Docs / Studies / Studies.ClemDeal2026

Dependent Case by Agree: Ergative in Shawi

Clem, E. & Deal, A. R. (2026). Linguistic Inquiry, 57(2), 267–313

A probe flags each goal with all the features its head holds, so a second goal receives the first goal's φ-features, and a dependent case is the vocabulary item spelling out that φ on a nominal. The Shawi subject surfaces with -ri and the verb's object agreement when the strictly descending probe on v reaches it after a high object, and the same machinery predicts the rows of four languages. Since a hierarchy is a probe's run over a feature geometry, no probe puts third person above second, and no probe on v yields a reverse hierarchy, nor one on T a forward hierarchy.

Implementation notes

Items (36a) and (36c) tie in size on a first person, who bears both [SPKR] and [PART]; the order of objectItems resolves the tie, and the inclusive items, absent from (36), are not encoded.
objectAgreement states v after Kinyalolo's Constraint (35) has removed the subject's features; the impoverishment itself is not encoded.
The rows of (42), (45), (46) and (49) are read with the object positions of Shawi, as note 38 assumes for Shiwilu.
The paper's v is Cat.Voice and its categorizing v is Cat.v, the labels note 19 offers.

Errata

Generalizations (5) and (10) and table 7 give dependent case when the subject is at least as high as the object, but (15) rules out every first-person object, so the mechanism gives no ergative at 1→1 and does give it at 2→2. The tables have no reflexive rows.
(6b) prints the subject's object agreement as -nke(ma), but (36d) spells out [PART, AUG, v] as a whole and none of the routes to its absence in §3.3 deletes [AUG] alone, so bare -nke on the subject of a 2PL object is not derived.

TODO

positions stipulates that fronting and pro-drop require the high position; clause .low can still front, and a phase condition on movement should exclude it.

References

Clem & Deal 2026, Dependent Case by Agree: Ergative in Shawi
Deal 2024, Interaction, Satisfaction, and the PCC formalized
Deal 2025a, Current models of Agree formalized
Deal 2010, Ergative case and the transitive subject: A view from Nez Perce
Clem 2019, Amahuaca ergative as agreement with multiple heads
Béjar & Rezac 2009, Cyclic Agree formalized
Baker 2015, Case: Its Principles and Its Parameters formalized
Baker & Vinokurova 2010, Two Modalities of Case Assignment: Case in Sakha formalized
Bárány & Sheehan 2024, Challenges for dependent case theory
Valenzuela 2011, Argument encoding and pragmatic marking of the transitive subject in Shiwilu (Kawapanan)
Maslova 2003, A Grammar of Kolyma Yukaghir
van Urk 2015, A Uniform Syntax for Phrasal Movement: A Case Study of Dinka Bor
Bourdeau 2015, Ergativity in Shawi (Chayahuita)
Barraza de García 2005, El sistema verbal en la lengua shawi
Harley & Ritter 2002, Person and number in pronouns: A feature-geometric analysis formalized

Goal flagging

Feature transfer under Agree is transitive ((31)): a probe flags its first goal with its category feature alone and, if it goes on to a second, flags that with the first goal's φ-features as well.

theorem flagged_pair L137–145

(g : Deal2024.Grammar) (l : Locus) (nom : Baker2015.Arg → Nominal) (a b : Baker2015.Arg) :

flagged g l nom [a, b] = (a, {Feature.head l}) :: if Deal2024.Licit g { high := (nom b).1, low := (nom a).1 } then [(b, {Feature.head l} ∪ phi (nom a))] else []

Built from , , ,

The position of the object

Spell-out

The Shawi subject, as the probe's second goal, is spelled out as -ri followed by the object agreement of the verb ((33), (34), (36)), which is the sense in which ergative case and object agreement on the subject are both features of the object realized on the subject.

theorem forms_of_licit L416–422

{s o : Nominal} (h : Deal2024.Licit Deal2024.strictlyDescending { high := s.1, low := o.1 }) :

shawi.forms s o ObjectPosition.high Baker2015.Arg.subject = Kawapanan.Shawi.ergativeMarker :: objectAgreement o

Built from , , ,

L431–434

L437–462

Table 4

The rows

Every row of Shawi, Shiwilu, Sakha and Kolyma Yukaghir but the bare -nke of (6b) is judged as the run of the language's probe and its vocabulary predict: a sentence is accepted exactly when some position open to its object spells out what it prints, on the argument it prints it on.

theorem rows_predicted L559–564

Datum.Checks stimulus? predicted (Examples.all.erase Examples.ex_6b_nke) 42

Built from ,

L567–570

Configurational rules (§1, §4.1)

Agree cannot do what (37) can: no probe yields the rule over a ranking with third person above second, since a probe that gets past a second person gets past a third (§4.1).

theorem not_forall_iff_rule37 L611–625

(g : Deal2024.Grammar) {rank : Person → ℕ} (h : rank Person.second < rank Person.third) (n : Number) :

¬∀ (s o : Person), s.prominence ≠ o.prominence → (DependentCase g Locus.v (nominals (s, n) (o, n)) (goals Locus.v ObjectPosition.high) ↔ Option.map (fun (x : Case × Case.Mechanism) => x.1) (rule37 rank (s, n) (o, n) Baker2015.Arg.subject).value? = some Case.erg)

Built from , , ,

The typology of global case splits (§5)

L629–632

L635–644

L652–662

The Directionality Correlation (54). Whatever a hierarchy effect t that bans a first-person direct object under a third-person indirect object, no probe on v yields its reverse and no probe on T its forward form, since a first goal's person is what bleeds a probe: only the object's can bleed ergative and only the subject's accusative. This excludes ergative where the object outranks the subject, as reported for Fore and Dalabon (fns. 29, 44).

theorem directionality L670–683