Documentation

Linglib.Phenomena.Ellipsis.Studies.CitkoGracaninYuksek2025

Economy in PF Reduction #

@cite{citko-gracanin-yuksek-2025}

Coordinated wh-questions (CWHs) and coordinated sluices (CSs) are two PF-reduced constructions with wh-phrase remnants. Despite superficial similarity, they differ in structure, derivational cost, and empirical properties. Economy governs the choice between ellipsis and multidominance as the PF reduction mechanism.

Key Claims #

  1. CWHs use non-bulk-sharing MD: each conjunct CP contains one wh-phrase; functional heads (C, T) are multiply dominated. No ellipsis is involved.

  2. CSs use bulk-sharing MD + ellipsis: the entire C' is shared between conjuncts; both wh-phrases originate inside the shared vP. The E-feature on C triggers TP deletion (cf. FeatureVal.ellipsis in Core/Features.lean, which models this E-feature).

  3. Economy selects the structure: MD is preferred over ellipsis (fewer operations, fewer lexical items). Bulk-sharing MD is more economical than non-bulk-sharing, but is blocked for CWHs by the MWF parameter. CSs can't have the CWH structure because of Pronunciation Economy.

Empirical Contrasts #

Substrate engagement #

Integration #

The two PF-reduced wh-coordination constructions.

  • CWH : WHCoordType

    Coordinated wh-question: "What and when did John teach?"

  • CS : WHCoordType

    Coordinated sluice: "I forgot what and when."

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

      An empirical datum contrasting CWHs and CSs.

      • sentence : String
      • construction : WHCoordType
      • whPhrases : List String
      • grammatical : Bool
      • notes : String
      Instances For
        Equations
        • One or more equations did not get rendered due to their size.
        Instances For
          Equations
          Instances For

            CWHs ban coordination of obligatory wh-arguments (ex 5a).

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

              CWHs ban wh-object + wh-adjunct with obligatorily transitive verbs (ex 6a). Distinct from (5a): here only one wh-phrase is an obligatory argument; the other is an adjunct. The ban persists because the verb ('buy') is obligatorily transitive.

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

                  CSs allow coordination of obligatory wh-arguments (ex 5b).

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

                    CSs allow wh-object + wh-adjunct coordination (ex 6b).

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

                      All CWH/CS empirical contrasts as a single drift-sentry table. Replaces seven trivial rfl readouts (cwh_bans_obligatory_args etc.) with one decide-checked aggregate that catches drift in any row.

                      Equations
                      • One or more equations did not get rendered due to their size.
                      Instances For
                        theorem CitkoGracaninYuksek2025.whCoordContrastTable_consistent :
                        (whCoordContrastTable.all fun (x : WHCoordDatum × Bool) => match x with | (d, expected) => d.grammatical == expected) = true

                        Reading type for wh-coordination.

                        • paired : ReadingType

                          Each wh-phrase ranges over a single value (n-tuple as the answer).

                        • nonpaired : ReadingType

                          Each wh-phrase ranges over multiple values (cross-product answer).

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

                            Reading availability datum.

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

                                CWHs: only nonpaired reading available (ex 8a).

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

                                  CWHs: paired reading unavailable (ex 8a).

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

                                    CSs: paired reading available with obligatorily transitive verbs (ex 8b). The bulk-sharing CS structure puts both wh-phrases inside a single vP.

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

                                      CSs: nonpaired reading available with optionally transitive verbs (§6.1, ex 43). With optionally transitive verbs, the CWH-style structure becomes available, yielding a nonpaired reading.

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

                                        Drift sentry: the four reading datums encode the paper's central paired/nonpaired contrast — CWH bans paired, CS allows both.

                                        The "what" conjunct of cs_basic as a standalone sluice: "John taught something, but I forgot what."

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

                                          The "when" conjunct of cs_basic as a standalone sluice: "John taught (then), but I forgot when."

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

                                            Each CS decomposes into sluices: the wh-phrases in a CS are each sluice remnants.

                                            The "what" and "to whom" conjuncts of cs_obligatory_args_allowed.

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

                                                Bulgarian, German, Greek MWF data live in Phenomena/Questions/MultipleWh.lean (textbook-consensus typology). The intra-English variety A/B split is paper-specific to @cite{citko-gracanin-yuksek-2025} (p. 19): both varieties are non-MWF in matrix questions, but they diverge on multiple sluicing — variety A bans it, variety B allows it. The paper derives this asymmetry from where the PF asterisk lands (vP only vs both vP and CP edges), not as a free parameter. The tri-valued MWFParameter in MultipleWh.lean captures this derivation: variety B is .nonFrontsVPOnly, variety A is .nonFrontsBothEdges.

                                                English (default): variety A — both vP and CP asterisks; no sluicing repair.

                                                Equations
                                                Instances For

                                                  English bans multiple wh-specifiers at a phase edge.

                                                  English variety A: non-MWF, bans multiple sluicing. The CP-edge asterisk survives ellipsis (vP-edge deletion is not enough).

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

                                                    English variety B: non-MWF, allows multiple sluicing. Only the vP edge has the asterisk; vP-edge deletion repairs it.

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

                                                      Variety A bans multiple sluicing, variety B allows it. Derived from the parameter via EllipsisRepairsMWF, not stipulated as an independent Bool.

                                                      Cost defs unchanged in shape from earlier revisions. The 4-Nat parameterization (sm/sl/nsm/nsl coarse, hm/hl/pm/pl/wm/wl fine) is ergonomically cryptic but matches the paper's argument granularity:

                                                      Coarse (MD vs ellipsis): parameterized by total shared/non-shared cost. Used for Theorems 1–2.

                                                      Fine (non-bulk vs bulk sharing): decomposes shared material into heads vs phrasal structure. Used for Theorem 3.

                                                      Cost of CWH via non-bulk-sharing MD (adopted structure, paper's (10b)). Shared material built once; no ellipsis.

                                                      Equations
                                                      Instances For

                                                        Cost of CWH via ellipsis (biclausal alternative, paper's (11b)). Shared material duplicated; one E-feature deletion.

                                                        Equations
                                                        Instances For

                                                          Cost of CWH via non-bulk-sharing MD — fine-grained (paper's (10b)). Individual heads (C, T) shared; per-conjunct phrasal structure (assembling C', TP, vP around shared heads) built in each conjunct.

                                                          Equations
                                                          Instances For

                                                            Cost of CWH via bulk-sharing MD (excluded, paper's (12b)). Entire C' shared — phrasal structure built only once. More economical than non-bulk-sharing, but blocked by MWF.

                                                            Equations
                                                            Instances For

                                                              Cost of CS via bulk-sharing MD + single ellipsis (adopted, paper's (20b)). C' shared; E-feature on C deletes TP once.

                                                              Equations
                                                              Instances For

                                                                Cost of CS via double ellipsis, no MD (excluded, paper's (19b)). Both conjuncts built in full; E-feature in each.

                                                                Equations
                                                                Instances For

                                                                  Theorem 1: For CWHs, the MD derivation is strictly more economical than the ellipsis alternative (paper's (10b) vs (11b)).

                                                                  The MD derivation saves sm Merge ops and sl lexical items, and avoids the ellipsis op entirely. This holds for ANY shared material — even sm = sl = 0, the ellipsis op alone makes the alternative costlier.

                                                                  Theorem 2: For CSs, the bulk-sharing derivation is strictly more economical than the double-ellipsis alternative (paper's (20b) vs (19b)).

                                                                  The bulk-sharing derivation saves sm Merge operations and sl lexical items, and uses one fewer ellipsis operation.

                                                                  theorem CitkoGracaninYuksek2025.cwh_bulk_beats_nonbulk (hm hl pm pl wm wl : ) (h : 0 < pm 0 < pl) :
                                                                  Minimalist.strictlyMoreEconomical (cwhBulkMDCost hm hl pm pl wm wl) (cwhNonBulkCost hm hl pm pl wm wl)

                                                                  Theorem 3: Bulk-sharing is strictly more economical than non-bulk-sharing for CWHs (paper's (12b) vs (10b)).

                                                                  The paper's crucial insight: the MOST economical derivation for CWHs (bulk-sharing, which builds C' once) is blocked by an independent constraint (MWF), forcing the LESS economical non-bulk-sharing structure. The precondition 0 < pm ∨ 0 < pl ensures there is at least some phrasal structure to share — which holds for any non-trivial clause.

                                                                  Identifying economy winners #

                                                                  The Theorems 1-3 above establish strict-domination facts; combining each with Minimalist.economy_winner_of_pair (which discharges the existence guarantee from economy_admits_winner for the binary case) identifies the actual economy winner of the relevant 2-candidate reference set. These are the load-bearing wirings: C&G-Y's selection arguments need both "X dominates Y" AND "the reference set has a winner" to deliver "X is THE winner".

                                                                  theorem CitkoGracaninYuksek2025.cwh_md_is_economy_winner (sm sl nsm nsl : ) (alt : Minimalist.DerivationCost) :
                                                                  alt {cwhMDCost sm sl nsm nsl, cwhEllipsisCost sm sl nsm nsl}¬Minimalist.strictlyMoreEconomical alt (cwhMDCost sm sl nsm nsl)

                                                                  CWH economy winner: of the reference set {cwhMDCost, cwhEllipsisCost}, cwhMDCost is an economy winner — no element of the set is strictly more economical than it. Wires cwh_md_beats_ellipsis through Minimalist.economy_winner_of_pair.

                                                                  theorem CitkoGracaninYuksek2025.cs_bulk_is_economy_winner (sm sl nsm nsl : ) (alt : Minimalist.DerivationCost) :
                                                                  alt {csBulkCost sm sl nsm nsl, csDoubleEllipsisCost sm sl nsm nsl}¬Minimalist.strictlyMoreEconomical alt (csBulkCost sm sl nsm nsl)

                                                                  CS economy winner: of the reference set {csBulkCost, csDoubleEllipsisCost}, csBulkCost is an economy winner. Wires cs_bulk_beats_double_ellipsis through Minimalist.economy_winner_of_pair.

                                                                  The CS (bulk-sharing) structure places both wh-phrases inside a single vP. Both must pass through the vP phase edge, creating a phase node with multiple wh-specifiers (paper's (36b)/(37b)).

                                                                  In a non-MWF language like English, this configuration receives an asterisk at PF, crashing the derivation.

                                                                  Unlike CSs, CWHs do NOT involve ellipsis — the vP edge survives to PF, so the asterisk is not deleted and the crash is unavoidable.

                                                                  CSs survive because the E-feature on C triggers TP deletion (including the offending vP edge), removing the asterisk before PF interprets it.

                                                                  The bulk-sharing structure for CWHs crashes in English: 2 wh-specifiers at vP edge in a non-MWF language.

                                                                  CWHs have no ellipsis to repair the MWF violation — the vP edge survives to PF, and the asterisk crashes.

                                                                  In English variety B (vP-only asterisk), CSs survive the same MWF configuration because ellipsis deletes the vP edge. Variety A (both-edges asterisk) is not repairable — CP-edge survives.

                                                                  Paper p. 32 (re. ex (45c)): "If both Cs had the E feature, … Pronunciation Economy would be violated since the TP would be elided twice (once by the E feature on each C)."

                                                                  The argument depends on the CS structure where a shared TP has two Cs each bearing the E-feature. Both Cs fire the E-feature on the same TP — but only the first deletion has any PF effect; the second targets material the first already removed. This is the per-operation vacuity that requires the new pronunciationEconomy : List PFOperation → Prop shape: a whole-derivation pfBefore ≠ pfAfter check would let this through (the first deletion ensures the whole derivation's PF differs).

                                                                  Initial PF state of the shared TP: contains "John taught".

                                                                  Equations
                                                                  Instances For

                                                                    The first C's E-feature deletion: removes "John taught" from the shared TP. Non-vacuous.

                                                                    Equations
                                                                    Instances For

                                                                      The second C's E-feature deletion: tries to delete the same shared TP, but it is already empty after the first deletion. Vacuous.

                                                                      Equations
                                                                      Instances For

                                                                        The second deletion is vacuous (its PF before/after coincide).

                                                                        The two-E-feature CS structure violates Pronunciation Economy at the second deletion. This is the paper's (45c) ban — the per-op vacuity is what rules it out, even though the full derivation's pfBefore (["John", "taught"]) ≠ pfAfter ([]) is satisfied.

                                                                        A whole-derivation PF≠ check would pass this configuration incorrectly; the per-op pronunciationEconomy correctly fails it.

                                                                        Whole-derivation PF differs (initial ["John","taught"] → final []) even though the second op is vacuous. This is exactly what makes a whole-derivation PF≠ check inadequate for the paper's argument.

                                                                        The adopted CWH structure: non-bulk-sharing MD, no ellipsis (paper's (10b)). Shared nodes: C and T are individually multiply dominated.

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

                                                                          The adopted CS structure: bulk-sharing MD + ellipsis (paper's (20b)). The E-feature on C (cf. FeatureVal.ellipsis in Core/Features.lean) triggers TP deletion, repairing the MWF violation at the vP edge. Shared nodes: entire C' is shared (includes C, TP, vP, VP).

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

                                                                            Structural drift sentry over the four key properties of cwhStructure and csStructure. Replaces five individual rfl theorems (cwh_uses_md_only, cs_uses_both, cwh_shares_heads_cs_shares_constituent, cwh_shared_categories, cs_shared_categories).

                                                                            The paper's central contribution is explaining WHY CWHs and CSs must have different structures — and therefore different empirical properties — despite their superficial similarity.

                                                                            For CWHs (§4.1):

                                                                            1. Bulk-sharing MD is most economical (cwh_bulk_beats_nonbulk)
                                                                            2. But bulk-sharing creates MWF violation at vP (cwh_bulk_crashes_in_english)
                                                                            3. CWHs have no ellipsis to repair MWF (cwh_no_ellipsis_repair)
                                                                            4. So non-bulk-sharing MD is selected (cwh_md_beats_ellipsis over ellipsis)

                                                                            For CSs (§4.2):

                                                                            1. Bulk-sharing MD + ellipsis is most economical (cs_bulk_beats_double_ellipsis)
                                                                            2. Bulk-sharing creates MWF at vP, but ellipsis repairs it (cs_ellipsis_repairs_mwf_varietyB, in vP-only varieties)
                                                                            3. Non-bulk-sharing (CWH structure) violates Pronunciation Economy (cs_twoEFeature_violates_pronunciation_economy)
                                                                            4. So bulk-sharing MD + ellipsis is selected

                                                                            The two constructions differ in which mechanisms they use.

                                                                            The two constructions differ in sharing type.

                                                                            End-to-end: CWHs cannot use bulk-sharing (most economical) because the MWF violation at vP cannot be repaired without ellipsis. Combines Theorem 3, cwh_bulk_crashes_in_english, and cwh_no_ellipsis_repair.

                                                                            End-to-end: CSs use bulk-sharing because it is most economical AND the MWF violation is repaired by ellipsis (in vP-only varieties); the CWH (two-E-feature) structure is excluded by per-op Pronunciation Economy.

                                                                            Section 6.2 of the paper extends the economy analysis to Right Node Raising (RNR). RNR is a key test case because it can involve BOTH ellipsis and MD simultaneously — the "mix and match" analysis of @cite{belk-neeleman-philip-2023} (and earlier @cite{barros-vicente-2011}).

                                                                            The core claim: economy favors MD over ellipsis when both yield the same string and interpretation. In RNR, the shared pivot (rightmost material) is multiply dominated. When the pivot and antecedent are not morphologically identical (vehicle change), ellipsis is additionally required for the non-shared material in the first conjunct.

                                                                            RNR pivot type: what is shared at the right edge.

                                                                            • minimal : RNRPivotType

                                                                              Only the rightmost constituent (e.g., PP) is shared.

                                                                            • extended : RNRPivotType

                                                                              A larger constituent (e.g., VP) is shared.

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

                                                                                An RNR datum captures the structural analysis.

                                                                                • sentence : String
                                                                                • pivot : String

                                                                                  The shared pivot string (right-edge material).

                                                                                • morphologicalIdentity : Bool

                                                                                  Does the pivot exhibit morphological identity with the antecedent?

                                                                                • mdProperties : Bool

                                                                                  Does the pivot exhibit properties of MD? (e.g., cumulative agreement, internal reading of relational adjective)

                                                                                • involvesEllipsis : Bool

                                                                                  Does the construction involve ellipsis in addition to MD?

                                                                                • pivotType : RNRPivotType
                                                                                • notes : String
                                                                                Instances For
                                                                                  Equations
                                                                                  • One or more equations did not get rendered due to their size.
                                                                                  Instances For

                                                                                    Pure MD: identical verbs, no mismatch → economy selects MD only (ex 55).

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

                                                                                      Ellipsis + MD: verb morphology mismatch forces ellipsis for VP, but PP pivot is still shared via MD (ex 52/53).

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

                                                                                        Vehicle-change diagnostic from @cite{barros-vicente-2011} ex (47): morphological mismatch signals ellipsis (not MD).

                                                                                        The angled brackets mark the first conjunct's predicate as PF-reduced under identity with the second conjunct's overt predicate — backwards elision, not forwards RNR with a right-edge pivot. The morphological mismatch (wake vs wakes) confirms ellipsis (which tolerates vehicle change) over MD (which requires morphological identity for a single multiply-dominated node).

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

                                                                                          Relational adjective with internal reading signals MD.

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

                                                                                            RNR datum drift sentry: morphological-identity / MD-properties / ellipsis-involvement consistent with the paper's typology. Replaces rnr_economy_prefers_md and rnr_mismatch_forces_ellipsis.

                                                                                            def CitkoGracaninYuksek2025.rnrMDPivotCost (pivotMerge pivotLex nonsharedMerge nonsharedLex : ) :

                                                                                            RNR pivot cost: MD derivation for the shared pivot. The pivot is built once (MD) rather than twice (ellipsis).

                                                                                            Equations
                                                                                            • One or more equations did not get rendered due to their size.
                                                                                            Instances For
                                                                                              def CitkoGracaninYuksek2025.rnrEllipsisCost (pivotMerge pivotLex nonsharedMerge nonsharedLex : ) :

                                                                                              RNR cost with full ellipsis (no MD): pivot duplicated.

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

                                                                                                Theorem 4: For RNR, MD is strictly more economical than ellipsis when both yield the same string. Same shape as Theorem 1.

                                                                                                theorem CitkoGracaninYuksek2025.rnr_md_is_economy_winner (pm pl nm nl : ) (alt : Minimalist.DerivationCost) :
                                                                                                alt {rnrMDPivotCost pm pl nm nl, rnrEllipsisCost pm pl nm nl}¬Minimalist.strictlyMoreEconomical alt (rnrMDPivotCost pm pl nm nl)

                                                                                                RNR economy winner: of the reference set {rnrMDPivotCost, rnrEllipsisCost}, rnrMDPivotCost is an economy winner. Wires rnr_md_beats_ellipsis through Minimalist.economy_winner_of_pair.

                                                                                                @cite{bruening-2021}'s max-projection identity condition for ellipsis (formalized as Minimalist.Ellipsis.FormalMatching.bruening2021StructurallyIdentical) requires the antecedent and the elided constituent to have the same filtered max-projection path lists (up to permutation).

                                                                                                In the C&G-Y CS bulk-sharing structure, the elided TP literally is the antecedent TP — one node in the multidominance graph with two parents. So the filtered max-projection paths are physically identical, and List.Perm.refl discharges the identity condition.

                                                                                                The C&G-Y CS bulk-sharing structure is robust to identity-condition choice: it trivially satisfies @cite{bruening-2021}'s max-projection identity by List.Perm.refl on identical path lists, for any HeadFunction parameterizing the projection paths.

                                                                                                This is a positive cross-framework consilience: the C&G-Y MD architecture and the Bruening 2021 identity condition agree on CS, so the paper's predictions for CS sluicing licensing do not depend on identity-condition choice.