Clarification, Ellipsis, and the Nature of Contextual Updates in Dialogue

Intention, interpretation and the computational structure of language

“…I observed a number of directions for such integrations during the course of the paper. On the one hand, we need richer modeling of task context, perhaps along the lines suggested by Rich et al (2001), as well as richer modeling of the linguistic context, perhaps along the lines suggested by Ginzburg and Cooper (2004). At the same time, we need more detailed accounts of linguistic syntax, along the lines suggested by Frank (2002) and more detailed accounts of linguistic semantics, perhaps along the lines suggested by Gardent and Kallmeyer (2003).…”

“…We can describe the evolving status of contributions to the dialogue itself through systematic models of dialogue obligations and grounding, perhaps as Larsson and Traum (2000) propose. And we can characterize the shared information and coordinated attention in the linguistic context systematically las a function of the linguistic elements and the linguistic relationships that make up dialogue (Asher & Lascardies, 2003;Ginzburg & Cooper, 2004;Stone & Thomason, 2002. Of course, substantial research remains in giving unified analyses of dialogue by exploring the connections among these compatible models.…”

Pursuing and demonstrating understanding in dialogue

“…See Gregoromichelaki et al [2011] for a wide range of further cases. Other types of fragments, such as the reprise fragments of (1.1), seem to carry semantic constraints that closely tie their interpretation to those of prior utterances [Ginzberg andCooper, 2004, Purver, 2004]. Both cases create interpretive connections that readily signal grounding.…”

“…Our action set builds on stack-based models of clarification subdialogues [Ginzberg andCooper, 2004, Purver, 2004], collaborative negotiation models [Sidner, 1994, Carberry and Lambert, 1999, Eugenio et al, 2000, and the use of collaborative discourse theory to characterize user interface tasks [Rich et al, 2001]. For example, one action type, pushCollabRef[D, M, T ] lets the director D initiate collaborative reference with matcher M to a target T .…”

CLARIE: Handling Clarification Requests in a Dialogue System

“…The two classification schemes are inspired by different underlying formalisms: R&S follow (Schlangen, 2004)'s view of CRs being caused by problems at various grounding levels (acoustic, lexical, semantic etc. ); while PGH assume (Ginzburg and Cooper, 2004)'s view of CRs as being caused by individual contextually dependent parameters of utterances, and problems in their grounding. The two views seem reconcilable (see section 6); but PGH's scheme is taken here, mainly because its parameterised approach allows CR forms and readings to be linked specifically with grammatical analysis of the CR.…”