Controlling Logical Scope in Text Generation

Many applications of Natural Language Generation employ object-oriented knowledge representation formalisms such as loom. Each entity in the domain is represented by an object belonging to a specified class and possessing attributes suitable for that class. From a logical viewpoint, domain entities correspond to existentially quantified variables, but since loom and kindred formalisms provide no scoping mechanisms, it has to be assumed that all variables have wide scope. For technical material including conditional, negative or modal contexts this is a serious limitation. We suggest a simple extensions to standard objectoriented formalisms which allow encoding of logical scope, and show how logically complex material can be edited using the `wysiwym' approach. 1 Introduction Many Natural Language Generation (nlg) applications (Paris et al., 1995; Bateman, 1996) encode the content of the generated texts in an object-oriented knowledge representation formalism, such as loom (MacGregor a..