Supervised text analysis for ontology and terminology engineering

One of the means to reach a Semantic Web is to add some machine readable meta-data to documents (semantic annotations) and/or to improve the performance of information retrieval applications thanks to the use of semantic resources such as terminologies and ontologies. This position paper promotes the idea that these semantic resources cannot be universal but should rather be domain and even task specific in most cases. Moreover, we assert that their content is all the more relevant that it has been defined from document content analysis. And finally, we advocate in favor of a supervised learning process for their design from well selected texts. We shortly present two contributions, a method to build ontologies from texts and a tool for semantic relation identification, that illustrate these positions.