TxText architecture question

In the TxText model, a text consists of a double-linked list of spans,
containing the text content and its attributes. I wonder how to
efficiently map a hierarchical text model (like that of Pillar) to that.

Stephan