Whether you should use attributes or elements in your DTD 
depends in large part on the purpose for which you're 
designing the document type. The two extremes are best 
illustrated with examples.

"Traditional" textual practice is to include the "real" 
text (what would be printed) as character data content 
and keep the metadata (like line numbers) in attributes, 
from where they can more easily be isolated for analysis 
or special treatment like display in the margin or in a 
mouseover:

<l n="184"><sp>Portia</sp><text>The quality of mercy is not strain'd,</text></l> 

But from a systems point of view, there's nothing wrong 
with storing the data the other way around, especially 
where the volume of text data on each occasion is 
relatively small:

<line speaker="Portia" text="The quality of mercy is not strain'd,">184</line>

Much depends on what you want to do with the information 
and which bits of it are most easily accessed by each 
method. A good rule of thumb for conventional textual 
documents is that if the markup were all stripped away, 
the bare text should still be readable and usable, even 
if inconvenient. 

For database output, however, or other machine-generated 
documents, reading might not be meaningful. In this case, 
it's certainly possible to have documents where all of 
the data is in attributes and the document contains no 
character data in content models at all.