Grounded Dialogue Framework for Selective Belief Revision

A new data-driven framework manages turn-by-turn dialogue by deciding when to preserve or revise mental models under referential uncertainty. It demonstrates that updating solely on local mismatch leads to unstable conversational understanding compared to more conservative, prior-based strategies.