[Paper Review] Goal-oriented Dialog as a Collaborative Subordinated Activity involving Collective Acceptance

Sylvie Saget, Marc Guyomard|arXiv (Cornell University)|May 27, 2008

Multi-Agent Systems and Negotiation16 references4 citations

TL;DR

This paper proposes Collective Acceptance as a formalized social mental state to model Conversational Common Ground in goal-oriented dialog, offering a rational, computationally lighter alternative to shared beliefs. It formalizes referential acts as collaborative, subordinated activities governed by mutual acceptance of conceptual pacts, demonstrating how alignment emerges through pragmatic, context-sensitive coordination rather than strong epistemic commitments.

ABSTRACT

Modeling dialog as a collaborative activity consists notably in specifying the content of the Conversational Common Ground and the kind of social mental state involved. In previous work (Saget, 2006), we claim that Collective Acceptance is the proper social attitude for modeling Conversational Common Ground in the particular case of goal-oriented dialog. In this paper, a formalization of Collective Acceptance is shown, besides elements in order to integrate this attitude in a rational model of dialog are provided; and finally, a model of referential acts as being part of a collaborative activity is presented. The particular case of reference has been chosen in order to exemplify our claims.

Motivation & Objective

• To address the limitations of modeling Conversational Common Ground (CCG) using strong epistemic states like shared beliefs, which are computationally complex and not always necessary.
• To argue that Collective Acceptance—defined as a mutual, coordinated acceptance of a conceptual pact—is more appropriate for modeling CCG in goal-oriented dialog.
• To formalize the role of referential acts as collaborative, subordinated activities within a rational dialog model.
• To provide a framework where grounding emerges through pragmatic, context-sensitive alignment rather than full mutual belief states.
• To demonstrate that this model supports successful coordination in dialog while avoiding over-privileging epistemic commitments.

Proposed method

• Defining Collective Acceptance as a social mental state where agents mutually accept a conceptual pact (e.g., a description of a referent) without requiring full mutual belief or nested belief structures.
• Formalizing referential acts as part of a collaborative process involving speaker and addressee intentions: identification (picking out the referent) and grounding (confirming mutual understanding).
• Introducing a dual meta-goal structure: for the speaker, coordination of intention and collective acceptance; for the addressee, existential quantification over possible referents.
• Using logical formalism to represent mutual belief (MBel) and belief (Bel) states, distinguishing between speaker’s belief in mutual belief and actual shared understanding.
• Modeling dialog as a subordinated activity where task-level goals (e.g., meeting at a place) are coordinated through conversational acts that establish conceptual pacts.
• Applying the framework to a concrete example: Tom and Laura negotiating a meeting place using a description that requires clarification and eventual acceptance.

Experimental results

Research questions

• RQ1 Why is modeling Conversational Common Ground using shared beliefs or nested beliefs problematic in goal-oriented dialog?
• RQ2 How can Collective Acceptance serve as a more appropriate and computationally lighter alternative to strong epistemic states in dialog modeling?
• RQ3 In what way are referential acts inherently collaborative and subordinated to task-level goals in dialog?
• RQ4 How does the process of grounding emerge through conceptual pacts rather than full mutual belief?
• RQ5 What formal mechanisms can model the transition from failed identification to successful acceptance in dialog?

Key findings

• Collective Acceptance provides a more accurate and computationally efficient model of Conversational Common Ground than shared beliefs or weaker belief states.
• The example shows that when Laura fails to identify the referent, she triggers a clarification request, demonstrating that acceptance depends on successful identification.
• After Tom provides a revised description, Laura accepts it, leading to the formation of a Collective Acceptance: $CollAcc_{Laura,Tom}(referedBy( heta x. heta'(x), l'))$, which establishes mutual understanding.
• The model achieves mutual understanding ($MBel_{Laura,Tom}(meetingPlace = l')$) and task coordination through a sequence of collaborative acts, not through full epistemic commitment.
• The framework successfully models grounding as a dynamic, context-sensitive process, avoiding the high complexity of full belief-based systems.
• Collective Acceptance enables effective communication even when mutual belief is not fully established, supporting flexible, real-time coordination in dialog.