Notes from my meeting with Mark just before I went away from Christmas (20th Dec 2010).
Assume there's no dependency B - these Is are free to be interpreted differently.
Dependency A determines the semantics of the VI7.
The VI7 doesn't sound ambiguous, as it would be if the Is were semantically equal, until I1 is reached.
The I1 is the surprising part, because it's mistuned.
Dependencies in coordination - resolution of both V7s is identical, or else we'd be in two places at once.
II7 is unambiguous due to A. VI7 is ambiguous until the resolution fixes V07 and V17 to have equal semantics.
- This is the structure in the semantics that requires more than a FSM to interpret.
- Otherwise only adjacent dependencies, so FSM would do fine.
- Continue treating as before.
- Maximal first span is fine (though it doesn't matter).
- New idea for notation.
- Clearer than before - closer to the old notation.
- Use just one part for atoms in cadence categories.
- Produce both parts (S-E) as result of application.
- Forward category always looks only at start.
- Backward category looks at end.
- Need to work this through and see if it works.
- Instead of putting start/end ambiguity in the categories, makes much more sense to build it into the rules.
- Try rewriting rules to do this.
- Should be able to eliminate unification completely (or maybe just partially) - ought to make parsing faster.