On the generation of text and speech for database applications