The parts of speech and word features in traditional dictionaries are adequate for a purely syntactic parser. But a semantic interpreter requires a more detailed representation of the patterns of relationships among the concepts. This chapter surveys various semantic representations and shows how logic, in either the predicate calculus or conceptual graph notation, can be used to generalize and simplify those representations. The generalized semantic information is represented in canonical graphs, which for information extraction can be as simple as frame-like templates, but which can be extended to handle anything represented in feature structures, construction grammars, discourse representation structures, or event structures. The formal rules for manipulating and combining canonical graphs can generate any logical structure necessary for representing the semantic content of a sentence or larger discourse. Although the rules are easier to visualize in the conceptual graph form, they can be translated to an equivalent form in predicate calculus.
For an overview of material that will be incorporated in this chapter, see the papers by Sowa in the reference section and Chapters 4 and 5 of the book Knowledge Representation.
Next is Part III: Language Processing, or go back to the Introduction or to Part I: Problems and Issues.
Send comments to John F. Sowa.