What are real DTDs like?

DTDs have proved important in a variety of areas: transformations between XML and databases, XML storage, XML publishing, consistency analysis of XML specifications, typechecking, and optimization of XML queries. Much of this work depends on certain assumptions about DTDs, e.g., the absence of recursion and non-determinism. With this comes the need to justify these assumptions against DTDs in the real world. This paper surveys a number of DTDs collected from the Web, and provides statistics with respect to a variety of criteria commonly discussed in XML resesarch.