Using Machine Mental Imagery for Representing Common Ground in Situated Dialogue
Researchers have developed a new framework to improve how conversational agents maintain common ground during dialogues. This approach uses machine mental imagery, converting dialogue states into persistent visual histories that agents can retrieve for grounded responses. Evaluations on the IndiRef benchmark indicate that this visual scaffolding reduces "representational blur" and enhances grounding, especially when combined with traditional textual representations. AI
IMPACT Enhances conversational AI's ability to maintain context and grounding through multimodal representations.