Editing logically complex discourse meanings

WYSIWYM (What You See Is What You Meant) is a new methodology for editing knowledge representations by interacting with a generated NL feedback text which presents both the knowledge already deened and the options for extending and modifying it. This paper concerns novel problems that arise when this methodology is extended to deal with certain familiar and supposedly well-understood semantic phenomena. The basics of WYSIWYM editing are brieey explained, followed by discussion of the problems to do with coreference and quantiication and some partial solutions, including an experimental DRT-based prototype. Then we turn to more diicult problems in the area of quantiication and sketch some directions in which solutions may be found.