Hi,

> Interesting. Have you confirmed that this works? I am slightly
> worried by the fact that an LR parser reads one token ahead,
> i.e. one token past BEGIN_VERB might already have been read
> before the enter_verb semantic action is executed. If that is
> so, then this token would be read while the lexer is still in
> the wrong mode.

Yes, I was just thinking about that as well... :-)
I think I can pile another hack on top of the dummy action:
dummy tokens to take care of the readahead issue.  Though
this has the potential to get comically silly pretty quickly!

I'll report later...

cheers,
Dario Teixeira