Documentation

Linglib.Studies.GinzburgCooper2004

Ginzburg & Cooper (2004): Clarification, Ellipsis, and Contextual Updates #

[GC04]

Formalization of the core running example from [GC04]:

A: "Did Bo leave?" B: "Bo?"

This study applies the KOS framework (DGB, IS, C-PARAMS, coercion operations) to derive both CE readings — clausal and constituent — from the same antecedent sign, and demonstrates the speaker/addressee IS asymmetry.

Key Claims Formalized #

  1. Proper names introduce C-PARAMS (referent binding) — ex. 28
  2. The running example has 5 C-PARAMS — ex. 32
  3. Parameter focussing yields clausal CE reading — ex. 53/54
  4. Parameter identification yields constituent CE reading — ex. 59/60
  5. Both coercions take the antecedent sign and target the same SAL-UTT
  6. Existential generalization removes a parameter without clarification — ex. 77/78
  7. Speaker resolves all params; addressee may not — ex. 82
  8. Partial assignment triggers PENDING, not grounding
  9. Updates require structured representations (Hybrid Content Hypothesis) — ex. 2/16

The two readings of a clarification ellipsis. [GC04] ex. 4b–c.

  • clausal : CEReading

    "Are you asking whether p?" — polar question about propositional content. Paraphrasable as a polar interrogative. Presupposes shared belief about the sub-utterance's content.

  • constituent : CEReading

    "Who/what do you mean by X?" — wh-question about the referent/predicate. Paraphrasable as a wh-interrogative. No shared-belief presupposition.

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

      1994/2004 Clarification Ellipsis Apparatus #

      This section was previously in Discourse/Gameboard/Basic.lean §§6, 7, 8, 9, 10, 12, 15. It is paper-specific to [GC04]: in [Gin12], the corresponding machinery uses dgb-params (record types built on the shared CParam) rather than CtxtAssignment, and CCURs (Clarification Context Update Rules) rather than the three coercion operations (parameterFocussing, parameterIdentification, existentialGeneralization).

      We preserve the 2004 formulation here because this study replicates the 2004 paper directly — the apparatus is single-consumer paper-specific content, demoted from substrate to consumer per the linglib pattern (cf. Core/FormFrequency.leanStudies/Haspelmath2021.lean §0).

      The shared substrate primitives — CParam, CParamSet, SubUtterance — remain in Gameboard/Defs.lean since they survive into the 2012 framework as the dgb-params/sub-constituents apparatus.

      The four general theorems about coercion operations are namespaced under Apparatus to avoid colliding with this file's specific-instance theorems on the running example.

      A contextual assignment maps parameter indices to values.

      Grounding requires a total assignment (all C-PARAMS resolved). Clarification arises when the assignment is partial. [GC04] §6, ex. 81–82.

      • bindings : List (String × String)
      Instances For
        Equations
        • One or more equations did not get rendered due to their size.
        Instances For
          Equations
          Instances For

            Does the assignment resolve a given parameter?

            Equations
            Instances For

              Does the assignment resolve all parameters in a set?

              Equations
              Instances For

                Which parameters remain unresolved?

                Equations
                Instances For

                  An utterance skeleton: a sign with C-PARAMS and CONSTITS.

                  The CONSTITS feature (ex. 30) provides access to all sub-utterances. C-PARAMS (ex. 28–29) are the contextual dependencies introduced by the sign, amalgamated from daughters via the Non-local Amalgamation Constraint. [GC04] §3.

                  Instances For
                    Equations
                    • One or more equations did not get rendered due to their size.
                    Instances For
                      def GinzburgCooper2004.Apparatus.instDecidableEqUttSkeleton.decEq (x✝ x✝¹ : UttSkeleton) :
                      Decidable (x✝ = x✝¹)
                      Equations
                      • One or more equations did not get rendered due to their size.
                      Instances For

                        Find the constituent whose CONT matches a parameter index.

                        Equations
                        Instances For

                          A sign paired with a contextual assignment.

                          [GC04] ex. 81 p.353. The assignment f records which C-PARAMS have been resolved. Grounding checks whether f is total.

                          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

                                Clarification Ellipsis processing state.

                                [GC04]: MAX-QUD and SAL-UTT are processing state for the CE analysis. These are NOT part of the [Gin12] DGB or TIS (in 2012, MaxQUD is computed from the QUD poset's maximal element, not stored separately).

                                This state can be used alongside the 2012 TIS when CE processing is needed.

                                • maxQud : Option QContent

                                  The currently maximal question — for CE coercion operations

                                • The salient sub-utterance — target of clarification

                                • pendingUtts : List SignAssignment

                                  Pending utterances awaiting C-PARAMS resolution

                                Instances For

                                  The three coercion operations on signs with unresolved C-PARAMS. [GC04] §5.

                                  • paramFocussing : CoercionOp

                                    Clausal CE reading: polar question about content (ex. 53)

                                  • paramIdentification : CoercionOp

                                    Constituent CE reading: wh-question about speaker meaning (ex. 59)

                                  • existentialGeneralization : CoercionOp

                                    Ground without clarification: ∃-quantify a parameter (ex. 77)

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

                                      Output of a coercion operation: partial specification for the clarification context.

                                      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
                                            def GinzburgCooper2004.Apparatus.parameterFocussing (antecedent : UttSkeleton) (paramIdx : String) :

                                            Parameter focussing ([GC04] ex. 53): derive clausal CE reading.

                                            Takes the antecedent sign and a problematic parameter index. Finds the constituent that introduced the parameter. Produces MAX-QUD = polar question about the antecedent content.

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

                                              Parameter identification ([GC04] ex. 59): derive constituent CE reading.

                                              Produces MAX-QUD = wh-question about speaker meaning.

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

                                                Contextual existential generalization ([GC04] ex. 77): ground without clarifying.

                                                Removes a parameter from C-PARAMS by existentially quantifying it.

                                                Equations
                                                • One or more equations did not get rendered due to their size.
                                                Instances For
                                                  structure GinzburgCooper2004.Apparatus.IS (Fact QContent : Type) :

                                                  Information State for the [GC04] model.

                                                  Bundles a DGB with CE processing state (pending utterances). Uses String for both Fact and QContent, matching the string-based representations in the 2004 paper. The Participant type parameter is set to String, and the LocProp Cont is set to String since this is a 2004-era model.

                                                  This is NOT the [Gin12] TIS — it predates the genre/agenda private state. It exists to support the CE running example.

                                                  Instances For
                                                    def GinzburgCooper2004.Apparatus.IS.initial {Fact QContent : Type} :
                                                    IS Fact QContent

                                                    An empty IS.

                                                    Equations
                                                    Instances For
                                                      def GinzburgCooper2004.Apparatus.IS.integrateUtterance {Fact QContent : Type} [BEq Fact] (is_ : IS Fact QContent) (skel : UttSkeleton) (assign : CtxtAssignment) (toFact : StringFact) :
                                                      IS Fact QContent

                                                      Integrate an utterance into the IS.

                                                      If the assignment fully resolves all C-PARAMS, the utterance is grounded: its content goes to FACTS. Otherwise, it goes to PENDING. [GC04] §6, ex. 82.

                                                      Equations
                                                      • One or more equations did not get rendered due to their size.
                                                      Instances For
                                                        def GinzburgCooper2004.Apparatus.IS.integrateUtteranceStr (is_ : IS String String) (skel : UttSkeleton) (assign : CtxtAssignment) :
                                                        IS String String

                                                        String-specialized integration (content IS the fact).

                                                        Equations
                                                        Instances For
                                                          def GinzburgCooper2004.Apparatus.IS.applyCoercion {Fact QContent : Type} (is_ : IS Fact QContent) (co : CoercionOutput) (toQ : StringQContent) :
                                                          IS Fact QContent

                                                          Apply a coercion output to the IS: set MAX-QUD and SAL-UTT.

                                                          Equations
                                                          • One or more equations did not get rendered due to their size.
                                                          Instances For
                                                            def GinzburgCooper2004.Apparatus.IS.applyCoercionStr (is_ : IS String String) (co : CoercionOutput) :
                                                            IS String String

                                                            String-specialized coercion application.

                                                            Equations
                                                            Instances For

                                                              Convert an UttSkeleton to a string-content LocProp. Subsumes the 2004 skeleton representation in the 2012 LocProp framework.

                                                              Equations
                                                              Instances For

                                                                Convert a LocProp String back to an UttSkeleton. Plain function (not LocProp.toSkeleton) because LocProp lives in Discourse.Gameboard and dot notation looks there for the method, not in Apparatus. Use as locPropToSkeleton lp.

                                                                Equations
                                                                Instances For

                                                                  Round-trip: UttSkeleton → LocProp → UttSkeleton is identity.

                                                                  Both coercion operations target the same SAL-UTT.

                                                                  theorem GinzburgCooper2004.Apparatus.coercions_different_op (ant : UttSkeleton) (idx : String) (r1 r2 : CoercionOutput) (h1 : parameterFocussing ant idx = some r1) (h2 : parameterIdentification ant idx = some r2) :
                                                                  r1.op r2.op

                                                                  The two coercion operations produce different operation types.

                                                                  A fully resolved assignment leaves no unresolved parameters.

                                                                  theorem GinzburgCooper2004.Apparatus.existential_gen_weakens (sk : UttSkeleton) (idx : String) :
                                                                  List.length (existentialGeneralization sk idx).cparams List.length sk.cparams

                                                                  Existential generalization never increases the parameter count.

                                                                  C-PARAM for "Bo": binds variable b to the referent named "Bo". [GC04] ex. 28.

                                                                  Equations
                                                                  Instances For

                                                                    C-PARAM for temporal precedence.

                                                                    Equations
                                                                    Instances For

                                                                      C-PARAM for speaker.

                                                                      Equations
                                                                      Instances For

                                                                        C-PARAM for addressee.

                                                                        Equations
                                                                        Instances For

                                                                          C-PARAM for utterance time.

                                                                          Equations
                                                                          Instances For
                                                                            Equations
                                                                            Instances For
                                                                              Equations
                                                                              Instances For
                                                                                Equations
                                                                                Instances For
                                                                                  Equations
                                                                                  Instances For

                                                                                    Full utterance skeleton for "Did Bo leave?" with all 5 C-PARAMS. [GC04] ex. 32.

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

                                                                                      Speaker (A) resolves all parameters: she knows who Bo is, who she is, who the addressee is, and the temporal parameters. [GC04] ex. 82b.

                                                                                      Equations
                                                                                      Instances For

                                                                                        Addressee (B) resolves all parameters EXCEPT b (Bo's referent). B doesn't know who "Bo" refers to. [GC04] ex. 82c.

                                                                                        Equations
                                                                                        Instances For

                                                                                          A's IS after uttering "Did Bo leave?": fully grounded. Speaker resolves all C-PARAMS, so the utterance goes straight to FACTS.

                                                                                          Equations
                                                                                          Instances For

                                                                                            B's IS after hearing "Did Bo leave?": partial assignment → pending. Addressee cannot resolve b, so the utterance goes to PENDING.

                                                                                            Equations
                                                                                            Instances For

                                                                                              Parameter focussing on "Bo" (parameter b): clausal CE reading. [GC04] ex. 53–54. Output: SAL-UTT = "Bo" constituent, MAX-QUD = ?b.ask(i,j,?.leave-rel(b,t)) Paraphrase: "Are you asking if b left?"

                                                                                              Equations
                                                                                              Instances For

                                                                                                Parameter identification on "Bo" (parameter b): constituent CE reading. [GC04] ex. 59–60. Output: SAL-UTT = "Bo" constituent, MAX-QUD = ?c.spkr-meaning-rel(addr,Bo,c) Paraphrase: "Who do you mean by Bo?"

                                                                                                Equations
                                                                                                Instances For

                                                                                                  Existential generalization on "Bo" (parameter b). [GC04] ex. 77–78. Removes b from C-PARAMS, weakens content to ∃b.ask(i,j,?.leave-rel(b,t)).

                                                                                                  Equations
                                                                                                  Instances For

                                                                                                    B applies parameter focussing to set up clarification context.

                                                                                                    Equations
                                                                                                    Instances For

                                                                                                      B applies parameter identification to set up clarification context.

                                                                                                      Equations
                                                                                                      Instances For

                                                                                                        The running example has exactly 5 C-PARAMS.

                                                                                                        The running example has 4 constituents (Did, Bo, leave, Did Bo leave).

                                                                                                        Addressee does NOT resolve all C-PARAMS (missing b).

                                                                                                        theorem GinzburgCooper2004.speaker_grounds :
                                                                                                        speakerIS.dgb.facts = ["ask(i,j,?.leave-rel(b,t))"]

                                                                                                        Speaker's utterance is grounded (added to FACTS).

                                                                                                        Addressee's utterance is NOT grounded (no new facts).

                                                                                                        Addressee's utterance goes to PENDING.

                                                                                                        Parameter focussing succeeds on "Bo".

                                                                                                        Parameter identification succeeds on "Bo".

                                                                                                        Both coercions target the same SAL-UTT (the "Bo" constituent).

                                                                                                        theorem GinzburgCooper2004.salUtt_is_bo :
                                                                                                        Option.map (fun (x : Apparatus.CoercionOutput) => x.salUtt.phon) focussingOnBo = some "Bo"

                                                                                                        The SAL-UTT is "Bo".

                                                                                                        Focussing and identification produce different operation types.

                                                                                                        Existential generalization removes exactly one parameter.

                                                                                                        The removed parameter is b.

                                                                                                        theorem GinzburgCooper2004.exist_gen_weakens_content :
                                                                                                        existGenOnBo.cont = "∃b.ask(i,j,?.leave-rel(b,t))"

                                                                                                        Existential generalization wraps content with ∃.

                                                                                                        theorem GinzburgCooper2004.focussing_maxqud :
                                                                                                        Option.map (fun (x : Apparatus.CoercionOutput) => x.maxQud) focussingOnBo = some "?b.ask(i,j,?.leave-rel(b,t))"

                                                                                                        Focussing MAX-QUD is a question about the antecedent content.

                                                                                                        theorem GinzburgCooper2004.identification_maxqud :
                                                                                                        Option.map (fun (x : Apparatus.CoercionOutput) => x.maxQud) identificationOnBo = some "?c.spkr-meaning-rel(addr,Bo,c)"

                                                                                                        Identification MAX-QUD is a speaker-meaning question.

                                                                                                        Hybrid Content Hypothesis ([GC04] ex. 2/16): The content updated in dynamic semantics consists of structure expressing detailed relationships between the content and formal properties (syntax, phonology etc) of the various parts of an utterance.

                                                                                                        Evidence: The same propositional content ("Bo left") yields different clarification potentials depending on phonological/syntactic structure. The utterance skeleton encodes this structure via CONSTITS and C-PARAMS.

                                                                                                        theorem GinzburgCooper2004.running_example_matches_ce_data :
                                                                                                        List.map (fun (x : String × Features.Judgment) => x.1) Examples.ex_4a_bo.readings = ["clausal", "constituent"]

                                                                                                        The proper-name CE (ex. 4a) carries both readings, in the order the coercion bridge predicts: clausal, then constituent.

                                                                                                        Shared-belief minimal pair (ex. 11 vs 12): when A and B are in different locations, "Here?" lacks the clausal reading (no shared belief about the content of "here"); when co-located, both readings survive.

                                                                                                        Indexical CE (ex. 13a): "I?" across speakers shifts reference, so shared belief about content fails and the clausal reading is blocked.