Abstract : Organizations' information systems contain different kinds of data, dispersed in several sources. The purpose of the managing heterogeneous data is to offer a transparent access to this set of sources. We are interested in the management of structured (relational databases) and unstructured ((multilingual) textual sources) data. We especially describe an approach for taking textual sources into account in an integration system. The approach we propose is based on the use of Semantic Web technologies and different kinds of ontologies. Ontologies are used to define the global schema (global ontology) and the sources to be integrated (local ontologies). Local ontologies are obtained in a semi-automatic way using reverse engineering techniques. Ontologies are also used for the hybrid representation of textual sources. The hybrid representation combines cataloguing information, vectors of terms and concepts and optionally named entities identified in documents. We have designed and implemented an ontology server to manage multiple ontologies and support queries. A first application domain of our work has been the brain field. We have developed or enriched ontologies for brain knowledge management and semantic characterization.