Toba Batak Extraction Restriction @cite{erlewine-2018}

@cite{elkins-torrence-brown-2026}

Empirical data on the extraction restriction in Toba Batak (only the pivot can undergo Ā-movement) plus @cite{erlewine-2018}'s Minimalist analysis (predicate fronting + nominal licensing).

Key Empirical Findings

1. Pivot-only extraction: Only the pivot argument can be Ā-extracted.
2. Voice determines pivot: Actor Voice → agent is pivot; Object Voice → patient is pivot.
3. Voice symmetry: AV makes the agent extractable but not the patient; OV makes the patient extractable but not the agent.
4. Long-distance extraction: The extracted element must be the pivot of the most deeply embedded clause.

The Analysis

The extraction restriction follows from the interaction of probing and nominal licensing:

1. Predicate fronting: C bears [PROBE:FOC], attracting the closest [+FOC] constituent — normally the vP — to Spec,CP, deriving V-initial word order. The subject/pivot is stranded in Spec,TP after vP fronts.

2. Nominal licensing: T bears [PROBE:D], which Case-licenses the subject DP in Spec,TP. If a non-subject DP were attracted to Spec,CP by [PROBE:FOC], it would end up in a position with no available Case licensor — the derivation crashes. Therefore only the pivot (already Case-licensed by T) can be Ā-extracted.

3. Non-DP extraction is unrestricted: Since the restriction is about nominal licensing, non-DP constituents (adverbs, PPs) can freely front to Spec,CP regardless of voice.

The descriptive generalization: extraction of a DP argument is grammatical iff it is the pivot for the given voice.

Cross-Linguistic Context

Toba Batak's extraction restriction contrasts with:

• Tagalog/Seediq: Philippine-style voice/Case determines extraction via agreement morphology — a voice-as-case system. @cite{erlewine-2018} argues TB is NOT this type.
• Mam (Mayan): Extraction is unrestricted, but oblique extraction triggers a dedicated morpheme as an Agree reflex on Voice⁰ (see @cite{elkins-torrence-brown-2026}).

Connection to Mam

Both TB and Mam involve successive-cyclic movement leaving morphological traces at clause boundaries. The shared abstraction is CyclicChain from Position.lean:

• Mam: Each intermediate Voice⁰ Agrees [+oblique], spelling out as a dedicated morpheme. The chain entries correspond to feature-valuation events.
• TB: Each intermediate C⁰ shows extraction voice morphology, reflecting the passage of the wh-element through Spec,CP.

theorem Phenomena.FillerGap.Studies.Erlewine2018.extraction_data_count : Fragments.TobaBatak.extractionData.length = 8

theorem Phenomena.FillerGap.Studies.Erlewine2018.av_agent_grammatical : Fragments.TobaBatak.avAgentExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.grammatical ∧ Fragments.TobaBatak.avAgentExtraction.extractsPivot = true

AV + agent (pivot): grammatical.

theorem Phenomena.FillerGap.Studies.Erlewine2018.av_patient_ungrammatical : Fragments.TobaBatak.avPatientExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.ungrammatical ∧ Fragments.TobaBatak.avPatientExtraction.extractsPivot = false

AV + patient (non-pivot): ungrammatical.

theorem Phenomena.FillerGap.Studies.Erlewine2018.av_oblique_ungrammatical : Fragments.TobaBatak.avObliqueExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.ungrammatical ∧ Fragments.TobaBatak.avObliqueExtraction.extractsPivot = false

AV + oblique (non-pivot): ungrammatical.

theorem Phenomena.FillerGap.Studies.Erlewine2018.ov_patient_grammatical : Fragments.TobaBatak.ovPatientExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.grammatical ∧ Fragments.TobaBatak.ovPatientExtraction.extractsPivot = true

OV + patient (pivot): grammatical.

theorem Phenomena.FillerGap.Studies.Erlewine2018.ov_agent_ungrammatical : Fragments.TobaBatak.ovAgentExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.ungrammatical ∧ Fragments.TobaBatak.ovAgentExtraction.extractsPivot = false

OV + agent (non-pivot): ungrammatical.

theorem Phenomena.FillerGap.Studies.Erlewine2018.ov_oblique_ungrammatical : Fragments.TobaBatak.ovObliqueExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.ungrammatical ∧ Fragments.TobaBatak.ovObliqueExtraction.extractsPivot = false

OV + oblique (non-pivot): ungrammatical.

theorem Phenomena.FillerGap.Studies.Erlewine2018.av_adjunct_grammatical : Fragments.TobaBatak.avAdjunctExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.grammatical

AV + adjunct: grammatical (adjuncts don't need Case).

theorem Phenomena.FillerGap.Studies.Erlewine2018.ov_adjunct_grammatical : Fragments.TobaBatak.ovAdjunctExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.grammatical

OV + adjunct: grammatical (adjuncts don't need Case).

theorem Phenomena.FillerGap.Studies.Erlewine2018.dp_extraction_iff_pivot : (Fragments.TobaBatak.extractionData.all fun (d : Fragments.TobaBatak.ExtractionDatum) => match d.extracted with | Typology.Extractee.dpArg a => d.extractsPivot == (d.judgment == Fragments.TobaBatak.ExtractionJudgment.grammatical) | Typology.Extractee.adjunct => true) = true

For DP arguments: extraction is grammatical iff the extracted element is the pivot for the given voice.

theorem Phenomena.FillerGap.Studies.Erlewine2018.adjunct_always_grammatical : (Fragments.TobaBatak.extractionData.all fun (d : Fragments.TobaBatak.ExtractionDatum) => match d.extracted with | Typology.Extractee.adjunct => d.judgment == Fragments.TobaBatak.ExtractionJudgment.grammatical | Typology.Extractee.dpArg a => true) = true

For non-DP adjuncts: extraction is always grammatical regardless of voice, because adjuncts don't need Case licensing.

theorem Phenomena.FillerGap.Studies.Erlewine2018.voice_symmetry : Fragments.TobaBatak.avAgentExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.grammatical ∧ Fragments.TobaBatak.ovPatientExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.grammatical ∧ Fragments.TobaBatak.avPatientExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.ungrammatical ∧ Fragments.TobaBatak.ovAgentExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.ungrammatical

Voice symmetry: AV and OV are mirror images — each makes exactly one argument extractable (the pivot) and blocks all others.

theorem Phenomena.FillerGap.Studies.Erlewine2018.tb_strategy : Fragments.TobaBatak.tbExtractionProfile.strategy = Typology.ExtractionMarkingStrategy.voiceAlternation

TB uses structural restriction, not dedicated morpheme or voice alternation in the Tagalog sense.

theorem Phenomena.FillerGap.Studies.Erlewine2018.tb_distinguishes : Fragments.TobaBatak.tbExtractionProfile.distinguishesPosition = true

TB distinguishes extracted positions via voice (which role is pivot).

Connects three independent data sources through the AH bridge:

1. Individual extraction datums (from @cite{erlewine-2018})
2. The ExtractionProfile summary (markedPositions)
3. RelClauseMarkers (from @cite{keenan-comrie-1977})

If any link is wrong — e.g., listing .directObject as extractable when relativization markers don't cover DO — the chain breaks.

Note: the extraction data uses ArgumentRole (agent, patient) while the ExtractionProfile and AH use ExtractionTarget (subject, directObject). Voice promotion means the patient's default position is DO, but its surface extraction position is always subject (the pivot).

theorem Phenomena.FillerGap.Studies.Erlewine2018.profile_matches_data : (Fragments.TobaBatak.tbExtractionProfile.markedPositions.all fun (pos : Typology.ExtractionTarget) => Fragments.TobaBatak.extractionData.any fun (d : Fragments.TobaBatak.ExtractionDatum) => match d.extracted with | Typology.Extractee.dpArg role => decide ((role.defaultPosition == pos) = true ∧ (d.judgment == Fragments.TobaBatak.ExtractionJudgment.grammatical) = true ∧ d.extractsPivot = true) | Typology.Extractee.adjunct => false) = true

Link 1→2: The ExtractionProfile's marked positions are exactly the positions for which some voice makes a DP argument grammatically extractable. Only .subject qualifies (the pivot position).

theorem Phenomena.FillerGap.Studies.Erlewine2018.extractable_positions_are_relativizable : (Fragments.TobaBatak.tbExtractionProfile.markedPositions.all fun (et : Typology.ExtractionTarget) => have ahPos := Typology.Relativization.extractionTargetToAH et; Fragments.TobaBatak.relMarkers.any fun (x : Core.RelClauseMarker) => x.covers ahPos) = true

Link 2→3: Every extractable position (ExtractionTarget) maps to an AH position that is covered by some Toba Batak relativization marker.

theorem Phenomena.FillerGap.Studies.Erlewine2018.extraction_profile_relativization_chain : (Fragments.TobaBatak.extractionData.all fun (d : Fragments.TobaBatak.ExtractionDatum) => match d.extracted with | Typology.Extractee.dpArg a => d.extractsPivot == (d.judgment == Fragments.TobaBatak.ExtractionJudgment.grammatical) | Typology.Extractee.adjunct => true) = true ∧ Fragments.TobaBatak.tbExtractionProfile.Marks Typology.ExtractionTarget.subject ∧ (Fragments.TobaBatak.tbExtractionProfile.markedPositions.all fun (et : Typology.ExtractionTarget) => have ahPos := Typology.Relativization.extractionTargetToAH et; Fragments.TobaBatak.relMarkers.any fun (x : Core.RelClauseMarker) => x.covers ahPos) = true

Full chain (all three links as a single conjunction).

def Phenomena.FillerGap.Studies.Erlewine2018.predictExtraction (voice : Fragments.TobaBatak.Voice) (extractee : Typology.Extractee) : Fragments.TobaBatak.ExtractionJudgment

Predict whether extraction is grammatical from voice + extractee.

The nominal licensing analysis predicts:

• DP arguments: extraction is grammatical iff the DP is the pivot, because only the pivot is Case-licensed (by T's [PROBE:D] in Spec,TP) before Ā-extraction.
• Non-DP adjuncts: always grammatical, because adjuncts don't need Case licensing.

Equations

• One or more equations did not get rendered due to their size.
• Phenomena.FillerGap.Studies.Erlewine2018.predictExtraction voice Typology.Extractee.adjunct = Fragments.TobaBatak.ExtractionJudgment.grammatical

theorem Phenomena.FillerGap.Studies.Erlewine2018.bridge_av_agent : predictExtraction Fragments.TobaBatak.Voice.av (Typology.Extractee.dpArg Typology.ArgumentRole.agent) = Fragments.TobaBatak.ExtractionJudgment.grammatical ∧ Fragments.TobaBatak.avAgentExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.grammatical

AV + agent (pivot): Case-licensed in Spec,TP → extractable.

theorem Phenomena.FillerGap.Studies.Erlewine2018.bridge_av_patient : predictExtraction Fragments.TobaBatak.Voice.av (Typology.Extractee.dpArg Typology.ArgumentRole.patient) = Fragments.TobaBatak.ExtractionJudgment.ungrammatical ∧ Fragments.TobaBatak.avPatientExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.ungrammatical

AV + patient (non-pivot): no Case licensor in Spec,CP.

theorem Phenomena.FillerGap.Studies.Erlewine2018.bridge_av_oblique : predictExtraction Fragments.TobaBatak.Voice.av (Typology.Extractee.dpArg Typology.ArgumentRole.oblique) = Fragments.TobaBatak.ExtractionJudgment.ungrammatical ∧ Fragments.TobaBatak.avObliqueExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.ungrammatical

AV + oblique (non-pivot): no Case licensor in Spec,CP.

theorem Phenomena.FillerGap.Studies.Erlewine2018.bridge_ov_patient : predictExtraction Fragments.TobaBatak.Voice.ov (Typology.Extractee.dpArg Typology.ArgumentRole.patient) = Fragments.TobaBatak.ExtractionJudgment.grammatical ∧ Fragments.TobaBatak.ovPatientExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.grammatical

OV + patient (pivot): Case-licensed in Spec,TP → extractable.

theorem Phenomena.FillerGap.Studies.Erlewine2018.bridge_ov_agent : predictExtraction Fragments.TobaBatak.Voice.ov (Typology.Extractee.dpArg Typology.ArgumentRole.agent) = Fragments.TobaBatak.ExtractionJudgment.ungrammatical ∧ Fragments.TobaBatak.ovAgentExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.ungrammatical

OV + agent (non-pivot): no Case licensor in Spec,CP.

theorem Phenomena.FillerGap.Studies.Erlewine2018.bridge_ov_oblique : predictExtraction Fragments.TobaBatak.Voice.ov (Typology.Extractee.dpArg Typology.ArgumentRole.oblique) = Fragments.TobaBatak.ExtractionJudgment.ungrammatical ∧ Fragments.TobaBatak.ovObliqueExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.ungrammatical

OV + oblique (non-pivot): no Case licensor in Spec,CP.

theorem Phenomena.FillerGap.Studies.Erlewine2018.bridge_av_adjunct : predictExtraction Fragments.TobaBatak.Voice.av Typology.Extractee.adjunct = Fragments.TobaBatak.ExtractionJudgment.grammatical ∧ Fragments.TobaBatak.avAdjunctExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.grammatical

AV + adjunct: no Case needed → freely extractable.

theorem Phenomena.FillerGap.Studies.Erlewine2018.bridge_ov_adjunct : predictExtraction Fragments.TobaBatak.Voice.ov Typology.Extractee.adjunct = Fragments.TobaBatak.ExtractionJudgment.grammatical ∧ Fragments.TobaBatak.ovAdjunctExtraction.judgment = Fragments.TobaBatak.ExtractionJudgment.grammatical

OV + adjunct: no Case needed → freely extractable.

theorem Phenomena.FillerGap.Studies.Erlewine2018.all_predictions_match : (Fragments.TobaBatak.extractionData.all fun (d : Fragments.TobaBatak.ExtractionDatum) => predictExtraction d.voice d.extracted == d.judgment) = true

The structural analysis correctly predicts every monoclausal datum.

theorem Phenomena.FillerGap.Studies.Erlewine2018.prediction_matches_pivot : (Fragments.TobaBatak.extractionData.all fun (d : Fragments.TobaBatak.ExtractionDatum) => match d.extracted with | Typology.Extractee.dpArg a => (predictExtraction d.voice d.extracted == Fragments.TobaBatak.ExtractionJudgment.grammatical) == d.extractsPivot | Typology.Extractee.adjunct => true) = true

For DP arguments, the prediction function agrees with extractsPivot.

theorem Phenomena.FillerGap.Studies.Erlewine2018.structural_grounding : (Fragments.TobaBatak.extractionData.all fun (d : Fragments.TobaBatak.ExtractionDatum) => match d.extracted with | Typology.Extractee.dpArg a => predictExtraction d.voice d.extracted == if d.extractsPivot = true then Fragments.TobaBatak.ExtractionJudgment.grammatical else Fragments.TobaBatak.ExtractionJudgment.ungrammatical | Typology.Extractee.adjunct => true) = true

For DP arguments, the prediction function IS the pivot check — they agree extensionally on the extraction data.

theorem Phenomena.FillerGap.Studies.Erlewine2018.nonDP_unrestricted (v : Fragments.TobaBatak.Voice) : predictExtraction v Typology.Extractee.adjunct = Fragments.TobaBatak.ExtractionJudgment.grammatical

The nominal licensing analysis predicts non-DPs extract freely.

theorem Phenomena.FillerGap.Studies.Erlewine2018.predicate_fronting_yields_vi_order : (Minimalist.HeadFunction.leftSpine.phonYield ColeHermon2008.tobaBatakVOS.final).head? = some "mangatuk"

VP-raising (@cite{cole-hermon-2008}, Toba Batak) and predicate fronting (@cite{erlewine-2018}, Toba Batak) share the same core prediction: the predicate phrase moves above the subject, yielding V-initial surface order.

Both analyses predict:

1. The predicate c-commands the subject at surface structure
2. Only the subject (pivot) can subsequently Ā-extract
3. Non-DP adjuncts extract freely (not subject to Case licensing)

The key parametric difference:

• @cite{cole-hermon-2008}: VP moves to Spec,TP; subject stranded in Spec,vP
• @cite{erlewine-2018}: vP moves to Spec,CP; subject stranded in Spec,TP

vP-to-Spec,CP analysis

@cite{erlewine-2018}'s analysis differs structurally from @cite{cole-hermon-2008}:

• @cite{cole-hermon-2008}: VP → Spec,TP
• Erlewine: Subj → Spec,TP then vP → Spec,CP

Both derive the same VOS surface order, but the derived tree is structurally different: Erlewine's has an additional CP layer, and the fronted constituent is vP (containing a subject trace) rather than bare VP.

def Phenomena.FillerGap.Studies.Erlewine2018.v_head_e : Minimalist.SyntacticObject

Little v (Erlewine's analysis, unique ID).

Equations

• Phenomena.FillerGap.Studies.Erlewine2018.v_head_e = Minimalist.mkLeaf Minimalist.Cat.v [Minimalist.Cat.V] 101

def Phenomena.FillerGap.Studies.Erlewine2018.t_head_e : Minimalist.SyntacticObject

T head (Erlewine's analysis, unique ID).

Equations

• Phenomena.FillerGap.Studies.Erlewine2018.t_head_e = Minimalist.mkLeaf Minimalist.Cat.T [Minimalist.Cat.v] 102

def Phenomena.FillerGap.Studies.Erlewine2018.c_head_e : Minimalist.SyntacticObject

C head bearing [PROBE:FOC] (Erlewine's analysis, unique ID).

Equations

• Phenomena.FillerGap.Studies.Erlewine2018.c_head_e = Minimalist.mkLeaf Minimalist.Cat.C [Minimalist.Cat.T] 103

def Phenomena.FillerGap.Studies.Erlewine2018.vP_traced : Minimalist.SyntacticObject

The vP after subject extraction: [vP tSubj [v' v [VP V Obj]]].

The trace marks where the subject DP originated before moving to Spec,TP.

Equations

• Phenomena.FillerGap.Studies.Erlewine2018.vP_traced = (Minimalist.mkTrace 0).node (Phenomena.FillerGap.Studies.Erlewine2018.v_head_e.node ColeHermon2008.vp)

def Phenomena.FillerGap.Studies.Erlewine2018.erlewineDerivation : Minimalist.Derivation

Erlewine's vP-to-Spec,CP derivation for Toba Batak VOS.

Steps (bottom-up):

1. EM-R Obj → [VP V Obj]
2. EM-L v → [v' v VP]
3. EM-L Subj → [vP Subj [v' v VP]]
4. EM-L T → [TP T [vP Subj [v' v VP]]]
5. IM Subj → [TP Subj [T' T [vP tSubj [v' v VP]]]]
6. EM-L C → [CP C [TP Subj [T' T [vP tSubj [v' v VP]]]]]
7. IM vP → [CP [vP tSubj v VP] [C' C [TP Subj [T' T tvP]]]]

Equations

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

theorem Phenomena.FillerGap.Studies.Erlewine2018.erlewine_yields_vos : Minimalist.HeadFunction.leftSpine.phonYield erlewineDerivation.final = ["mangatuk", "biangi", "dakdanakan"]

Erlewine's derivation yields VOS word order.

theorem Phenomena.FillerGap.Studies.Erlewine2018.cole_erlewine_agree_on_order : Minimalist.HeadFunction.leftSpine.phonYield ColeHermon2008.tobaBatakVOS.final = Minimalist.HeadFunction.leftSpine.phonYield erlewineDerivation.final

Both analyses agree on VOS surface order despite different structural heights.

theorem Phenomena.FillerGap.Studies.Erlewine2018.erlewine_two_movements : erlewineDerivation.movedItems.length = 2

Erlewine has TWO movements (Subj → Spec,TP + vP → Spec,CP) vs @cite{cole-hermon-2008}'s ONE (VP → Spec,TP).

theorem Phenomena.FillerGap.Studies.Erlewine2018.different_derived_structure : ColeHermon2008.tobaBatakVOS.final.nodeCount ≠ erlewineDerivation.final.nodeCount

Different derived structures despite the same word order. Phase 1.0 sorry: .shape no longer typechecks; the count-based weakening also collapses (both have the same count).

theorem Phenomena.FillerGap.Studies.Erlewine2018.vP_ccommands_subject_erlewine : Minimalist.cCommandsIn erlewineDerivation.final vP_traced ColeHermon2008.n_dakdanakan

Fronted vP c-commands the subject in Erlewine's derived tree.

The fronted vP is in Spec,CP; its sister C' dominates the subject in Spec,TP. This yields the same binding prediction as @cite{cole-hermon-2008}'s VP-in-Spec,TP analysis: the predicate phrase c-commands the subject.

theorem Phenomena.FillerGap.Studies.Erlewine2018.same_vp : ColeHermon2008.tobaBatakVOS.stageAt 1 = erlewineDerivation.stageAt 1

Same VP base in both analyses (stage after first merge).

Erlewine's vP-to-Spec,CP fronting as a RemnantFronting instance

@cite{erlewine-2018}'s analysis is a genuine remnant fronting: the subject DP first moves out of vP to Spec,TP (step 5 of erlewineDerivation), and only then does the remnant vP — now containing only the subject trace — front to Spec,CP (step 7).

This is structurally different from @cite{cole-hermon-2008}'s analysis: there, the VP fronts with the subject still in vP-internal position; under Erlewine's analysis, the subject has evacuated before the larger constituent fronts. Erlewine's case fits RemnantFronting more directly.

def Phenomena.FillerGap.Studies.Erlewine2018.erlewineVPRemnant : Minimalist.Movement.RemnantFronting

Erlewine's vP-to-Spec,CP fronting as a remnant-fronting witness. The subject DP n_dakdanakan evacuates to Spec,TP before the vP fronts to Spec,CP, leaving the trace mkTrace 0 behind.

Equations

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