Attractions of this approach
- Does not limit the strengths and potentials of flat LMNL
- E.g., support for arbitrary overlap including mixing ranges of the same type (“self-overlap”)
- Does not require up-front modeling
- Can be discovered heuristically
- Supports “exploratory” tagging
(mark up a text before models are determined)
- Range of alternatives for syntactic representation
- Out-of-band declarations (a la SGML DTDs)
- Range qualification via namespaces (crude, but sometimes effective)
- Other ideas?
- Consistent with pipelining architecture
- Base limina (content and annotations) come from the syntax
- Interpolate and extract derivative limina