Documentation

Linglib.Theories.Dialogue.KOS.RepriseContent

The Reprise Content Hypothesis (RCH) #

@cite{purver-ginzburg-2004}, @cite{ginzburg-2012}

The Reprise Content Hypothesis is a methodological criterion linking fragment-reprise data to denotational adequacy. It comes in two strengths (@cite{ginzburg-2012} ex. 98):

RCH operates per reading type — fragment reprises admit multiple readings (Clausal Confirmation, Intended Content, Repetition, Correction; @cite{ginzburg-2012} Ch. 6 §6.2.1, Table 6.1). A reprisedContent operation returns the type at which each licensed query operates, parameterized by reading.

Architectural role #

RCH is not a load-bearing constraint inside the dialogue grammar — nothing in KOS needs RCH to function. It is a meta-theoretic predicate over denotation assignments: given a function from fragments to LocProps, RCH asks whether the q-params structure on each LocProp licenses exactly (or at least) the queries that a reprise empirically generates. RCH is therefore a Prop predicate over assignment functions; specific denotation proposals become theorems about whether they satisfy or violate it.

What this file provides #

The GQ-violation theorem and the q-params/dgb-params satisfaction theorem live in Phenomena/Dialogue/Studies/PurverGinzburg2004.lean, which consumes this file.

The four reading types for a fragment reprise.

@cite{ginzburg-2012} Ch. 6 §6.2.1: empirical CR-corpus work establishes that any single reprise fragment admits up to four readings, distinguished by what they query.

This enum mirrors the CRReading enum in Phenomena/Dialogue/Studies/Ginzburg2012.lean. We declare it here so that KOS/RepriseContent.lean does not depend downstream on any Phenomena file.

  • clausalConfirmation : RFReading

    "Are you asking/saying that p?" — confirms propositional content

  • intendedContent : RFReading

    "What do you mean by X?" — requests intended referent / predicate

  • repetition : RFReading

    "Can you repeat X?" — requests phonological repetition

  • correction : RFReading

    "Did you say X or Y?" — corrective alternative

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

      The semantic type at which a fragment-reprise query operates.

      @cite{ginzburg-2012} §8.5.1's argument against generalized-quantifier denotations turns on a type-mismatch: GQ denotations predict reprise queries at functional type (e → t) → t, while the empirical record shows reprises operating only at individual / property type. We reify the relevant type distinctions as a finite enum so that the GQ-violation theorem in PurverGinzburg2004.lean can be a structural disagreement between predicted and observed QueryType sets, not a numerical mismatch on probability values.

      • individual : QueryType

        Query at type e — "what individual?" (witness for an IND-typed q-param)

      • property : StringQueryType

        Query at type e → t — "what property/restriction?"

      • functional : QueryType

        Query at type (e → t) → t — generalized-quantifier-typed

      • repetition : QueryType

        Query at the phonological form — "what string?"

      Instances For
        Equations
        • One or more equations did not get rendered due to their size.
        Instances For
          def Dialogue.KOS.instDecidableEqQueryType.decEq (x✝ x✝¹ : QueryType) :
          Decidable (x✝ = x✝¹)
          Equations
          Instances For
            def Dialogue.KOS.reprisedContent {Cont : Type} (host : LocProp Cont) (sub : SubUtterance) :

            Query types observed for a reprise of sub in host LocProp under reading.

            @cite{ginzburg-2012} Ch. 6 §6.2.1, Table 6.1 + §8.5.1:

            • repetition queries the phonological form (always .repetition).
            • intendedContent queries the q-params record of the sub-utterance — one .individual query per index, plus a .property query per restrictor. When the host's qcparams is empty (referential NP), the only intended- content query is a .property over the sub-utterance's content string.
            • clausalConfirmation confirms propositional content — a single .individual query about the witness.
            • correction is ad-hoc and not characterized in §8.5; we return [].
            Equations
            Instances For
              structure Dialogue.KOS.RepriseEvent (Cont : Type) :

              A reprise scenario: a host utterance, a sub-utterance fragment within it, and a reading type.

              Instances For
                @[reducible, inline]

                A predictor maps a reprise event to the query types its denotation theory predicts should be licensed. Different denotation proposals (GQ, q-params split, HOU, ...) yield different predictors; RCH judges them.

                Equations
                Instances For

                  The q-params/dgb-params predictor (@cite{ginzburg-2012} §8.5.1).

                  For each reprise event, this predictor licenses exactly the query types that the q-params record on the host's LocProp would expose — namely .individual (witness) plus a .property per restrictor — plus the reading-specific queries (.repetition for phonological echoes).

                  Equations
                  Instances For
                    def Dialogue.KOS.WeakRCH {Cont : Type} (predict : RchPredictor Cont) :

                    Weak RCH (@cite{ginzburg-2012} ex. 98a, p. 323): every observed reprise query type is predicted by the denotation theory.

                    Read as: the theory does not under-generate. It may predict more than is observed, but it covers everything that is observed.

                    Equations
                    Instances For
                      def Dialogue.KOS.StrongRCH {Cont : Type} (predict : RchPredictor Cont) :

                      Strong RCH (@cite{ginzburg-2012} ex. 98b, p. 323): predicted and observed query type sets coincide.

                      Read as: the theory neither under- nor over-generates. Strong RCH is too strong to hold of arbitrary fragments — Ginzburg himself documents exceptions (the multi-reading phenomenon in Table 6.1). It is included for completeness because Ginzburg states both versions. Theorems about it are typically refutations.

                      Equations
                      Instances For
                        theorem Dialogue.KOS.strongRCH_implies_weakRCH {Cont : Type} (predict : RchPredictor Cont) (h : StrongRCH predict) :
                        WeakRCH predict

                        Strong RCH implies Weak RCH — the predicted set is a superset of the observed set under either reading.

                        The q-params predictor satisfies Weak RCH by construction: it predicts exactly what reprisedContent reports.

                        The q-params predictor satisfies Strong RCH for the same reason.