Inputs to the encoder are XML text which is read via SAX events and the pull parser, and a schema from the repository, going through a vocabulary manger.