Tag
This paper proposes an assistive AI agent architecture based on hierarchical compositionality and simple heuristics to resolve ambiguity in object references, outperforming data-driven baselines while supporting adaptation to user preferences.