Frame of Reference: Addressing the Challenges of Common Ground Representation in Situational Dialogs

TL;DR

This work advances robust, long-horizon grounding in situated dialogs by introducing IndiRef, a benchmark that tests relational reference resolution to established common ground. It analyzes Meetup and Spot the Difference to categorize relational references and demonstrates that standard LLMs struggle to resolve such references, especially under constrained context. The authors compare full-dialog-context baselines with resource-limited representations (summarization, chunking, ontology) and show notable gains from prompt engineering and a reinforcement-learning framework trained on synthetic data (GRPO). The findings highlight the need for explicit, retrievable common-ground representations and provide a scalable path via synthetic data and learning-based grounding strategies, while acknowledging current limitations and directions for future work in multi-session belief tracking and larger-scale models.

Abstract

Common ground plays a critical role in situated spoken dialogues, where interlocutors must establish and maintain shared references to entities, events, and relations to sustain coherent interaction. For dialog systems, the ability to correctly ground conversational content in order to refer back to it later is particularly important. Prior studies have demonstrated that LLMs are capable of performing grounding acts such as requesting clarification or producing acknowledgments, yet relatively little work has investigated how common ground can be explicitly represented and stored for later use. Without such mechanisms, it remains unclear whether acknowledgment or clarification behaviors truly reflect a grounded understanding. In this work, we evaluate a model's ability to establish and exploit common ground through relational references to entities within the shared context in a situational dialogue. We test multiple methods for representing common ground in situated dialogues and further propose approaches to improve both the establishment of common ground and its subsequent use in the conversation.

Figures (12)

• Figure 1: An example of people referring back to the common ground.
• Figure 2: Pipeline for the synthetic data generation involving three main phases: (1) Grounded World Generation, creates the factual basis of the scenario; (2) Dialog Synthesis, where LLM generates conversation constrained by the script; and (3) Q/A Synthesis, which extracts and formats question-answer pairs for training.
• Figure 3: Test example for inferred grounding from Meetup dataset
• Figure 4: Test example for attributive grounding from Meetup dataset.
• Figure 5: Test example for spatial grounding from Spot the difference dataset