previous | start | next

---

Diagnosis and conclusions

• The errors were in the natural language content of the document and weren't captured by the markup.
• This wasn't due to the markup being too coarse grained: even at the finest level of granularity, the errors were distributed across the document structure in unpredictable ways.
• This will be almost inevitable because any structure of the underlying meaning of natural language content would almost always cross-cut any plausible markup-level structure.
• Other than in highly formal document types (stock quotes, weather and traffic bulletins) document meaning is more likely to be carried by character data than by markup.
• Character data is not amenable to machine driven markup-level syntactic transformation: it needs human assistance.
• Unless 'device-independent' authoring is only intended to be applicable within very narrowly circumscribed classes of device then we need a different approach ...