Towards a corpus-based distinction between language-specific and universal features of mediated discourse