A major difficulty which needs to be overcome in the design of knowledge representation systems is the gap between lexical items (with their lexical or syntactic relations) and concepts (with their semantic relations). An extensive collection of attempts at combining lexical semantics and conceptual modeling can be found in Pustejovsky (1993). The most obvious differences between the lexical level and the conceptual level are synonymy, homography, polysemy, metaphors, and metonymy - all demonstrating structures that exist in one level, but not in the other. On the other hand, lexical and conceptual structures are also dependent on and complementary to each other. Human speakers unconsciously apply a fixed set of lexical items to an open set of concepts. Polysemy, metaphor, and metonymy allow speakers, for example, to be creative, to relate concepts, and to connote associations to concepts. For a computer and for knowledge representation purposes, polysemy, metaphor, and metonymy provide difficulties: does each polysemous meaning or each metaphorical use of a lexical item represent a new concept which should be stored separately or should the polysemous meanings of a lexical item be computed according to certain semantic rules?
Kilgarriff (1995) develops a graphical representation of lexical and conceptual items using the semantic network DATR (Evans & Gazdar, 1989) which favours the computation of polysemous meanings. In contrast to Kilgarriff's theory we apply Relational Concept Analysis (Priss, 1997) to lexical databases and graphically represent polysemous meanings as separate concepts. While the DATR modeling seems to be adequate for a computer knowledge base that functions without human intervention, the Relational Concept Diagrams are easier to comprehend for humans and therefore adequate for human-computer interactive systems. Human users gain more control over the lexical and semantic relations; updates and consistency checks are made easier. Furthermore, the Relational Concept Analysis approach can be utilized to visualize relations among metaphorical uses of lexical items. And it facilitates insights into the general interrelationships between lexical and conceptual structures. This poster will present graphical representations of conceptual and lexical structures and their relation to each other using Relational Concept Analysis. Polysemy and, to some extent, metaphor are visualized in their regularities and in their dependency on conceptual structures.
Keywords: computer models, polysemy, relational concept analysis