The Data Format Description Language (DFDL or "daffodil") is emerging as a good notation for describing mappings between data in formatted files (text as well as binary) and a corresponding XML representation. A DFDL specification takes the form of an XML Schema with "application annotations" that make the correspondence between file characters (or bytes or even bits) and XML data values precise.
DFDL is being developed by the Data Format Description Language working group of the Open Grid Forum.
Our Virtual XML Garden release includes a prototype implementation.