Eve Maler and Jeanne el Andaloussi’s out-of-print book Developing SGML DTDs: from Text to Model to Markup has just been put online I see. (Through the magic of Docbook!)
Even though it looks dated in its SGML examples, it really is about a methodology for analysing and designing schemas (especially for literature, i.e. “documents” rather than “data”) that is just as useful today. We might call SGML XML, and we might use “MIME type” or “data type” instead of “notation”, but the development issues this book addresses never went away. Anyone who wants to be an expert in XML schemas and document analysis needs to be aware of it, IMHO.
A good taster might be Learning to recognize semantic components.