Documentation

Linglib.Phenomena.Definiteness.Studies.Jenks2018

Jenks (2018): Articulated Definiteness without Articles #

@cite{jenks-2018}

Mandarin distinguishes unique definites (bare N) from anaphoric definites (Dem-Clf-N), with a documented exception for matrix subjects (§5.3). A Chierchia-style ι type-shift handles the bare-N route (@cite{chierchia-1998}, @cite{yang-2001}, @cite{dayal-2004}, @cite{jiang-2012}); demonstratives carry an indexical argument supplying ι^x (a Schwarz-style strong article); an Index! principle (Heim 1990 Maximize Presupposition + @cite{schlenker-2012} Gricean reduction) selects between them.

The substrate already operationalizes the inventory layer: Core.Nominal.NominalKind.{unique,anaphoric} are ι and ι^x; Core.Nominal.ArticleInventory.toMarkingStrategy derives the four-cell typology directly named after Jenks 2018 in Features.Definiteness. The Mandarin Fragment commits articleInventory_marking := .markedAnaphoric. This file focuses on what is distinctly Jenks: the typological prediction (§3 + §6), the bare/Dem competition (§5 paper), the covarying-readings argument for index binding (§4.3 paper), and the subject-as-topic exception (§5.3 paper). The post-Jenks Shan refutation lives in Moroney2021.lean per chronology discipline.

Jenks 2018 §6.2 (Table 2) lays out a four-cell typology of definiteness marking. The cells in linglib substrate vocabulary:

InventoryStrategyLanguages
both forms, distinct.bipartiteGerman, Lakhota
only anaphoric form.markedAnaphoricMandarin, Akan, Wu
both forms, syncretic.generallyMarkedCantonese, English
only unique form(unattested)

Jenks proposes the unattested fourth cell as a typological gap and gives it a historical explanation (Greenberg 1978: definite articles typically grammaticalize from demonstratives and so first appear in anaphoric contexts).

The three Jenks-attested marking strategies (Table 2). Lifted to the substrate level here so consumers (e.g. Moroney's refutation) can import the prediction without reciting the list.

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

    Mandarin's article inventory derives .markedAnaphoric — its Jenks (2018) Table 2 cell.

    Cantonese's article inventory derives .generallyMarked — paper §6 (Table 1, Table 2): [Clf-N] is an ambiguous definite article like English the, covering both unique and anaphoric environments.

    Mandarin and Cantonese instantiate distinct Jenks cells — the central typological contrast of paper §6.

    Bare N is universally licensed by the article inventory; in Mandarin it serves unique definites (paper §3.1 examples 10–11: yueliang sheng shang lai le 'the moon has risen'; Hufei he-wan-le tang 'Hufei finished the soup'; Gou yao guo malu 'the dog wants to cross the road').

    The anaphoric kind (.anaphoric R d) is licensed in Mandarin via the demonstrative paradigm (paper §3.2: anaphoric definites surface as Dem-Clf-N constructions). The licensing proceeds through the left disjunct of licensesKind .anaphoric.

    Mandarin demonstratives are licensed (the na/zhe paradigm — paper fn. 8: speakers prefer na 'that' to zhe 'this' in most simple anaphoric environments).

    A Mandarin bare definite and its .unique counterpart over the same restrictor pick the same referent — the bare-N route to unique definiteness (paper §4.1: ι via Chierchia type-shift) is extensionally the NominalKind.unique denotation at the API layer. Parallels Moroney2021.shan_bare_unique_agreement.

    A Mandarin demonstrative-marked anaphoric and the bare anaphoric over the same restrictor and discourse index pick the same entity: the deictic feature is a presupposition filter, not a referent selector (paper §4.2 cited Hanink/Schwarz analysis). Parallels Moroney2021.shan_demonstrative_anaphoric_agreement.

    Paper §3.1 (Mandarin): part-whole bridging takes bare N (chezi … paizhao 'car … license plate'); producer-product bridging takes Dem-Clf-N (shi … #(na wei) shiren 'poem … #(that) poet'). Paper §3.3 (Mandarin donkey): demonstratives required in both ruguo/dou-conditionals (example 18) and relative-clause donkey configurations (example 20: Mei ge … #(na zhi) shuiniu); bare N is infelicitous (example 19). The substrate's Features.Definiteness.bridgingPresupType and useTypeToPresupType encode these splits at the use-type level — proved here for Mandarin by re-using Schwarz2009.lean lemmas.

    Producer-product bridging projects familiarity (paper §3.1 example 14b: shi … #(na wei) shiren — Dem-Clf-N realization).

    Donkey anaphora projects familiarity at the use-type layer (paper §3.3; @cite{schwarz-2009} §3). For Mandarin's .markedAnaphoric strategy this predicts demonstrative use in both donkey sub-configurations.

    Paper §3.3 distinguishes two donkey environments: bare conditionals (use indeterminate pronouns only — no def expression) and ruguo/dou-conditionals plus relative-clause donkey (definite expression required, instantiated as Dem-Clf-N). @cite{cheng-huang-1996} originally observed the contrast. The use-type collapse to .familiarity covers the second environment (the one where definite expressions are licensed); bare conditionals are out of the scope of useTypeToPresupType because they involve no definite at all.

    Paper p. 514 introduces the Blocking Principle (eq. 23: "Don't do covertly what you can do overtly"), inherited from @cite{chierchia-1998}. In Mandarin, the absence of an overt unique article means ι is unblocked — bare N can route to a unique-definite reading via the type-shift hierarchy of @cite{dayal-2004}.

    The substrate has Dayal's hierarchy (Theories/Semantics/Kinds/MeaningPreservation.lean), already used by Moroney2021.lean. The Mandarin parallel theorem is the same selectShift instance: when no shift is blocked, ι is selected first (Meaning Preservation).

    theorem Phenomena.Definiteness.Studies.Jenks2018.iota_is_last_resort :
    Semantics.Kinds.MeaningPreservation.selectShift { number := Semantics.Kinds.MeaningPreservation.NumberFeature.neutral, downDefined := false, iotaBlocked := false, iotaAnaphoricBlocked := false, existsBlocked := false, instantiationAccessible := true } = some Semantics.Kinds.MeaningPreservation.TypeShift.iota

    Mandarin's number-neutral nouns under no blocking select ι as the preferred type-shift — paper p. 514's "ι is the type-shifter relevant for definite type-shifting", lifted to the Dayal substrate Moroney2021 already consumes.

    Bridge to Mandarin's classifier strategy: per Phenomena.Classifiers.Studies.NMP.mandarinStrategy, Mandarin's classifier denotation atomizes the noun (CLF-for-N). This is the Trinh 2011 / Krifka 1995 / Chierchia 1998 denotation Jenks adopts in §4.1 (eq. 21). Parallels Moroney2021.shan_clf_is_atomization.

    Jenks (2018, p. 513) is explicit that his anaphoric article ι^x takes an index argument of type ⟨e,t⟩ (a property), departing from @cite{schwarz-2009}/@cite{schwarz-2013}'s type ⟨e⟩ (an individual). The substrate's NominalKind.anaphoric R d carries d : Nat — a discourse-index slot resolved through the entity assignment, which matches Schwarz's individual-typed index, not Jenks's property-typed one. For Schwarz-style and ordinary demonstrative cases this divergence is inert (the assignment returns an entity), but the property-typed index is load-bearing in paper §4.4 (Pred type-shift, examples 32–38: proper names + demonstratives composing as Pred(Zhangsan) + Dem-Clf-N).

    Faithfully formalizing §4.4 requires a property-typed-index variant on NominalKind.anaphoric. This is recorded as a TODO at the substrate level (Core/Nominal/Description.lean) rather than encoded as a placeholder theorem.

    Paper p. 524 (eq. 50): Index! = "Represent and bind all possible indices". Jenks derives this from Heim 1990 Maximize Presupposition via Schlenker 2012's Gricean reduction. The substrate provides Theories.Semantics.Presupposition.MaximizePresupposition.mpConstraintOf, parametric over an arbitrary candidate type and strength function — the natural slot for any MP-instance.

    The IndexCandidate carrier below is a minimal 2-bit witness type sufficient to demonstrate the principle's qualitative behavior (prefer indexed when index is available; neutral otherwise). A fuller instantiation would parameterize over NominalKind F and the discourse-context predicate licensing the index — that refactor belongs in a substrate file (Theories/Semantics/Presupposition/Index.lean) when a second consumer needs it.

    Index! candidate: an indexed alternative is in the running only when an index can be supplied (paper p. 523-524: "an index is required to be licensed by explicit prior mention in discourse").

    • isIndexed : Bool
    • indexAvailable : Bool
    Instances For
      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

          Index! strength function: an indexed candidate gets strength 1 only when an index can actually be supplied (paper's prior-mention condition). Bare candidates always get strength 0.

          Equations
          Instances For

            Index! as MP: the principle is mpConstraintOf 1 indexStrength — the substrate's general MP construction at strength 1, applied to the binary indexed/non-indexed competition. Per paper p. 524, "Index! is a specific instance of Maximize Presupposition! (Heim 1990)".

            Equations
            Instances For
              theorem Phenomena.Definiteness.Studies.Jenks2018.index_prefers_indexed_when_available :
              indexConstraint.eval { isIndexed := true, indexAvailable := true } < indexConstraint.eval { isIndexed := false, indexAvailable := true }

              When both candidates are available and an index can be supplied (a discourse antecedent exists), the indexed candidate has fewer Index! violations than the bare candidate.

              theorem Phenomena.Definiteness.Studies.Jenks2018.index_neutral_when_unavailable :
              indexConstraint.eval { isIndexed := true, indexAvailable := false } = indexConstraint.eval { isIndexed := false, indexAvailable := false }

              When no index can be supplied (no discourse antecedent), Index! is neutral — both candidates incur the same number of violations, correctly leaving bare N as the only option in unique-definite contexts (paper §3.1; explains why ι^x is unavailable there).

              Paper §5.3 documents the exception to Index!'s prediction: in matrix subject position, bare N is licensed for anaphoric reference (not just unique). Examples 51–53 establish that the exception is pragmatic — bare N in subject position marks the noun as a continuing topic (in the sense of @cite{roberts-2003} QUD-relativized topics), which short-circuits Index! because the topic-marking pragmatic function takes precedence.

              Two empirical points the paper makes (p. 524-526):

              The substrate has Roberts2012 QUD machinery in Phenomena/Discourse/Strategy/QUDStack.lean (per memory: project_qud_dissolution.md). A faithful formalization of paper §5.3 would need a topic predicate over NominalKind configurations co-occurring with QUD-stack state — substrate the linglib has but that this study file does not yet plug into.

              Stated below as a sorry'd theorem to mark the analytical commitment without forcing the discharge.

              A topic-aware Index! candidate carries the additional isTopic flag (continuing-topic status under Roberts QUD).

              • isIndexed : Bool
              • indexAvailable : Bool
              • isTopic : Bool
              Instances For
                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

                    Topic-overridden Index! strength: the topic-marking pragmatic function (paper §5.3) neutralizes Index!'s preference. A bare candidate marked as a continuing topic gets the same strength as an indexed candidate when an index is available.

                    Equations
                    Instances For

                      The topic-aware Index! constraint.

                      Equations
                      • One or more equations did not get rendered due to their size.
                      Instances For
                        theorem Phenomena.Definiteness.Studies.Jenks2018.subject_topic_overrides_index :
                        topicAwareIndexConstraint.eval { isIndexed := false, indexAvailable := true, isTopic := true } = topicAwareIndexConstraint.eval { isIndexed := true, indexAvailable := true, isTopic := true }

                        Subject-position exception (paper §5.3): when a bare-N candidate is marked as a continuing topic, Index!'s preference for the indexed alternative is neutralized. Both candidates incur the same number of violations, restoring the apparent free variation paper §3.2 documents for matrix subjects.

                        Decision-procedure proof over the four relevant TopicCandidate configurations.

                        theorem Phenomena.Definiteness.Studies.Jenks2018.non_topic_keeps_index_preference :
                        topicAwareIndexConstraint.eval { isIndexed := true, indexAvailable := true, isTopic := false } < topicAwareIndexConstraint.eval { isIndexed := false, indexAvailable := true, isTopic := false }

                        Without topic marking, the original Index! preference holds (paper §5.2): the indexed candidate has strictly fewer violations than the bare candidate when an index is available.

                        Paper §4.3 (examples 27–30) is the empirical anchor for treating ι^x's index as a bound variable: bare N can covary with a quantificational topic, demonstratives in the same configuration cannot. The covarying part-whole bridging (example 30: Mei ge mai le fangzi de ren dou xuyao xiuli #(na ge) wuding 'every house-buyer needed to fix the/#that roof') is the load-bearing case — the bare restrictor's resource situation can vary with the topic adverb's binding, while the demonstrative's index forces a strict reading.

                        The substrate has Hanink-style resource-situation binding in Hanink2021.lean (tableAtSit0 over a Room index, with gsKitchen /gsLiving as situation assignments). A Mandarin parallel theorem would instantiate that pattern with a Mandarin restrictor and a quantificational topic situation; left as a sorry pending the property-typed-index substrate gap (the demonstrative side of the contrast requires the §5 ⟨e,t⟩-typed index to forbid covariation properly).

                        Strict reading of demonstratives across situation variation: a .demonstrative/.anaphoric description's referent is fixed by the entity assignment g at index d, independent of the situation assignment gs. Concretely: if the predicate R is itself invariant across two situation assignments at the indexed entity, then both demonstratives return the same referent — the demonstrative cannot covary through the situation slot.

                        This is one half of paper §4.3's covariation contrast (the strict half). The other half — bare N covarying via situation binding — requires the property-typed index variant on NominalKind.anaphoric flagged in §5 to express cleanly, and is deferred.