Ginzburg and Cooper (2004): Clarification, Ellipsis, and the Nature of Contextual Updates in Dialogue #

This file formalizes the account of clarification ellipsis of [GC04], on which a sign carries the contextual parameters its sub-utterances introduce (LocutionaryProposition.parameters, LocutionaryProposition.constituents), grounding an utterance is finding an assignment for all of them (Grounds, grounds_iff), and an addressee who cannot updates the context instead by a coercion operation on the sign: parameter focussing raises the question ?i.p, the content with the problematic parameter abstracted (parameterFocussing), parameter identification the question what the speaker meant by the sub-utterance (parameterIdentification), both making that sub-utterance salient (focusEstablishing_parameterFocussing), and contextual existential generalization quantifies the parameter away so that the weaker content can be grounded (existentialGeneralization, grounds_existentialGeneralization). Since the operations read the constituents of the sign, utterances with the same content have different clarification potentials (potential_ne_of_constituents): the Hybrid Content Hypothesis, against the purely semantic updates of dynamic semantics.

The utterance-processing protocol integrates a pending utterance whose parameters the participant's assignment resolves and otherwise clarifies it (IS.ground, IS.clarify); after "Did Bo leave?" the speaker's and the addressee's information states differ in MAX-QUD and SAL-UTT (Ex32.differ), and the speaker comprehends the addressee's fragment by applying the coercion to her own latest move (IS.backtrack, backtrack_eq_clarify). A fragment resolves against the clarification context when it matches the salient sub-utterance in category (DeclFrag); the clausal reading identifies its index with the parameter and so presupposes that the participants share the sub-utterance's content, the constituent reading does not (rows_clausal, rows_constituent), and a fragment of the wrong category has neither (rows_parallelism).

Implementation notes #

Signs are the substrate's LocutionaryProposition, whose constituents' denotations name the parameter each contributes; contents are opaque, so existential generalization takes the binder as an argument, and parameter labels and contents are written as the paper's strings. A contextual parameter is its label: the restrictions of (28) are not represented, grounding asks only for a value, and the parameters an assignment leaves unresolved are those contextual instantiation of its labels leaves.
A coercion operation takes the problematic sub-utterance itself, whose denotation is the parameter; the clarification context it yields is the substrate's InformationStructure, with MAX-QUD as its question and SAL-UTT, a set of at most one sign, as its focus-establishing constituents.
The information state keeps the components that (82) displays: LATEST-MOVE as a sign with its assignment, PENDING, MAX-QUD and SAL-UTT; the FACTS update (81) and the grounding conditions (80) are not modelled.
The constituent reading is licensed by categorial parallelism alone, the reformulation without utterance anaphora of footnote 55, so that the non-identical fragments of (8) get it; the categories of (10) are distinguished by the case or verb form the contrasts turn on.

Ginzburg and Cooper (2004): Clarification, Ellipsis, and the Nature of Contextual Updates in Dialogue #

Implementation notes #

References #

Contextual parameters and grounding #

Sign coercion #

Integrating utterances in information states #

The running example, "Did Bo leave?" (32) #

Clarification ellipsis and its readings #