flat : flatOutranksForSubject kissSubjectProfile kissObjectProfile = true

Same result under Dowty's flat counting.

theorem Dowty1991.kiss_asymmetry_is_volition : kissSubjectProfile.pAgentScore = kissObjectProfile.pAgentScore + 1

Volition adds exactly 1 to the subject's P-Agent score.

theorem Dowty1991.kiss_collective_alternation : ArgumentStructure.AllowsAlternation kissSubjectProfile kissSubjectProfile

The collective intransitive ("Kim and Sandy kissed") is predicted: when both participants have symmetric volition, neither outranks.

Psych verbs come in doublets (like/please, fear/frighten) with reversed argument configurations. Under the stative reading, Experiencer and Stimulus have equal P-Agent scores → alternation is predicted.

theorem Dowty1991.psychStative_alternation : ArgumentStructure.AllowsAlternation ThetaRole.experiencer.canonicalProfile ThetaRole.stimulus.canonicalProfile

Stative: Experiencer and Stimulus have incomparable P-Agent sets ({S,IE} ⊥ {C,IE}) and equal P-Patient (both 0) → alternation.

def Dowty1991.expInchoativeProfile : ArgumentStructure.EntailmentProfile

Under inchoative interpretation, the Experiencer enters a new mental state → gains changeOfState (P-Patient entailment a).

Equations

• Dowty1991.expInchoativeProfile = { sentience := true, independentExistence := true, changeOfState := true }

def Dowty1991.stimProfile : ArgumentStructure.EntailmentProfile

Equations

• Dowty1991.stimProfile = ThetaRole.stimulus.canonicalProfile

theorem Dowty1991.psych_inchoative_stimulus_is_subject : ArgumentStructure.OutranksForSubject stimProfile expInchoativeProfile

Inchoative breaks the tie: Stimulus outranks Experiencer for subject because the Experiencer now has more P-Patient → Experiencer is a "better" object → Stimulus is subject. Predicts StimExp frame.

[Dow91] identifies three classes based on CoS distribution across non-subject arguments. When CoS is symmetric (both or neither), alternation is possible. When asymmetric, the CoS argument is fixed as DO. The comparison with [Lev93]'s alternation judgments lives in Studies/Levin1993.lean (dowty_* theorems).

def Dowty1991.cosSymmetric (arg1 arg2 : ArgumentStructure.EntailmentProfile) : Bool

Equations

• Dowty1991.cosSymmetric arg1 arg2 = (arg1.changeOfState == arg2.changeOfState)

def Dowty1991.sprayLoadTheme : ArgumentStructure.EntailmentProfile

Equations

• Dowty1991.sprayLoadTheme = { independentExistence := true, changeOfState := true, causallyAffected := true }

def Dowty1991.sprayLoadLocation : ArgumentStructure.EntailmentProfile

Equations

• Dowty1991.sprayLoadLocation = { independentExistence := true, changeOfState := true, causallyAffected := true, stationary := true }

theorem Dowty1991.sprayLoad_cos_symmetric : cosSymmetric sprayLoadTheme sprayLoadLocation = true

def Dowty1991.breakDirectObject : ArgumentStructure.EntailmentProfile

Equations

• Dowty1991.breakDirectObject = { changeOfState := true, incrementalTheme := true, causallyAffected := true, stationary := true }

def Dowty1991.breakInstrument : ArgumentStructure.EntailmentProfile

Equations

• Dowty1991.breakInstrument = { movement := true, independentExistence := true }

theorem Dowty1991.break_cos_asymmetric : cosSymmetric breakDirectObject breakInstrument = false

theorem Dowty1991.break_DO_more_patient : breakDirectObject.pPatientScore > breakInstrument.pPatientScore

def Dowty1991.hitArg1 : ArgumentStructure.EntailmentProfile

Equations

• Dowty1991.hitArg1 = { independentExistence := true, causallyAffected := true, stationary := true }

def Dowty1991.hitArg2 : ArgumentStructure.EntailmentProfile

Equations

• Dowty1991.hitArg2 = { movement := true, independentExistence := true, causallyAffected := true }

theorem Dowty1991.hit_cos_symmetric : cosSymmetric hitArg1 hitArg2 = true

theorem Dowty1991.hit_no_cos : hitArg1.changeOfState = false ∧ hitArg2.changeOfState = false

theorem Dowty1991.hit_no_IT : hitArg1.incrementalTheme = false ∧ hitArg2.incrementalTheme = false

[Dow91] Table 1: the interaction of agentivity (most salient P-Agent property) and telicity (most salient P-Patient property) predicts the unergative/unaccusative split. Only the two "pure" cells are stable; the mixed cells are where cross-linguistic variation occurs.

inductive Dowty1991.IntransClass : Type

• unergative : IntransClass
• unaccusative : IntransClass
• unstable : IntransClass

@[instance_reducible]

instance Dowty1991.instDecidableEqIntransClass : DecidableEq IntransClass

Equations

• Dowty1991.instDecidableEqIntransClass x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯

def Dowty1991.instReprIntransClass.repr : IntransClass → ℕ → Std.Format

Equations

• One or more equations did not get rendered due to their size.

@[instance_reducible]

instance Dowty1991.instReprIntransClass : Repr IntransClass

Equations

• Dowty1991.instReprIntransClass = { reprPrec := Dowty1991.instReprIntransClass.repr }

def Dowty1991.table1 (agentive telic : Bool) : IntransClass

Equations

• Dowty1991.table1 true false = Dowty1991.IntransClass.unergative
• Dowty1991.table1 false true = Dowty1991.IntransClass.unaccusative
• Dowty1991.table1 agentive telic = Dowty1991.IntransClass.unstable

theorem Dowty1991.cell1_unergative : table1 true false = IntransClass.unergative

theorem Dowty1991.cell4_unaccusative : table1 false true = IntransClass.unaccusative

theorem Dowty1991.cell2_unstable : table1 true true = IntransClass.unstable

theorem Dowty1991.cell3_unstable : table1 false false = IntransClass.unstable

theorem Dowty1991.run_cell1 : table1 ArgumentStructure.selfMotion.subjectProfile.volition ArgumentStructure.selfMotion.subjectProfile.changeOfState = IntransClass.unergative

theorem Dowty1991.die_cell4 : table1 ArgumentStructure.disappearance.subjectProfile.volition ArgumentStructure.disappearance.subjectProfile.changeOfState = IntransClass.unaccusative

The key divergence between [Dow91]'s flat counting and the modern priority-based ASP. Flat counting gets arrive wrong because it counts movement + IE (2 P-Agent) > changeOfState (1 P-Patient), predicting unergative. The modern ASP correctly identifies arrive as unaccusative because it lacks the priority features (volition, causation).

theorem Dowty1991.arrive_flat_wrong : flatPredictsUnaccusative ArgumentStructure.directedMotion.subjectProfile = false

Flat counting predicts arrive is NOT unaccusative (2 P-Ag > 1 P-Pat).

theorem Dowty1991.arrive_modern_correct : ArgumentStructure.PredictsUnaccusative ArgumentStructure.directedMotion.subjectProfile

Modern priority-based ASP correctly predicts arrive IS unaccusative.

theorem Dowty1991.arrive_table1_correct : table1 ArgumentStructure.directedMotion.subjectProfile.volition ArgumentStructure.directedMotion.subjectProfile.changeOfState = IntransClass.unaccusative

Table 1 also correctly predicts arrive as unaccusative (non-agentive + telic = cell 4).

theorem Dowty1991.arrive_anomaly_summary : table1 ArgumentStructure.directedMotion.subjectProfile.volition ArgumentStructure.directedMotion.subjectProfile.changeOfState = IntransClass.unaccusative ∧ ArgumentStructure.PredictsUnaccusative ArgumentStructure.directedMotion.subjectProfile ∧ flatPredictsUnaccusative ArgumentStructure.directedMotion.subjectProfile = false ∧ English.Predicates.Verbal.arrive.unaccusative = true

Agreement: Table 1 and the modern ASP converge on arrive being unaccusative. Flat counting diverges — this is the anomaly that motivated [DK00b]'s priority refinement.

theorem Dowty1991.die_both_agree : flatPredictsUnaccusative ArgumentStructure.disappearance.subjectProfile = true ∧ ArgumentStructure.PredictsUnaccusative ArgumentStructure.disappearance.subjectProfile

Flat counting gets die right (both methods agree).

theorem Dowty1991.run_both_agree : flatPredictsUnergative ArgumentStructure.selfMotion.subjectProfile = true ∧ ArgumentStructure.PredictsUnergative ArgumentStructure.selfMotion.subjectProfile

Flat counting gets run right (both methods agree).

These theorems verify that the English Fragment verb entries store exactly the Levin-class role-list profiles of RoleList.lean — the stored fields are derivable from the class map (sweep_instr is the deliberate instrument-sense override).

theorem Dowty1991.kick_subject_profile_matches : English.Predicates.Verbal.kick.subjectEntailments = some ArgumentStructure.mannerContact.subjectProfile

theorem Dowty1991.kick_object_profile_matches : English.Predicates.Verbal.kick.objectEntailments = some ArgumentStructure.contactObject

theorem Dowty1991.eat_subject_profile_matches : English.Predicates.Verbal.eat.subjectEntailments = some ArgumentStructure.consumption.subjectProfile

theorem Dowty1991.eat_object_profile_matches : English.Predicates.Verbal.eat.objectEntailments = some ArgumentStructure.consumptionObject

theorem Dowty1991.build_subject_profile_matches : English.Predicates.Verbal.build.subjectEntailments = some ArgumentStructure.creation.subjectProfile

theorem Dowty1991.build_object_profile_matches : English.Predicates.Verbal.build.objectEntailments = some ArgumentStructure.creationObject

theorem Dowty1991.run_subject_profile_matches : English.Predicates.Verbal.run.subjectEntailments = some ArgumentStructure.selfMotion.subjectProfile

theorem Dowty1991.arrive_subject_profile_matches : English.Predicates.Verbal.arrive.subjectEntailments = some ArgumentStructure.directedMotion.subjectProfile

theorem Dowty1991.see_subject_profile_matches : English.Predicates.Verbal.see.subjectEntailments = some ArgumentStructure.perception.subjectProfile

theorem Dowty1991.sweep_subject_profile_matches : English.Predicates.Verbal.sweep.subjectEntailments = some ArgumentStructure.wipeManner.subjectProfile

theorem Dowty1991.sweep_instr_subject_profile_matches : English.Predicates.Verbal.sweep_instr.subjectEntailments = some ArgumentStructure.wipeInstrument.subjectProfile

theorem Dowty1991.stored_profiles_derivable_from_class : (English.Predicates.Verbal.kick.objectEntailments = English.Predicates.Verbal.kick.levinClass.bind fun (x : ArgumentStructure.LevinClass) => x.objectProfile) ∧ (English.Predicates.Verbal.eat.objectEntailments = English.Predicates.Verbal.eat.levinClass.bind fun (x : ArgumentStructure.LevinClass) => x.objectProfile) ∧ (English.Predicates.Verbal.build.objectEntailments = English.Predicates.Verbal.build.levinClass.bind fun (x : ArgumentStructure.LevinClass) => x.objectProfile) ∧ (English.Predicates.Verbal.see.subjectEntailments = English.Predicates.Verbal.see.levinClass.bind fun (x : ArgumentStructure.LevinClass) => x.subjectProfile) ∧ (English.Predicates.Verbal.sweep.subjectEntailments = English.Predicates.Verbal.sweep.levinClass.bind fun (x : ArgumentStructure.LevinClass) => x.subjectProfile) ∧ English.Predicates.Verbal.sweep_instr.subjectEntailments ≠ English.Predicates.Verbal.sweep_instr.levinClass.bind fun (x : ArgumentStructure.LevinClass) => x.subjectProfile

The stored fragment profiles coincide with the Levin-class fallback — effectiveSubjectEntailments/effectiveObjectEntailments would be unchanged without them. sweep_instr is the deliberate exception: its instrument sense overrides the wipe-class (manner-subclass) default.

theorem Dowty1991.arrive_prediction_matches_fragment : decide (ArgumentStructure.PredictsUnaccusative ArgumentStructure.directedMotion.subjectProfile) = English.Predicates.Verbal.arrive.unaccusative

Agreement: arrive prediction matches the fragment annotation.

theorem Dowty1991.run_prediction_matches_fragment : decide (ArgumentStructure.PredictsUnaccusative ArgumentStructure.selfMotion.subjectProfile) = English.Predicates.Verbal.run.unaccusative

Agreement: run prediction matches the fragment annotation.

The paper's explicit per-argument attributions (generated rows in Data.ProtoRoles.Dowty1991, one per stated entailment, with locators) checked against the class-template profiles. Positive checks confirm the templates encode what Dowty states; the divergence theorems record what they deliberately do NOT encode.

def Dowty1991.matchesProfile (d : ProtoRoleDatum) (p : ArgumentStructure.EntailmentProfile) : Bool

Every entailment a data row explicitly attributes agrees with profile p; fields the paper is silent or hedged about are unconstrained.

Equations

• One or more equations did not get rendered due to their size.

theorem Dowty1991.primary_transitives_match_templates : matchesProfile Rows.buildSubject ArgumentStructure.accomplishmentSubjectProfile = true ∧ matchesProfile Rows.writeSubject ArgumentStructure.accomplishmentSubjectProfile = true ∧ matchesProfile Rows.murderSubject ArgumentStructure.accomplishmentSubjectProfile = true ∧ matchesProfile Rows.eatSubject ArgumentStructure.accomplishmentSubjectProfile = true ∧ matchesProfile Rows.washSubject ArgumentStructure.accomplishmentSubjectProfile = true ∧ matchesProfile Rows.murderObject ArgumentStructure.accomplishmentObjectProfile = true ∧ matchesProfile Rows.washObject ArgumentStructure.accomplishmentObjectProfile = true ∧ matchesProfile Rows.writeObject ArgumentStructure.creationObject = true ∧ matchesProfile Rows.eatObject ArgumentStructure.consumptionObject = true

The (35) primary transitive verbs: each stated subject attribution (V+S+C+M, no P-Patient, p. 577) holds of the accomplishment template subject, and each stated object attribution (CoS+CA) holds of the accomplishment/creation/consumption objects.

theorem Dowty1991.hit_class_matches_contact_object : matchesProfile Rows.hitObject ArgumentStructure.contactObject = true ∧ matchesProfile Rows.hitObject hitArg1 = true ∧ matchesProfile Rows.hitInstrument hitArg2 = true

The hit-class attributions ((64 III), §9.3.3): the direct object is non-moving, unchanged, and a non-incremental-theme — exactly consistent with the mannerContact/wipeManner contacted object (CA+St, no CoS) and with this file's §3 hit-class argument profiles.

theorem Dowty1991.exemplars_match_templates : matchesProfile Rows.seeSubject ArgumentStructure.perception.subjectProfile = true ∧ matchesProfile Rows.needSubject ArgumentStructure.desire.subjectProfile = true ∧ Option.map (matchesProfile Rows.needObject) ArgumentStructure.desire.objectProfile = some true ∧ Option.map (matchesProfile Rows.seekObject) ArgumentStructure.desire.objectProfile = some true

The (29)/(30) single-entailment exemplars land in the right templates: see (29b) in the perception subject, need (29e) in the desire subject, need's de dicto object (30e) in the desire object.

theorem Dowty1991.psych_desire_split_justified : matchesProfile Rows.likeSubject ArgumentStructure.psychState.subjectProfile = true ∧ matchesProfile Rows.needSubject ArgumentStructure.desire.subjectProfile = true ∧ matchesProfile Rows.needSubject ArgumentStructure.psychState.subjectProfile = false

The split of the former single state-subject profile is Dowty's own: admire-class experiencers are sentience-entailed ((38): like), so psychState fits them; want-class subjects are NOT ((29e)/p. 573: need entails subject existence "but none of (a)-(d)"), so the sentient profile mis-states them — desire fits instead.

theorem Dowty1991.psych_doublet_tie : ArgumentStructure.AllowsAlternation ArgumentStructure.psychState.subjectProfile ArgumentStructure.stimulusProfile ∧ Option.map (matchesProfile Rows.likeObject) ArgumentStructure.psychState.objectProfile = some true ∧ matchesProfile Rows.pleaseSubject ArgumentStructure.stimulusProfile = true

The (38) doublet tie: the experiencer ({S,IE}) and stimulus ({C,IE}) profiles are Proto-Agent-incomparable with no Proto-Patient difference, so neither outranks — Corollary 1 predicts both lexicalizations (like/please). The stated stimulus attributions (causation, no sentience) hold of the template's stimulus object.

theorem Dowty1991.inchoative_experiencer_matches : matchesProfile Rows.surpriseObjectInchoative expInchoativeProfile = true ∧ ArgumentStructure.OutranksForSubject stimProfile expInchoativeProfile

Croft's inchoative restriction (p. 580): under the inchoative reading the experiencer gains a change of state — the stated attribution matches this file's §2 inchoative experiencer profile, which the stimulus then outranks for subject.

theorem Dowty1991.buy_sell_tie : matchesProfile Rows.buySubject ArgumentStructure.possessionTransfer.subjectProfile = true ∧ matchesProfile Rows.buySeller ArgumentStructure.possessionTransfer.subjectProfile = true ∧ matchesProfile Rows.sellSubject ArgumentStructure.possessionTransfer.subjectProfile = true ∧ matchesProfile Rows.sellBuyer ArgumentStructure.possessionTransfer.subjectProfile = true ∧ ArgumentStructure.AllowsAlternation ArgumentStructure.possessionTransfer.subjectProfile ArgumentStructure.possessionTransfer.subjectProfile

§3.2: buyer and seller are both attributed volition and nothing distinguishes them ("nor are they different in any proto-role entailments", §8.3 p. 579) — the class map gives give-class and obtain-class subjects one shared profile, and Corollary 1 licenses the doublet: a profile never outranks itself.

theorem Dowty1991.build_object_stationary_divergence : matchesProfile Rows.buildObject ArgumentStructure.creationObject = false ∧ matchesProfile Rows.buildObject (have __src := ArgumentStructure.creationObject; { volition := __src.volition, sentience := __src.sentience, causation := __src.causation, movement := __src.movement, independentExistence := __src.independentExistence, changeOfState := __src.changeOfState, incrementalTheme := __src.incrementalTheme, causallyAffected := __src.causallyAffected, stationary := true, dependentExistence := __src.dependentExistence }) = true

Build-object stationary divergence: "all of 28" (pp. 572–573) includes (28d) stationary, but the creation template object (CoS+IT+CA+DE) deliberately omits St — the row disagrees with the template on exactly that field. Recorded as data, not flipped: the class-level template follows the (35)/p. 577 hedge ("(mostly) ... stationary"), which does not commit every creation object to (28d).

theorem Dowty1991.break_object_it_divergence : matchesProfile Rows.breakObject ArgumentStructure.accomplishmentObjectProfile = false ∧ matchesProfile Rows.breakObject (have __src := ArgumentStructure.accomplishmentObjectProfile; { volition := __src.volition, sentience := __src.sentience, causation := __src.causation, movement := __src.movement, independentExistence := __src.independentExistence, changeOfState := __src.changeOfState, incrementalTheme := true, causallyAffected := __src.causallyAffected, stationary := __src.stationary, dependentExistence := __src.dependentExistence }) = true

Break-object incremental-themehood divergence: (64 II) attributes "change of state (and Incremental Themehood)" to the break-class direct object, but the resultChange/accomplishment object template carries no IT (per-verb addition by design). The row disagrees with the template on exactly that field.

theorem Dowty1991.eat_object_de_tension : matchesProfile Rows.eatObject ArgumentStructure.consumptionObject = true ∧ ArgumentStructure.PersistenceLevel.fromPatientProfile ArgumentStructure.consumptionObject = ArgumentStructure.PersistenceLevel.exPersBeginning ∧ ArgumentStructure.PersistenceLevel.fromPatientProfile (have __src := ArgumentStructure.consumptionObject; { volition := __src.volition, sentience := __src.sentience, causation := __src.causation, movement := __src.movement, independentExistence := __src.independentExistence, changeOfState := __src.changeOfState, incrementalTheme := __src.incrementalTheme, causallyAffected := __src.causallyAffected, stationary := __src.stationary, dependentExistence := true }) = ArgumentStructure.PersistenceLevel.exPersEnd

Eat-object dependent-existence tension: (30e)(i) counts destruction — the argument "will not exist after the event" — as dependent existence, which would put DE on eat's object; the paper never states this for eat (the (35) hedge leaves DE open), and the consumption template omits DE because the Grimm bridge disambiguation is load-bearing: adding DE to an incremental theme flips PersistenceLevel.fromPatientProfile from exPersBeginning (consumption) to exPersEnd (creation), misclassifying eat as a creation verb. Recorded, not flipped.

theorem Dowty1991.spray_load_rows_match : matchesProfile Rows.loadTheme sprayLoadTheme = true ∧ matchesProfile Rows.loadLocation sprayLoadLocation = true

The spray/load rows ((64 I), p. 594): both nonsubject arguments are attributed a change of state, matching this file's §3 profiles.