Where MOE is going

Lexical classes I'll be writing up a set of classes that support much more of the original lexical content of documents as the parser comes to life.
XPath XPath support is critical for creating more sophisticated rules files. I'm not sure yet how to balance a streaming subset with MOE's capabilities for tree representation, but I'm working on it.
Easier integration with SAX filtering I need to concoct a pre-built SAX filter framework that makes it easier to use MOE internally. This may be part of the transition of Regular Fragmentations into MOE.
An interactive GUI, maybe an editor I'd really like to see MOEWorkshop flourish. It seems like MOE's design for filtering and its support for human-friendly lexical notation and incomplete markup should be a bonus, but this will take a long while to get right. Early stages should appear soon, I hope.
Beyond XML Regular Fragmentations has already taught me that XML isn't the only kind of markup, though it's useful stuff as a canonical form. I've started writing a generic parser which uses non-XML lexical notations to generate MOE events, but that'll be a while - creating a rules format that produces consistent results is tricky.