Dependent Case by Agree: Ergative in Shawi [CD26]

[CD26] argue that ergative case in Shawi (Kawapanan; Peru) arises when v Agrees with the subject after the object: the subject serves as the second goal for the v probe, receiving a goal-flag bundle that includes the object's φ-features. The ergative suffix -ri spells out the φ-root of that bundle, and the optional "object agreement on subject" (OAgr-on-S) morpheme spells out the inner φ-features of the same bundle.

The distribution of -ri in Shawi is a strictly descending / ultrastrong PCC pattern (1>2>3). [CD26]'s final generalization (their (10), p. 274) is: ergative appears when the subject is at least as high as the object on the person hierarchy 1>2>3 and both arguments are in the same syntactic domain. The hierarchy half is exactly [Dea24]'s strictlyDescending grammar (SAT:[SPKR], DynINT:[PART]↑), already formalized in Deal2024.lean; the same-domain half is the object's visibility to the v probe (objectVisible).

predictsErgative realizes both factors. Note it follows the [Dea24] mechanism the paper adopts, not the loose prose of (10): a 1st-person object satisfies SAT:[SPKR] and halts the probe, so predictsErgative predicts no ergative at the reflexive 1→1 cell, even though "at least as high as" (1 ≥ 1) would admit it. Shawi data do not document the reflexive 1→1/2→2 cells — Table 4 (p. 285) has no such rows — so this is a prediction of the mechanism, not a tested fact.

This study file does five things:

1. Map the Shawi clause to the Deal-2024 PCC framing: the lower goal (DO) is the object, the higher / second goal (IO) is the subject.
2. Predict the distribution of -ri over the cells of [CD26] Table 4 (p. 285) from the existing strictlyDescending grammar plus the high/low ambiguity for 3rd-person objects (the source of "(✓)" optionality).
3. Ground the predictions in the privative person geometry shared with [Dea24]'s dpBears (and via that, with [PZ18]'s satisfiesProminence and [BR09]'s personSpec).
4. Bridge to the existing person-rank order via Deal2024.sd_off_diagonal_iff_outranks — the Shawi pattern's 1>2>3 hierarchy is exactly the one that fell out of Deal-2024.
5. Counterexample the configurational case rules [CD26] discusses (§1 (1), after [Bak15]; §4.1 (37), a [BS24a]-style rule) by running linglib's own formalized dependent-case algorithm (Syntax.Case.assignCases) on Shawi cells, rather than a local strawman.

Genuinely new machinery — bidirectional Agree (goal flagging), Distributed-Morphology Vocabulary Insertion, Kinyalolo's Constraint — is not introduced here. The study file derives Shawi's empirical table from infrastructure linglib already has, and flags the new machinery as a separate follow-up.

def ClemDeal2024.shawiGrammar : Deal2024.DealGrammar

The v probe in Shawi sits between the object (lower) and the subject (higher). Cyclic Agree (Béjar & Rezac 2009) makes the object the first goal (G1 ≡ DO in Deal-2024 terms) and the subject the second goal (G2 ≡ IO). [CD26] (13) (cyclic expansion), (15)–(19).

Equations

• ClemDeal2024.shawiGrammar = Deal2024.strictlyDescending

def ClemDeal2024.objectVisible : Person → Kawapanan.Shawi.ObjectPosition → Bool

Whether v can interact with the object as G1. 3rd-person low objects sit inside the inner v_cat phase and are invisible to the v probe ([CD26] §3.2, (24), (30)). Every other object — local-person (1/2, including the clusivity cells) and high-positioned 3rd-person — is visible.

Two notes on the wildcard. (i) objectVisible .first .low and objectVisible .second .low return true vacuously: local-person objects never occupy the low position (mustBeHigh is true for them), so the case is structurally inaccessible. (ii) The impersonal .zero and the clusivity cells .firstInclusive/.firstExclusive fall through to true; only the bare .third low cell is invisible, matching the paper's treatment of 3rd-person objects as the persons that show no overt object agreement and may stay low.

Equations

• ClemDeal2024.objectVisible Person.third Kawapanan.Shawi.ObjectPosition.low = false
• ClemDeal2024.objectVisible x✝¹ x✝ = true

def ClemDeal2024.predictsErgative (subj obj : Person) (pos : Kawapanan.Shawi.ObjectPosition) : Bool

Predict whether -ri surfaces on the subject of a transitive clause with the given subject person, object person, and object position.

Two factors must coincide:

1. The object must be visible to v (objectVisible).
2. The probe must successfully Agree with the subject as a second goal — exactly Deal-2024's isLicit for the strictly descending grammar.

Equations

• ClemDeal2024.predictsErgative subj obj pos = (ClemDeal2024.objectVisible obj pos && Deal2024.isLicit ClemDeal2024.shawiGrammar subj obj)

theorem ClemDeal2024.erg_with_1p_obj_blocked (subj : Person) (pos : Kawapanan.Shawi.ObjectPosition) : predictsErgative subj Person.first pos = false

1P object always satisfies SAT:[SPKR], halting the probe before it reaches the subject — ergative is impossible regardless of the subject's features or the object's position. [CD26] (7a–b), (14a–b), §3 derivation in (15).

theorem ClemDeal2024.erg_with_2p_obj_iff_subj_part (subj : Person) : predictsErgative subj Person.second Kawapanan.Shawi.ObjectPosition.high = Deal2024.dpBears subj Deal2024.PersonFeature.part

2P high object: lacks [SPKR] (no SAT halt) but bears [PART], triggering dynamic narrowing. The subject is then visible as G2 only if it bears [PART] — i.e., is itself local-person. [CD26] §3.2, (16).

theorem ClemDeal2024.erg_with_3p_obj_high_always (subj : Person) : predictsErgative subj Person.third Kawapanan.Shawi.ObjectPosition.high = true

3P high object: lacks [SPKR] and lacks [PART], so the probe neither halts nor narrows. The subject is visible as G2 regardless of its features. [CD26] (8), (22).

theorem ClemDeal2024.erg_with_3p_obj_low_blocked (subj : Person) : predictsErgative subj Person.third Kawapanan.Shawi.ObjectPosition.low = false

3P low object: invisible to v. The probe finds only the subject as G1; no goal-flag bundle reaches the subject. [CD26] (24).

theorem ClemDeal2024.erg_off_diagonal_iff_subj_outranks_obj (subj obj : Person) (h_neq : Minimalist.decomposePerson subj ≠ Minimalist.decomposePerson obj) : predictsErgative subj obj Kawapanan.Shawi.ObjectPosition.high = true ↔ subj.prominence > obj.prominence

For non-diagonal (subject ≠ object) configurations with a visible object, ergative on the subject coincides with the subject outranking the object on the 1>2>3 person hierarchy. Inherits [Dea24]'s sd_off_diagonal_iff_outranks — Shawi's hierarchy effect is exactly the one that fell out of Deal-2024 on independent grounds.

theorem ClemDeal2024.table4 : predictsErgative Person.first Person.second Kawapanan.Shawi.ObjectPosition.high = true ∧ predictsErgative Person.first Person.third Kawapanan.Shawi.ObjectPosition.high = true ∧ predictsErgative Person.first Person.third Kawapanan.Shawi.ObjectPosition.low = false ∧ predictsErgative Person.second Person.first Kawapanan.Shawi.ObjectPosition.high = false ∧ predictsErgative Person.second Person.third Kawapanan.Shawi.ObjectPosition.high = true ∧ predictsErgative Person.second Person.third Kawapanan.Shawi.ObjectPosition.low = false ∧ predictsErgative Person.third Person.first Kawapanan.Shawi.ObjectPosition.high = false ∧ predictsErgative Person.third Person.second Kawapanan.Shawi.ObjectPosition.high = false ∧ predictsErgative Person.third Person.third Kawapanan.Shawi.ObjectPosition.high = true ∧ predictsErgative Person.third Person.third Kawapanan.Shawi.ObjectPosition.low = false

[CD26] Table 4 (p. 285) — the full ergative distribution by person of subject and object, here with the high/low split of the 3rd-person-object rows giving 10 cells. Each reduces by rfl from predictsErgative. Table 4 lists 7 (subject, object) person pairs; it has no reflexive 1→1 or 2→2 rows (see the header note on the 1→1 mechanism prediction).

theorem ClemDeal2024.optionality_only_with_3p_object (subj obj : Person) : predictsErgative subj obj Kawapanan.Shawi.ObjectPosition.high ≠ predictsErgative subj obj Kawapanan.Shawi.ObjectPosition.low → obj = Person.third

Optionality is exactly the (✓) cells: 1→3, 2→3, 3→3.

theorem ClemDeal2024.local_object_position_irrelevant (subj obj : Person) (h : obj.IsSAP) (pos : Kawapanan.Shawi.ObjectPosition) : predictsErgative subj obj pos = Deal2024.isLicit shawiGrammar subj obj

For any local-person object, the position parameter is irrelevant to the prediction: ergative depends only on isLicit shawiGrammar. Reason: objectVisible is true for any non-3P object regardless of position.

theorem ClemDeal2024.local_object_must_be_high (p : Person) (h : p.IsSAP) : Kawapanan.Shawi.mustBeHigh p = true

Bookkeeping: for any local-person object, the Fragment-level mustBeHigh constraint forces the high position, which is exactly what the probe-visibility analysis presupposes.

def ClemDeal2024.shawiDomain (_subj _obj : Person) : List Syntax.Case.NPInDomain

A Shawi monotransitive mapped to a dependent-case Spell-Out domain: the subject c-commands the object (earlier = structurally higher), and neither bears lexical case. Person-blind, exactly as a configurational case rule is.

Equations

• ClemDeal2024.shawiDomain _subj _obj = [{ label := "subj", lexicalCase := none }, { label := "obj", lexicalCase := none }]

def ClemDeal2024.configErg (subj obj : Person) : Bool

Baker's configurational rule for ergative ([Bak15]; [CD26] (1)): "if there are two distinct NPs in the same spell-out domain such that NP1 c-commands NP2, then value the case feature of NP1 as ergative unless NP2 has already been marked for case." Rather than restate this as a local strawman, we run linglib's own dependent-case algorithm — Syntax.Case.assignCases .ergative, which values the higher of two caseless NPs ergative — over the Shawi domain.

Equations

• ClemDeal2024.configErg subj obj = (Syntax.Case.getCaseOf "subj" (Syntax.Case.assignCases Syntax.Case.CaseLanguageType.ergative (ClemDeal2024.shawiDomain subj obj)) == some Case.erg)

theorem ClemDeal2024.configErg_person_blind (subj obj : Person) : configErg subj obj = true

The configurational rule is person-blind: assignCases .ergative values every transitive subject ergative, regardless of person.

theorem ClemDeal2024.configurational_rule_overgenerates_2_to_1 : configErg Person.second Person.first = true ∧ predictsErgative Person.second Person.first Kawapanan.Shawi.ObjectPosition.high = false

It therefore overgenerates exactly where Shawi bans -ri because the object outranks the subject. 2→1: [CD26] (7a). This is the paper's own argument against rule (1).

theorem ClemDeal2024.configurational_rule_overgenerates_3_to_1 : configErg Person.third Person.first = true ∧ predictsErgative Person.third Person.first Kawapanan.Shawi.ObjectPosition.high = false

And 3→1: [CD26] (7b).

def ClemDeal2024.augmentedConfigErg (subj obj : Person) : Bool

The augmented configurational rule [CD26] (37) writes the person hierarchy into rule (1) itself: NP1 gets ergative when it c-commands a caseless NP2 and "NP1 is at least as high as NP2 on the person hierarchy 1>2>3" — note ≥ (the paper's "at least as high as"), so it fires on the person diagonal too. This is [CD26]'s rendering of an option explored by [BS24a]; linglib has no formalized person-augmented configurational rule, consistent with the paper's third concern (§4.1) that such rules face no principled limit on the hierarchies they may stipulate.

Equations

• ClemDeal2024.augmentedConfigErg subj obj = decide (subj.prominence ≥ obj.prominence)

theorem ClemDeal2024.augmented_config_matches_on_documented_cells : augmentedConfigErg Person.first Person.second = predictsErgative Person.first Person.second Kawapanan.Shawi.ObjectPosition.high ∧ augmentedConfigErg Person.first Person.third = predictsErgative Person.first Person.third Kawapanan.Shawi.ObjectPosition.high ∧ augmentedConfigErg Person.second Person.first = predictsErgative Person.second Person.first Kawapanan.Shawi.ObjectPosition.high ∧ augmentedConfigErg Person.second Person.third = predictsErgative Person.second Person.third Kawapanan.Shawi.ObjectPosition.high ∧ augmentedConfigErg Person.third Person.first = predictsErgative Person.third Person.first Kawapanan.Shawi.ObjectPosition.high ∧ augmentedConfigErg Person.third Person.second = predictsErgative Person.third Person.second Kawapanan.Shawi.ObjectPosition.high ∧ augmentedConfigErg Person.third Person.third = predictsErgative Person.third Person.third Kawapanan.Shawi.ObjectPosition.high

[CD26] do not refute (37) empirically — they grant it "is simple to state" (§4.1), and it is descriptively adequate: on every documented (high-object) cell of Table 4 it predicts exactly the Agree analysis.

theorem ClemDeal2024.augmented_config_blind_to_object_position : augmentedConfigErg Person.third Person.third = true ∧ predictsErgative Person.third Person.third Kawapanan.Shawi.ObjectPosition.high = true ∧ predictsErgative Person.third Person.third Kawapanan.Shawi.ObjectPosition.low = false

The paper's objections to (37) are architectural, not empirical (§4.1): a person-augmented configurational rule (i) severs the case↔agreement (OAgr-on-S) connection, (ii) cannot unify the case-split person hierarchy with the identical hierarchy seen in agreement (e.g. Spanish ditransitives), and (iii) places no principled limit on which hierarchies a case rule may stipulate. linglib's internal counterpart is expressive: being a rule over person alone, (37) is blind to object position, so it cannot produce the position-conditioned optionality of 3rd-person objects (Table 4 rows b, d, g; (8), (23)). It predicts ergative for 3→3 (3 ≥ 3) — the high-object reading, (22) — but has no way to derive the low-object, no-ergative reading, (24), that the Agree analysis gets from object visibility.

theorem ClemDeal2024.outranking_subj_blocks_postverbal_obj (subj obj : Person) (h_neq : Minimalist.decomposePerson subj ≠ Minimalist.decomposePerson obj) (h_rank : subj.prominence > obj.prominence) : Kawapanan.Shawi.objectSyntaxLicit (predictsErgative subj obj Kawapanan.Shawi.ObjectPosition.high) Kawapanan.Shawi.ObjectSyntax.overtPostverbal = false

[CD26] (9), (21): when the subject bears -ri, the object cannot remain overt-postverbal. This combines the Fragment-level objectSyntaxLicit with the predictedness of ergative — a high-object subject that outranks the object forces the object out of postverbal position (to OSV or pro-drop).

theorem ClemDeal2024.fronted_and_prodropped_always_licit (subj obj : Person) (pos : Kawapanan.Shawi.ObjectPosition) : Kawapanan.Shawi.objectSyntaxLicit (predictsErgative subj obj pos) Kawapanan.Shawi.ObjectSyntax.overtFronted = true ∧ Kawapanan.Shawi.objectSyntaxLicit (predictsErgative subj obj pos) Kawapanan.Shawi.ObjectSyntax.proDropped = true

The fronted (OSV) and pro-drop options remain licit regardless of whether the subject bears -ri.

"Object agreement on subject" (OAgr-on-S) can spell out the inner φ-features of the goal-flag bundle that -ri exposes; if no such bundle is present (i.e. no ergative), there is nothing for the OAgr-on-S morpheme to attach to. [CD26]'s empirical generalization (§2, p. 274, with footnote 7, elaborated in §3.3) is that OAgr-on-S obtains only if the subject is ergative — an only if, not an iff: an ergative subject may appear with or without OAgr-on-S.

We deliberately do not ship a separate oagrOnSAvailable predicate. Its availability coincides with predictsErgative only by construction here, so a definition oagrOnSAvailable := predictsErgative plus a "bridge" theorem would merely rename predictsErgative and re-export its theorems. The substantive content — the only-if/not-iff asymmetry — is not derivable without the goal-flagging machinery flagged in §11; until then it is an empirical generalization about predictsErgative, not a separate object.

theorem ClemDeal2024.shawi_licit_count : Deal2024.licitCount shawiGrammar = 5

Across the 9 (subject, object) cells, the strictly-descending grammar licenses 5 — exactly the 5 cells where ergative is licit in Shawi (rows a, b, d, g of Table 4, with row b/d/g being the high-object subcases). Derived from Deal2024.sd_licit_count.

theorem ClemDeal2024.high_object_matches_deal2024 (subj obj : Person) : predictsErgative subj obj Kawapanan.Shawi.ObjectPosition.high = Deal2024.isLicit shawiGrammar subj obj

Among the high-object cells, predicted ergative coincides with isLicit on the strictly-descending grammar — Shawi inherits the Deal-2024 typology wholesale once we condition on "object visible to v".

Follow-ups

[clem-deal-2024]'s analysis crucially depends on three pieces of machinery that linglib does not yet have:

1. Bidirectional Agree (goal flagging) — the probe-to-goal direction of feature transfer. Without this, the claim that -ri is the object's φ-features (rather than a primitive [ERG]) cannot be stated structurally. Syntax/Minimalism/Agree.lean currently models only valuation (goal→probe).
2. Distributed Morphology / Vocabulary Insertion — for the VI rule ri ↔ φ / ___ [φ,D] ([CD26] (34)) and Kinyalolo's Constraint impoverishment ((35)). The Core/Lexical/MorphRule.lean skeleton is Bybee-flavored and does not yet capture context-sensitive realization.
3. Feature provenance — distinguishing "native" features on a goal from features deposited there by Agree ([CD26] fn. 23). Required to state the ergative VI without overgenerating.

Once (1)–(3) are in place, the OAgr-on-S generalization noted in §9 — currently an empirical only if about predictsErgative — can be derived as a theorem, including the only-if/not-iff asymmetry.