Documentation

Linglib.Phenomena.Morphology.Studies.Wood2023

@cite{wood-2023} — Icelandic Nominalizations and Allosemy #

@cite{wood-2023} @cite{wood-2015} @cite{wood-marantz-2017}

Icelandic Nominalizations and Allosemy. Oxford University Press. DOI: 10.1093/oso/9780198865155.001.0001

Overview #

@cite{wood-2023} argues that Icelandic deverbal nominalizations are built on the structure [nP n [vP v √ROOT]] (a complex head, NOT a phrasal VoiceP), and that the ambiguity between CEN, SEN, and RN readings arises from allosemy of v and n — one syntactic terminal with multiple context-dependent meanings:

Key Claims Formalized #

  1. No Voice in nominalizations (Ch. 3, Ch. 5): The external argument is introduced by a Poss head (= i* from @cite{wood-marantz-2017}), NOT by Voice. Voice diagnostics in nominalizations really test for agentive semantics, which Poss can also provide.

  2. Borer's Generalization (Ch. 5 §5.1.5): CEN reading entails the existence of a morphologically related verb with the same meaning. This follows from the architecture: CEN requires v, and n cannot trigger root suppletion past v.

  3. P-prefixing patterns (Ch. 4): Three patterns of preposition-verb interaction in nominalizations, depending on whether P conditions root meaning.

  4. marg- and endur- diagnostics (Ch. 6): Iterative marg- 'many-' is only compatible with CEN; repetitive endur- 're-' is compatible with CEN, SEN, and result/product RN, but not simple entity RN.

  5. -vaeða verbs always compositional (Ch. 6 §6.5): Because -vaeða is a compound (√VAEÐA adjoins to v), the root cannot interact idiosyncratically with n past v. Therefore -vaeðing nominals never have idiosyncratic RN readings.

Wood's reading derivation: v and n alloseme combinations.

@cite{wood-2023} Ch. 5 (5.4a–e):

  • v eventive + n zero → CEN (noun = verb meaning)
  • v zero + n simpleEvent → SEN (event-entity reading)
  • v zero + n entity → simple entity (entity reading)
  • v eventive + n result → result/product (entity from event)

Whether a nominalization has Voice (it doesn't, per @cite{wood-2023}).

@cite{wood-2023} Ch. 5 §5.1.3: "I will assume, as discussed in Chapter 3, that there is in fact no Voice head in the structure." The external argument is introduced by Poss (= i*), not Voice.

Equations
Instances For

    Poss head semantics: parallel to Voice but for nominals. @cite{wood-2023} Ch. 5 (5.22): ⟦Poss⟧ ↔ λxλe. agent(x)(e) / __ agentive nP

    @cite{wood-marantz-2017}: Voice and Poss are the same head i*, appearing in different categories (vP vs nP).

    Instances For
      @[implicit_reducible]
      Equations
      Equations
      • One or more equations did not get rendered due to their size.
      Instances For
        def Wood2023.possReading (nPisCEN : Bool) :

        Poss gets agent reading only with agentive (CEN) nP. @cite{wood-2023} Ch. 5 (5.24): "i* ↔ λxλe. agent(x)(e) / __ (agentive event)"

        Equations
        Instances For

          Three patterns of preposition-verb interaction in nominalizations.

          @cite{wood-2023} Ch. 4:

          • Pattern 1: P conditions root meaning, must be prefixed, can also appear as complement PP (ráða umumráðun á)
          • Pattern 2: P conditions root meaning, must be prefixed, cannot be doubled (gera viðviðgerð á, not *viðgerð við)
          • Pattern 3: P does NOT condition root meaning, is not prefixed (hugsa umhugsun um, not umhugsun)
          Instances For
            @[implicit_reducible]
            Equations
            Equations
            • One or more equations did not get rendered due to their size.
            Instances For

              Pattern assignment for fragment nominalizations.

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

                Verbal prefixes that diagnose nominalization readings.

                @cite{wood-2023} Ch. 6 §6.4:

                • marg- 'many-' adds iterativity to the event. Only compatible with CEN, because only CEN has an event variable at the v level.
                • endur- 're-' adds presupposition of prior eventuality. Compatible with CEN, SEN, and result/product RN (all have event variables), but NOT with simple entity RN or simple state (no event variable). Per (6.46)–(6.53): endurprentun 'reprint' (result RN) is OK, but endur-þvottur 'laundry' (simple entity) is not.
                Instances For
                  @[implicit_reducible]
                  Equations
                  Equations
                  • One or more equations did not get rendered due to their size.
                  Instances For

                    Borer's Generalization: CEN reading entails the existence of a morphologically related verb with the same meaning.

                    @cite{wood-2023} Ch. 5 §5.1.5: This follows from two assumptions: (a) verbs are semantically special (they introduce event variables), (b) n cannot trigger root suppletion past v.

                    Equations
                    Instances For

                      All CEN-capable nominals in the fragment have base verbs (Borer's Generalization holds).

                      -vaeða verbs are compounds: √VAEÐA adjoins directly to v. Because √VAEÐA is meaningless (like English do-support √DO), the root it compounds with must be categorized (n) first.

                      @cite{wood-2023} Ch. 6 (6.60): [v [n ... √ROOT n] [v √VAEÐA v]]

                      This structure entails:

                      • Root cannot idiosyncratically select complement PPs
                      • -vaeðing nominals never have idiosyncratic RN readings
                      • PP complements of -vaeða verbs are always compositional
                      • -vaeða verbs cannot select ApplP
                      • idiosyncraticPP : Bool

                        Root can condition idiosyncratic complement PP?

                      • idiosyncraticRN : Bool

                        Nominalization can have idiosyncratic RN reading?

                      • selectsApplP : Bool

                        Verb can select ApplP?

                      • alwaysCompositional : Bool

                        Meaning of nominalization always compositional?

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

                            -vaeða verbs have maximally restricted properties.

                            Equations
                            • Wood2023.vaedaProps = { idiosyncraticPP := false, idiosyncraticRN := false, selectsApplP := false, alwaysCompositional := true }
                            Instances For

                              All -vaeða restrictions hold simultaneously.

                              opnun 'opening' connects to opnast 'open-ST' (anticausative). The nominalization is built on the same root as the -st verb; the -st voice morphology does not appear in the nominal (nominalizations lack Voice). The Wood-2015 stType for opnast is sourced from opnast_info in the Wood2015 study file (the Fragment carries only consensus lexical data).

                              Anticausative -st verbs can be nominalized: the nominalization lacks Voice (hence no -st), but retains the root's meaning. @cite{wood-2023} Ch. 3: -st and nominalization both require non-agentive contexts, but the nominal achieves this by lacking Voice entirely rather than having non-agentive Voice.

                              All nominalizing suffixes spell out the same head n. Different suffixes do NOT indicate different functional heads — this is allomorphy of n, not different morphemes. @cite{wood-2023} Ch. 2 (2.1), Ch. 3.

                              The same suffix (-un) can yield different readings: opnun has CEN + simple entity, notkun has CEN only. The reading is determined by allosemy, not by the suffix.

                              Wood vs Panagiotidis on the n head #

                              @cite{wood-2023} Ch. 1 §1.2.3 + Ch. 5: nominalisations are systematically ambiguous (CEN/SEN/RN/SimpleState/SimpleEntity). All readings stem from ONE structure; the interpretive variation comes from n (and v) having multiple allosemesn can be zero / sortal / relational / alienator / content / simpleEvent / result / state / entity (9 cases per Theories/Morphology/DM/Allosemy.lean).

                              @cite{panagiotidis-2015} treats n as a uniform categoriser bearing the interpretable feature [N]. There is no alloseme machinery in Panagiotidis: interpretation of an n-headed projection follows from [N] (sortal perspective) plus what the complement contributes. Per p. 95, "categorizers are not functional" and per Ch. 4 "categorial features [N] and [V] are LF-interpretable" — features, not allosemes, do the interpretive work.

                              The two frameworks are incommensurable on n's interpretive contribution: Wood says context-determined alloseme choice; Panagiotidis says uniform [N] feature. Applied to @cite{mcnally-deswart-2011}'s InflectedAnalysis rivals (which all involve some n head), they diverge on what additional commitment is required.

                              This is the cross-framework divergence the @cite{mcnally-deswart-2011} bridge in Panagiotidis2015.lean (namespace MdSBridge) noted as open — addressed here.

                              Wood's framework requires every n-headed nominalisation to commit to an NAlloseme (one of the 9 cases). Panagiotidis's framework requires no such commitment — n is uniformly [N].

                              For the @cite{mcnally-deswart-2011} hetAsCap analysis (where M&deS posit a trope interpretation, an entity correlate of a property uniquely instantiated in one bearer per @cite{moltmann-2004}), Wood's framework would need to specify which alloseme het selects. None of the 9 NAlloseme cases is "trope":

                              • relational, sortal, alienator, content — semantically wrong type (relational arguments / kind sortation / possessor closure / CP-complement-selection)
                              • zero — identity function, would inherit the AP's adjectival meaning verbatim, missing the trope-reification
                              • simpleEvent, result, state — event-nominalisation allosemes, conceptually for V→N transitions
                              • entity — closest fit, but M&deS distinguish tropes from ordinary entities (§3.4 + cite of @cite{moltmann-2004} ontology)

                              So Wood's framework would require extending NAlloseme to model M&deS's trope analysis. Panagiotidis's framework would not.

                              Equations
                              Instances For

                                The Panagiotidis-side prediction: for any rival, what does Panagiotidis say n does interpretively? Answer (from referential_predicative_asymmetry): bears [N], makes the projection referential. No alloseme choice required.

                                Equations
                                Instances For

                                  The Wood-side requirement: each rival must specify an alloseme, or the framework must be extended to cover it.

                                  Equations
                                  Instances For

                                    The substantive divergence. Wood's framework requires an alloseme commitment for every rival; Panagiotidis's framework requires none. The two frameworks make incommensurable demands on a theory of het rode van X (and of nominalisations in general).

                                    The substantive gap on hetAsCap. Wood's NAlloseme inventory has no case obviously fitting M&deS's trope analysis. Wood's framework would need extension; Panagiotidis's framework would not. Concrete: woodAllosemeForRival .hetAsCap = none, recording the gap.

                                    The 9 NAllosemes are exactly: relational, sortal, alienator, content, zero, simpleEvent, result, state, entity. None of these is "trope". The bridge documents this as a substantive limitation of Wood's inventory when applied to M&deS's adjectival nominalisation data.

                                    Three-way framework dialogue #

                                    The Subkinds substrate now anchors a three-way cross-framework dialogue on Dutch nominalisation morphology:

                                    Each framework's distinctive theoretical primitives generate distinct empirical commitments on the same Dutch data — exactly the kind of cross-framework incompatibility linglib is designed to surface.