Detecting when a parsed intent is underspecified or has more than one plausible resolution, and either asking a targeted clarifying question or offering the most likely alternatives — rather than silently guessing and grounding the wrong data.