PRED-features necessary?

[Bruce Mayo]

Alex Alsina writes (14 Jun 96)
>Ah, now I see the problems that Avery Andrews is having with the PRED
>feature.  Once you have introduced the features REL and ARGS into the
>picture, it makes no sense to still have a PRED sitting around waiting
>for some function to fall on it.  In a sense you have renamed PRED as REL
>and ARGS.  Therefore, the role of PRED is taken up by REL and ARGS and
>there is no reason to keep PRED around.  (See relevant citation below
>from Avery Andrews' latest msge.)

I'd like to make another plea, again from the standpoint of a computational
implementor, for the PRED feature and for the independence of f-structure.
I remember thinking, after reading  Andrews and Manning 1993, this makes it
pretty clear that PREDs as an interface to the lexicon aren't necessary -
in a formal sense. But computationally PREDs make things a whole lot easier
and clearer. They're like index tables in a relational data base - they
provide quick and easy access to data that would sometimes be extremely
hard to compute on the fly. I don't doubt that psychologically they have a
function of this sort - they fix the mapping from lcs argument struture to
sentence surface structure in a readily available form, so that we don't
have to figure it out every time we use a word. Of course, sometimes we do
have to figure it out, when we stumble over neologisms - that's probably
what the stubling is about.