about PREDs

[Alex Alsina]

In my latest posting to this list I say something to the effect that
possibly the only use of the feature PRED is to take the argument
structure as its value.  This appears to be in conflict with the many
uses that Joan Bresnan and Ron Kaplan find for it, as stated in their
messages.  And maybe it is, but I would like to think it isn't really.

The following is the list of uses that the PRED has, according to Ron
Kaplan:

> 1.  The assignment of grammatical functions (subcategorization)
> 2.  The mapping of grammatical functions to semantic arguments
> 3.  The individuation of semantic entities (by the instantiation property)
> 4.  The name of the semantic relation.
> 5.  The nucleus of predication

If we assume that there is such as thing as an argument structure,
which is supposed to constrain the grammatical functions that are
associated with a certain predicate, subcategorization need not (and
therefore should not) be stated in terms of grammatical functions.
The argument structure is the subcategorization information, and then
there is a set of principles that links (or maps) the argument slots
in the argument structure to grammatical functions in the f-structure.
Thus, the information about what grammatical functions a given
predicate takes (item 1 above) is divided between the argument
structure, the theory that maps argument structure to grammatical
functions, and possibly other principles.

Item 2 is the responsibility of the theory that maps argument slots to
grammatical functions.  So, once we have developed this theory and a
theory of argument structure, this item get absorbed by arg structure
and the mapping theory.

As for item 3, I must confess I'm out of my depth here.  (It's better
to confess one's own ignorance than to have someone else expose it.)
My guess is that it's the information about which semantic object
corresponds to which argument slot.  If that is the case, all we need
to do is to assume that there is a semantic representation, along with
f-structure, a-structure and others, with links between all of these
representations.  Once we have the relevant theories relating all
these representations to each other, we will have taken care of item
3.  In any case, Ron Kaplan says:

> We can assume also that the
> semantics itself will worry about individuation, so that that property of
> the semantic form can also be dispensed with.

Item 4: the name of the semantic relation.  Now that doesn't sound
like a very vital piece of information for the syntax.  But if we must
have it, we can have it as part of the argument structure, as many of
us do.  (Although, thinking about it, it looks more like a decoration
than anything, since it is not involved in any syntactic process.)

Item 5: The nucleus of predication.  Defining the nucleus of
predication can be done perfectly well if we assume that the value of
PRED is an argument structure.

So, it seems perfectly reasonable to assume that the function of PRED
is to take an argument structure as its value.  From this and the
attendant mapping theories, the five items noted above will follow,
although some of them will no longer be information explicitly encoded
in the PRED value.  

Now, one could say: since the argument structure is a level of
representation distinct from f-structure, we shouldn't represent it as
the PRED, and therefore in the f-structure.  So, it would be perfectly
reasonable to put the argument structure outside the f-structure,
which I believe is what Avery Andrews has in mind, and have it be the
value of the ARGS feature.  That's fine with me, but then there is no
reason to keep the PRED in the f-structure and it would obviously be
an embarrassment to keep it there.  In a theory you only have what you
need to have.

Alex Alsina