Reacting to Unexpected Events and Communicating in Spite of Mixed Ontologies

Abstract

Abstract. We describe our experiences in building agents (and their environment) that could solve important problems in agent to agent communication, in manners that are not pre-programmed or reactive: (1) Two agents may have different ontologies. We do not assume that agents share the same classification of concepts: they have each one its own ontology or concept categorization. Agents can not exchange concepts: they have to exchange symbols (words of natural language), which the receiving agent has to map to the intended concept, for interaction to be meaningful.

Finding the Most Similar Concepts in Two Different Ontologies

“…[9] seeks to communicate several agents sharing a single ontology. The authors have been motivated [6,7] by the need of agents to communicate with unknown agents, so that not much a priori agreement between them is possible. With respect to concept comparison, an ancestor of our COM ( §2, appears first in [13]) matching mechanism is [2], based on the theory of analogy.…”

“…6 We have found useful the threshold 0.5: more than half of the compared entities must coincide. 7 If p B is found more than three levels up, the "semantic distance" is too high and sim says "no match." …”

Hierarchies Measuring Qualitative Variables

“…Its value cannot be a set. 4 By symbolic we mean qualitative, as opposed to numeric, vector or quantitative variables.…”

“…Perhaps infinite, perhaps empty 3. The symbol means: end of definition 4. Variable, attribute and property are used interchangeably.…”