Semantics of Case ([Gri11]) #
Study file for [Gri11]: canonical verbs mapped through the agentivity
lattice (Semantics/ArgumentStructure/) to case regions, a
lattice-derived account of differential object marking checked against the
[Ais03] DOM profiles, and the paper's engagement with [Dow91]'s
Argument Selection Principle. The DOM substrate reconstructs the
referential-property treatment the paper attributes to [Gri05a].
Main results #
subject_region_iff_causation/kill_subject_range: for every [Dow91] profile the subject is NOM/ERG iff it entails causation, and every node above kill's minimal agent qualifies — both from the substrate's interval characterization of the regions.russian_dom_matches_lattice/spanish_dom_within_lattice/animate_dom_object_is_dative_node: the lattice predicts DOM for exactly {animate, human}, and the animate DOM object is literally the dative node.domPredicted_monotone,latticeDOM_isMonotone,latticeDOM_matches_aissen_type2: the lattice derives [Ais03]'s monotonicity universal and reproduces OT Type 2.genitive_requires_entailment_free: only the entailment-free object profile reaches the genitive node — the §5.2 intensionality restriction as a theorem over all [Dow91] profiles.wellFormedPair_not_preserved,kiss_outranking_from_dominance,arrive_cross_theory: what the lattice projection of Dowty's profiles loses, and the ASP re-derived from dominance (outranks_of_lattice_dominance) rather than checked per verb.lattice_diverges_from_dependent_case: the semantic-case vs structural-case fault line on animate objects, made explicit against the dependent-case pipeline inStudies/Aissen2003.lean.
The canonical verb chain (§2.2, p.523–524) #
Subjects populating a maximal chain: positional verbs at ⊥, then know/see (sentience), discover (+instigation), look at (+motion), assassinate (+volition, = ⊤).
sit/stand subject: ⊥, p.523.
Equations
Instances For
know/see subject: sentience only, p.523–524.
Equations
- Grimm2011.knowAgentivity = ArgumentStructure.Agentivity.mk false true false false
Instances For
discover subject: sentience + instigation, p.524.
Equations
- Grimm2011.discoverAgentivity = ArgumentStructure.Agentivity.mk false true true false
Instances For
look at subject: sentience + instigation + motion, p.524.
Equations
- Grimm2011.lookAtAgentivity = ArgumentStructure.Agentivity.mk false true true true
Instances For
assassinate subject: all four features, p.524.
Equations
Instances For
Each verb adds one feature: a strict chain from ⊥ to ⊤.
All chain positions satisfy volition → sentience.
The perception template's subject projects to exactly the know/see node — the [Dow91] bridge lands on the chain.
Tsunoda's transitivity hierarchy (§3, example 8) #
Resultative Effective Action (kill, break) >> Contact (shoot, hit) >> Pursuit (search, seek): the lower the patient's persistence, the more prototypically transitive the verb.
Class I/II patients are in the transitivity region; Class III (pursuit) is outside.
Patient nodes are ordered III ≤ I ≤ II: lower persistence, more affected.
Fig. 5 placements: the shared agent (Ia/IIa) is NOM/ERG; Class I/II patients are ACC/ABS; the pursuit patient falls outside the core object region.
Case regions for canonical verb classes (§4) #
Levin-class templates mapped through the lattice: instigation puts subjects
in NOM/ERG; ⊥ agentivity plus change-from-the-beginning persistence puts
objects in ACC/ABS. Contact objects use the project-canonical
no-entailed-change profile (contactObject_persistence), which exits
ACC/ABS — [Gri11]'s own Fig. 5 keeps them inside at quPersBeginning.
kick: subject → NOM/ERG; object → oblique under the canonical contact profile — a flagged mis-prediction (English gives contact objects plain ACC) inherited from the Fig. 5 deviation.
kick/eat objects are in the transitivity region; build objects (not existing at event start, p.529–530) are not.
build: subject → NOM/ERG; the created object (exPersEnd) → oblique.
eat: subject → NOM/ERG; the consumed object → ACC/ABS, with destroyed objects.
buy/sell: subject → NOM/ERG. Buyer and seller share one profile — the [Dow91] §3.2 alternation tie.
see: sentience without instigation → oblique. Experiencer subjects
land outside the dative region — fromSubjectProfile fixes total
persistence, the dative region needs quPersBeginning.
run: volition + sentience + motion without instigation → oblique, matching unergative behaviour in split-S systems.
arrive: motion only → oblique.
die: the sole argument, read as patient, → ACC/ABS; ergative readout ABS — the unaccusative pattern.
Instigation exactly divides the subjects: for every [Dow91]
profile, the subject lands in NOM/ERG iff it entails causation —
the per-verb facts above are instances. From the interval
characterization and Agentivity.le_iff, not enumeration.
Acceptable-argument ranges (§2.3, p.528) #
Argument acceptability is closure-based: any node above a verb's minimal agent requirement qualifies. Since kill's agent requires only instigation, its subject range at total persistence is the whole NOM/ERG interval — from natural forces to intentional agents.
Everything above kill's minimal agent is NOM/ERG: upward closure plus
the interval characterization toCaseRegion_eq_nomErg_iff.
The range's endpoints, derived: electricity (instigation only) and the assassin (⊤) are both acceptable kill-subjects.
Accusative and ergative alignment (§4, Fig. 6) #
The poles read out to the core markers of both systems: NOM/ACC in an accusative alignment, ERG/ABS in an ergative one.
Differential object marking (§4, p.534; [Gri05a]) #
[Gri11] p.534: "it is a combination of verbal and nominal properties which trigger DOM", deferring the referential side to [Gri05a]. This section supplies one lattice encoding — a formaliser construction, not the paper's: animacy contributes a baseline agentive position, the verb the object's persistence, and DOM is predicted iff the node is in the transitivity region but outside ACC/ABS. Grimm himself (following [Ais03]) keeps referential prominence on a separate axis.
Ceiling agentivity of a referent type: human ↦ {V, S}, animate ↦ {S}, inanimate ↦ ⊥. Instigation and motion are event-bound, so never contributed; denying animates volition keeps the hierarchy strict — a modelling choice, not Grimm's.
Equations
- Grimm2011.animacyToAgentivity Features.Prominence.AnimacyLevel.human = ArgumentStructure.Agentivity.mk true true false false
- Grimm2011.animacyToAgentivity Features.Prominence.AnimacyLevel.animate = ArgumentStructure.Agentivity.mk false true false false
- Grimm2011.animacyToAgentivity Features.Prominence.AnimacyLevel.inanimate = ⊥
Instances For
All animacy-derived nodes satisfy volition → sentience.
Higher animacy → higher agentivity.
Animacy-derived agentivity × the verb's object persistence.
Equations
- Grimm2011.objectNodeWithAnimacy a p = { agentivity := Grimm2011.animacyToAgentivity a, persistence := p }
Instances For
The key non-circular derivation: at quPersBeginning, toCaseRegion —
defined for general case theory, not DOM — separates inanimate objects
(ACC/ABS) from animate/human ones (dative, Fig. 7).
The animate DOM object at quPersBeginning IS sentientNonInstigator
— one lattice point shared with recipients and experiencers (Fig. 7).
On this account DOM marking is dative marking (Spanish a).
DOM is predicted: the object node is in the transitivity region but outside ACC/ABS.
Equations
Instances For
Equations
- Grimm2011.instDecidableDomPredictedByLattice a p = id inferInstance
DOM for animate/human but not inanimate objects, for both canonical
(quPersBeginning) and resultative (exPersBeginning) transitives.
Creation objects (exPersEnd) are outside the transitivity region at
every animacy level: DOM is structurally inapplicable.
[Ais03]'s monotonicity universal, derived: DOM prediction is
monotone in animacy at every persistence level — from
animacyToAgentivity_monotone and the region geometry, not
stipulation.
Checking attested DOM languages #
Russian (animate accusative) marks exactly the lattice-predicted cells.
Spanish (human a) is a proper subset of the lattice prediction: it
under-marks the predicted animate cell.
Hindi agrees on the animacy boundary: inanimate never marked, animate/human marked at some definiteness level.
The lattice's DOM prediction at a fixed persistence level, as a [Jus24]-style differential marking profile.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Lattice-derived profiles satisfy [Ais03]'s monotonicity
universal, structurally via isMonotoneP_of.
At quPersBeginning the lattice-derived profile coincides with
[Ais03]'s OT Type 2 (Hu + An; Aissen2003.anim_type_hu_an) —
two frameworks, independent premises, same prediction.
Limitation: total-persistence objects #
For totalPersistence objects, toCaseRegion yields oblique even at ⊥
agentivity, so DOM is over-predicted at every animacy level — the predicate
is informative only for the transitivity region's core.
DOM over-predicted at every animacy level at total persistence.
The verb-class effect on DOM (p.534; [vH08]) #
Grimm cites [vH08]'s finding that Spanish DOM regularized at different rates for matar 'kill', ver 'see', poner 'put'. This section operationalizes Grimm's subject–object opposition gauge through subject regions: NOM/ERG subjects give maximal contrast (DOM redundant, free to regularize); oblique-region subjects leave DOM discriminating. (Von Heusinger's own classification keys on object animacy — the other side of the same opposition.)
Accomplishment (kill-type) verbs: NOM/ERG subject and ACC/ABS object —
maximal contrast (matar). Perception verbs lose the contrast on the
subject side (see_case_region: oblique, ver); creation verbs have
NOM/ERG subjects (build_case_regions) but no DOM question at all
(creation_dom_inapplicable).
The Russian genitive/accusative alternation (§5.2, Fig. 8) #
Objects of intensional verbs (want, seek, await; p.539–541) take accusative under the specific reading and genitive under the non-specific one — a case contrast between two lattice nodes.
The specific reading (object exists, exPersBeginning) sits in
ACC/ABS; the non-specific reading (existence not entailed) is the
lattice bottom — Grimm's governed genitive at "the lowest node of the
lattice" (p.540).
Only the entailment-free P-Patient profile reaches the genitive node: the lattice form of "the alternation is limited to intensional verbs" (p.541), for every [Dow91] profile.
Limitation: [Dow91] (30e) codes de-dicto objects (needs a car)
with dependent existence, which the bridge reads as destruction — the
desire-class object lands at exPersBeginning, not at Grimm's ⊥
placement for want. Dowty's features cannot separate "never entailed
to exist" from "ceases to exist".
Grimm vs [Dow91] #
§2.1 recasts Dowty's ten entailments as four agentivity features plus
persistence; the projection kernel is
Agentivity.fromEntailmentProfile_eq_iff. Below: what the recast loses
(Dowty's pairing constraints) and what it fixes (the arrive anomaly) —
cross-theory checks absorbed from Studies/Dowty1991.lean, since the
comparison belongs to the later paper.
Dowty's WellFormedPair is invisible to the projection: a {C} and a
{C, IE} subject project to the same node (IE is dropped,
Agentivity.fromEntailmentProfile_eq_iff), yet against a {CoS}
object only the first satisfies the IE→DE pairing constraint.
The arrive anomaly resolved: Table 1, the priority ASP, and the lattice all say unaccusative; only flat counting diverges.
kick: ASP outranking and the lattice's subject region agree — NOM subject.
die: priority ASP, flat counting, and the lattice agree on unaccusativity.
kiss: the subject strictly dominates the object on the lattice —
Dowty1991.kiss_asymmetry_is_volition as order.
Subject selection for kiss from lattice dominance alone
(outranks_of_lattice_dominance): the ASP mechanized, re-deriving
Dowty1991.kiss_subject_outranks without counting or decide.
Dowty's count comparison follows from lattice dominance via
featureCount_monotone and pAgentScore_decomposition.
Grimm vs dependent case #
Studies/Aissen2003.lean Part II assigns abstract ACC to every transitive
object, with DOM a realization filter; Grimm assigns the case category by
lattice position. On Grimm's line Spanish a is dative, not flagged ACC —
the fault line between structural and semantic case.
Animate definite object of a canonical transitive: dependent case says ACC, the lattice says DAT — different categories, not different spell-outs.