[developers] Roots in ACE output

[Ann Copestake]

So I don't know whether anyone in DELPH-IN has explicitly formalised
the situation where there is more than one root symbol.  One could
extend the definition in the LKB book to allow a set of root symbols Q
and say that the well-formedness condition is that there is any root
symbol in Q that subsumes the relevant feature structure (which is
paired with string list S).  However, if that were all that were going
on, it would make no sense to have a subsumption relationship between
root symbols.  What's actually happening is that we're defining
different classes of well-formed structures according to the
specificity of the root symbol.  Technically, one would probably have
to say there are multiple grammars.  Hence I believe there is some
formal work to be done on the ordered list idea, though I think it's
doable.  (I might have a go tomorrow, since I have a train journey and
plane flight.)

Cheers,

Ann