Ontology Evolution trough Interaction

From International Center for Computational Logic

Ontology Evolution trough Interaction

Talk by Irina Dragoste
This paper shows how agents can evolve their ontologies while trying to communicate, following a protocol inspired by cultural evolution experiment methodology. While other studies have dealt with evolving ontology alignments, we are interested in refining the classification precision, which is accomplished during the protocol: while communicating about objects encountered in the environment, agents may discover the need to increase their classification precision in order to reach an agreement with the interlocutor. They do so by creating and adopting new categories from other agents that would help them discriminate better. We conducted several experiments to show that a population playing this game will always converge to a state of full communication success, when agents have stabilized their knowledge and cannot learn any new categories.