As claimed in the Semantic Web project, a huge amount of physically distributed interacting software agents could find the semantic of available resources and answer more relevantly to users' requests if the content of these resources would be represented with formal semantic concepts defined in ontologies. Because Web information sources are highly dynamic and conceptually heterogeneous, one of the most challenging problems in the Semantic Web research is the proper and frequent ontology updating in keeping with knowledge changes. To tackle this problem, we have developed a self-organizing multi-agent system -Dynamo- able to create an ontology draft from automatic text processing. Because it is well-known that only a part of a domain description is explicitly described in texts, Dynamo enables an ontology co-construction with a domain expert in a fully interactive way. In this paper, we present the principles of this approach and related experiments.