With the proliferation of execution contexts (PDAs, mobile phones, blind-friendly interfaces, etc.), medias must be adapted according to the target context. In order to ensure universal access to information, we propose an architecture which uses information retrieval techniques in order to replace incompatible media items with compatible ones. More specifically, we advertise the use of the Web as a repository of potential alternative media items. Moreover, the chosen media item must convey the same message as the original ones, while satisfying the target profile. We show that Web-based search engines can already achieve this to a limited extent. Nonetheless, some results are unsatisfactory because media annotations lack semantics, are partial and are heterogeneous. Hence, we demonstrate the value of Semantic Web technologies, such as RDF descriptions, ontologies, ontology merging and matching, for improving this adaptation framework.