Documentation

Linglib.Phenomena.Ellipsis.Studies.BenzSalzmann2025

Benz & Salzmann (2025): Evidence from German for N-stranding NP-ellipsis #

@cite{benz-salzmann-2025}

Evidence from German for N-stranding NP-ellipsis. Proceedings of the Chicago Linguistic Society 60.

Overview #

German has N-stranding NP-ellipsis: ellipsis of NP after N-to-n head movement, which strands the moved noun outside the deletion domain. What looks like deletion of individual constituents (PPs, relative clauses, genitives) is actually NP-ellipsis with N having evacuated.

Key Claims #

  1. X-stranding diagnostic (@cite{liptak-saab-2014}): if a language has both X-movement out of XP and XP-ellipsis, X-stranding XP-ellipsis should exist. German has N-to-n movement and NP-ellipsis → N-stranding NP-ellipsis exists.

  2. Variable [E] placement: [E] can be on D, Num, or n, producing different-sized deletion domains.

  3. Contrast condition: n[E] (N-stranding) requires the noun to be contrastive. When N is not contrastive, [E] must be on a higher head (Num or D), and N is deleted along with the rest.

  4. Against individual deletion: Apparent deletion of individual PPs or relatives is actually NP-ellipsis after N-movement — there is no mechanism for deleting arbitrary individual constituents within DP. No [E] position can delete prenominal modifiers without also deleting N.

  5. Against pragmatic recovery: Bound variable readings inside elided material prove these are genuine syntactic ellipsis, not pragmatic inference.

  6. Gender mismatch parallel: N-stranding tolerates gender mismatches (n hosts gender, n is external under n[E]) — paralleling voice mismatch tolerance in VPE (Voice external under Voice[E]).

Architecture #

This study imports DeletionDomain.lean: the generic DeletionSpine class, the clausal SpinePos instance, and the nominal NomSpinePos instance with NomEllipsisType and the xStranding theorem.

N-to-n head movement is assumed but not formally instantiated; the deletion-domain predictions are independent of the movement mechanism.

The core predictions are derived from the NomSpinePos deletion domain:

All already proved in DeletionDomain.lean §§ 11-12. This file adds the German empirical data and verifies it against those predictions.

Grammaticality judgment for a DP-ellipsis example.

  • sentence : String

    German sentence

  • translation : String

    English translation

  • grammatical : Bool

    Is the ellipsis grammatical?

  • Which [E] position licenses (or fails to license) this?

  • nounContrastive : Bool

    Is the noun contrastive?

  • source : String

    Source example number

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

      ex. (11): Deletion of more than just the noun — NP-ellipsis deletes N + PP modifier.

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

        ex. (12): PP complement of elided noun survives — the elided noun Angst selects vor, showing syntactic structure in the ellipsis site.

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

          ex. (6a): PP der Physik recovered in second conjunct — noun is contrastive (Studenten vs Professoren), enabling N-stranding. The PP dependent of the elided NP is interpreted in the ellipsis site.

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

            ex. (5): Non-contrastive noun blocks N-stranding (Spanish; @cite{liptak-saab-2014}) — only the numeral contrasts (tres vs dos), not the noun estudiantes. Since N is not contrastive, n[E] is unavailable. [E] must be on Num or higher, deleting N and its PP dependent. The PP de física cannot be recovered.

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

              ex. (27): N not contrastive but adjective contrastive → N-stranding blocked. Zerstörung is the same noun in both conjuncts; only the ordinal (erste vs zweite) contrasts. Since N is not contrastive, [E] must be on a higher head, deleting N along with the genitive argument Roms. The genitive cannot be recovered.

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

                ex. (25a): Prenominal adjective cannot be individually deleted — schönste cannot be elided while N (Auto/Motorrad) contrasts. No [E] position achieves this: n[E] leaves adj external; Num[E] and D[E] delete both adj and N. See no_individual_prenominal_deletion.

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

                  ex. (25b): Numeral cannot be individually deleted — zwei cannot be elided while N (Bücher/Romane) contrasts. D[E] is the only position that could delete Num, but it also deletes N, adj, and n.

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

                    ex. (26a): Prenominal genitive (in Spec,DP) cannot be deleted and recovered. Even D[E] (the highest [E] position) deletes only D's complement — Spec,DP is never in any deletion domain.

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

                      ex. (26b): Postnominal von-PP CAN be deleted — this IS NP-ellipsis after N-stranding, not individual constituent deletion.

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

                        ex. (28a): One PP (von jedem Studenten) remains while the second (vor seinem Professor) is attempted to be recovered — UNGRAMMATICAL. If one postnominal modifier survives, NP-ellipsis did not apply, and the second modifier cannot be individually deleted.

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

                          ex. (28b): BOTH PPs deleted — GRAMMATICAL. This is genuine N-stranding NP-ellipsis: Angst vs Respekt are contrastive, n[E] applies, and the entire NP (both PPs) is deleted. The bound pronoun seinem (bound by jedem Studenten) in the elided material proves genuine syntactic ellipsis — pragmatic recovery cannot bind variables.

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

                            The contrast condition on [E] placement: n[E] (N-stranding) is available only when the noun is contrastive. When N is not contrastive, [E] must be on a higher head (Num or D), causing N to be deleted too.

                            This predicts:

                            • Contrastive N + postnominal modifier → N-stranding OK (ex. 6a)
                            • Non-contrastive N + postnominal modifier → modifier not recovered (ex. 5, 27)
                            Equations
                            Instances For

                              When N is contrastive, [E] is on n. The n head (to which N has moved) is external → the noun survives via its moved copy at n.

                              When N is not contrastive, [E] is on Num. Both N (base position) and n are in the deletion domain — the noun is deleted along with all NP-internal and nP-internal material.

                              The central asymmetry (@cite{benz-salzmann-2025} §2.3):

                              • Prenominal modifiers (adjectives, numerals, prenominal genitives) CANNOT be individually deleted → they are above n
                              • Postnominal modifiers (PPs, relative clauses, postnominal genitives) CAN be deleted under N-stranding → they are inside NP (below n)

                              This follows directly from the NomSpinePos deletion domain: under n[E], only material below n (= NP-internal) is deleted.

                              Individual prenominal deletion is impossible: any [E] position that puts the prenominal adjective in the deletion domain also puts N in the deletion domain. So you cannot delete an adjective without also deleting the noun — ruling out individual adjective deletion when N is contrastive (ex. 25a).

                              This follows from NP_adj being structurally above N: N is below NP_adj in the complement hierarchy, so any complement that contains NP_adj also contains N.

                              Individual numeral deletion is impossible: any [E] position that puts Num in the deletion domain also puts N in the deletion domain. Only D[E] deletes Num, and D[E] also deletes everything else (ex. 25b).

                              Under N-stranding, if one postnominal modifier remains, ALL must remain (ex. 28a vs 28b). Ellipsis targets the entire NP projection, not individual constituents.

                              If [E] is NOT on n (no N-stranding), then [E] must be higher (Num or D), and all postnominal material plus N is deleted — so individual PP recovery is impossible. The only way to recover a PP is N-stranding, which deletes the entire NP.

                              German N-stranding (n[E]) and Spanish Num[E] are different [E] placements in the same nominal spine. German n[E] produces a strictly smaller deletion domain than Spanish Num[E].

                              This is the nominal analogue of the clausal vVPE vs. English VPE distinction: lower [E] → smaller deletion domain → more survivors.

                              Monotonicity: German N-stranding tolerates more "mismatches" (i.e., more positions survive) than Spanish nP-ellipsis, because n < Num. Adjectives survive German n[E] but not Spanish Num[E].

                              Gender mismatches under N-stranding parallel voice mismatches under VPE. Both involve the categorizer head being external to the deletion domain when [E] is at that categorizer:

                              • N-stranding (n[E]): n hosts gender features, n is external → gender mismatches tolerated
                              • English VPE (Voice[E]): Voice hosts voice features, Voice is external → voice mismatches tolerated

                              Conversely, when [E] is higher:

                              • nP-ellipsis (Num[E]): n is internal → gender mismatches blocked
                              • Sluicing (C[E]): Voice is internal → voice mismatches blocked

                              @cite{benz-salzmann-2025} p.62-63: "while gender mismatches are normally not possible in nominal ellipsis [...], such mismatches do become possible in our N-stranding NP-ellipsis data."