The standard DITA package comes with three basic information types for concept, task, and reference information. Each of these has a DTD that makes it easy to author the appropriate content. The output transforms also recognize the information type, and make use of it when generating or sorting links.
Each of these three topic types is authored using a specialized DTD. These specialized DTDs contain several new elements, although they also include a significant number of the block and phrase elements that are part of topic. One new element is the "steps" element in the task DTD; the contents of the element help guide authors to ensure they place the proper information in their steps. This element is based on the ordered list element in the original topic DTD, and contains an attribute that states this explicitly.
When an editor, translation tool, or output transform sees this attribute, it knows that it is really dealing with an ordered list in disguise. It can treat the element exactly the same as it would an ordered list, without ever having any knowledge of the actual element name, and without ever having seen the DTD beforehand.
The translation tool we use in IBM requires a customization for every DTD. Any time you create a new DTD from scratch, you need customize basic information, such as which elements are phrases and which elements (and attributes) should be translated. With DITA, this customization is done once -- it is based on the ancestry attribute rather than on the element name. This means that the steps element is immediately identifiable as an ordered list, using the basic DITA customization, regardless of whether the translation tool has ever seen or translated a steps element in the past.
After translation, our tool stores the new values in a translation memory, to speed up the process the next time around. Just as the translation tool recognizes the steps element as an ordered list, it recognizes the "step" element as a single list item -- that is, a single block of text. When the step is translated, the translation is stored as a single block of text. This tool will now match that sentence exactly when it shows up in a step, a list item, or any other element that was based on a list item -- regardless of whether it has seen that element before.
