Documentation

Linglib.Phenomena.ArgumentStructure.Studies.Storment2026

Unaccusativity Bridge @cite{storment-2026} #

Connects empirical diagnostic data to lexical annotations, syntactic theory (Voice, TransitivityClass, auxiliary selection, smuggling), and semantic proto-role predictions.

@cite{storment-2026}: QI via Smuggling #

Quotative inversion (QI) is an unaccusativity diagnostic: manner-of-speaking verbs like whisper permit QI ("whispered Mary") because they are unaccusative, while agentive communication verbs like speak block QI (*"spoke Mary") because they are unergative.

The syntactic mechanism is smuggling: VP moves to Spec-VoiceP, making the theme subject accessible to T⁰ for Case licensing. This is possible iff Voice is not a phase head (non-thematic Voice), which correlates with unaccusativity.

QI ∥ LI (Unified Smuggling) #

@cite{storment-2026} argues QI and locative inversion share the same mechanism: @cite{collins-2005} @cite{dowty-1991} @cite{lu-degen-2025} smuggling of VP to Spec-VoiceP. Both are blocked by the transitivity constraint (§5) and both require non-phase Voice. They differ in input (quote vs locative PP).

Dowty Divergence #

Proto-role counting predicts MoS verb subjects as proto-agents (volitional, sentient speaker), yet QI classifies these verbs as unaccusative. This is a known divergence: unaccusativity is a syntactic diagnostic that does not always align with semantic proto-role predictions.

Each theorem ties to exactly one verb's unaccusative field. Changing the annotation breaks exactly one theorem.

Each theorem connects a diagnostic datum to the corresponding lexical annotation. If either changes, the theorem breaks.

Derive TransitivityClass from VerbCore fields: unaccusative → .unaccusative, complement-taking → .transitive, otherwise → .unergative.

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

    Map unaccusativity to Voice flavor: unaccusative verbs project non-thematic Voice (no external argument); others project agentive Voice.

    Equations
    Instances For

      In split-auxiliary languages (Italian, French, German), unaccusative verbs select be and unergative verbs select have. We verify this via canonicalSelection ∘ deriveTransitivityClass.

      MoS verb subjects are volitional, sentient speakers — Dowty's proto-role counting yields P-Agent > P-Patient, predicting unergative. But QI says these verbs are unaccusative. This is a documented divergence between semantic proto-role predictions and syntactic diagnostics.

      MoS subject entailment profile: volitional (P-Agent), sentient (P-Agent), exists independently (P-Agent), but no P-Patient entailments.

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

        The smuggling analysis (@cite{collins-2005}, applied to QI by @cite{storment-2026}) derives QI licensing from two independently motivated properties:

        1. **Voice is non-phase** (= unaccusative): complement remains extractable
        2. **Verb has a complement** (the quote): there is something to smuggle
        
        We verify that `licensesQI` (defined in `Movement/Smuggling.lean` over
        Voice + complement availability) matches the empirical QI diagnostic data
        when applied to the Voice and complement type derived from each verb's
        lexical entry. 
        

        Derived QI licensing: combines Voice (from unaccusativity) and complement availability (from complementType).

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

          speak is blocked: agentive Voice makes vP a phase (blocks smuggling), AND it has no complement (.none) — doubly ruled out.

          arrive is unaccusative but has no complement to smuggle — so it doesn't license QI. This is correct: *"arrived Mary" requires a fronted locative (LI), not a fronted quote (QI).

          theorem Phenomena.ArgumentStructure.Unaccusativity.Bridge.qi_data_consistent :
          (Data.qi_whisper.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.whisper.toVerbCore = true) (Data.qi_murmur.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.murmur.toVerbCore = true) (Data.qi_shout.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.shout.toVerbCore = true) (Data.qi_cry.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.cry.toVerbCore = true) (Data.qi_scream.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.scream.toVerbCore = true) (Data.qi_mumble.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.mumble.toVerbCore = true) (Data.qi_mutter.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.mutter.toVerbCore = true) (Data.qi_shriek.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.shriek.toVerbCore = true) (Data.qi_yell.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.yell.toVerbCore = true) (Data.qi_groan.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.groan.toVerbCore = true) (Data.qi_grumble.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.grumble.toVerbCore = true) (Data.qi_hiss.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.hiss.toVerbCore = true) (Data.qi_sigh.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.sigh.toVerbCore = true) (Data.qi_whimper.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.whimper.toVerbCore = true) (Data.qi_snap.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.snap.toVerbCore = true) (Data.qi_speak.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.speak.toVerbCore = true) (Data.qi_talk.result = Data.DiagnosticResult.passes) = (derivedQI Fragments.English.Predicates.Verbal.talk.toVerbCore = true)

          Quotative inversion and locative inversion share the same mechanism (smuggling of VP to Spec-VoiceP, @cite{storment-2026} §6) but differ in their inputs and distribution:

          - QI requires a quote complement; LI requires a fronted locative PP
          - Both are subject to the transitivity constraint (§5): multiple DP
            arguments block inversion due to Case licensing limitations
          - QI degrades with pronominal subjects; LI categorically blocks them
          
          Despite sharing a mechanism, their distributional differences confirm
          they are distinct diagnostic constructions with different triggers. 
          

          LI blocks transitive verbs; QI with just a quote complement is fine, but both show a transitivity constraint with multiple DP arguments.

          Unified smuggling analysis (§6): LI with arrive works because arrive projects non-thematic (non-phase) Voice, permitting VP to smuggle to Spec-VoiceP — the same mechanism that licenses QI.

          MoS verbs participate in two seemingly opposite phenomena:

          1. **MoS islands**: wh-extraction FROM WITHIN
             the complement is degraded (discourse backgroundedness)
          2. **Quotative inversion**: VP-smuggling to Spec-VoiceP
             is grammatical (A-movement, not extraction from complement)
          
          These are not contradictory — they're different operations:
          - MoS islands: wh-movement from within CP complement (blocked by discourse)
          - QI smuggling: A-movement of VP to Spec-VoiceP (licensed by syntax)
          
          The asymmetry validates both analyses simultaneously: MoS islands are
          discourse-sourced (not syntactic), so they don't block the syntactic
          VP-movement that produces QI. 
          

          The smuggling derivation of QI assigns each major constituent to a structural position. Each position has observable consequences that are tested against the §3 structural evidence data.

          1. **Theme** → Spec-TP (A-movement): controls agreement, licenses
             raising, does NOT license parasitic gaps (A ≠ Ā)
          2. **Agent** → Spec-vP (in-situ): conjoint domain in Setswana,
             does not control agreement (defective circumvention)
          3. **VP** → Spec-VoiceP (smuggling): VP-internal material precedes
             Agent; vP-external adjuncts follow Agent
          4. **Quote** → DiscourseP (clause-external): can split around
             V + Agent, need not be grammatical 
          

          Structural position in QI derivation.

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

              The QI derivation assigns constituents to structural positions. Each position has testable predictions (§3).

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

                  The smuggling derivation of QI.

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

                    Each structural position predicts observable properties. We verify each prediction against the §3 data.

                    Theme in Spec-TP → agreement can track agent via defective circumvention: probe first hits phi-deficient Theme, then optionally re-probes and finds Agent (§3.1).

                    Theme in Spec-TP → Setswana SM surfaces as default SM17 because Theme is phi-deficient; probe doesn't reach Agent (§3.1).

                    Theme in Spec-TP → A-movement → parasitic gaps NOT licensed. A-movement chains cannot license PGs; only Ā-chains can (§3.2).

                    Theme in Spec-TP → compatible with subject-to-subject raising. Theme can A-move through intermediate Spec-TP positions (§3.3).

                    Agent in Spec-vP → Setswana disjoint morpheme blocked. Disjoint form requires the subject to be vP-external; Agent remains vP-internal in QI (§3.4).

                    Quote in DiscourseP → quote can split around verb + agent. A constituent in Spec-TP cannot split; DiscourseP can (§3.5).

                    Quote in DiscourseP → quote need not be grammatical. Clause-external status means quote is not subject to clausal well-formedness requirements (§3.5).

                    Levin & Rappaport @cite{levin-hovav-1995} predict unaccusativity from verb class membership: CoS classes (§45) predict unaccusative for their inchoative alternant, manner-of-motion (§51.3) predicts unergative, inherently directed motion (§51.1) predicts unaccusative, and emission classes (§43) predict unaccusative.

                    The function `LevinClass.predictsUnaccusative` (in RootDimensions.lean)
                    encodes these predictions. Here we verify them against Fragment entries
                    that carry both `.levinClass` and `.unaccusative` annotations. 
                    

                    MoS verbs: Levin class does NOT predict unaccusativity (agentive manner activity), but they ARE empirically unaccusative (@cite{storment-2026} QI diagnostic). This divergence motivates Storment's syntactic analysis (smuggling) over a purely lexical-semantic account of unaccusativity.

                    Break (§45.1): the Fragment entry is the transitive causative form (unaccusative = false), but the class predicts an unaccusative inchoative alternant ("the vase broke"). The causative/inchoative alternation is the bridge between the transitive entry and the predicted unaccusativity.