PRED-features

[Ron Kaplan]

Joan's point about the grammatical significance of PRED as defining a
predication-nucleus is well-taken.  I wasn't thinking so much about such
grammatical consequences when I made up my list of 4 items, but this is
also clearly an important property of PRED+semantic-forms.  Thus a fifth
item should be added to the list, so that a PRED+semantic-form represents:

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

Avery then remarked:
>But it seems to me that you don't really need a full-bodied PRED feature
>to perform these functions, an attribute like NUC +, in conjunction
>with the grammatical functions, would be sufficient (keeping in mind
>that the linear-logic based semantics is handling completeness and
>coherence).

Yes, a funny feature would serve the same formal purpose, and you could
even make it be a "property" of f-structure by introducing a notation such
as NUC(^), just like the GGF notation.

But apart from the motivations I mentioned in my earlier message, there is
really a substantive claim in the PRED+semantic-form arrangement, namely,
the claim that these 5 properties are universally correlated:  if a
structure has any of these properties, then it has all of them. This was
really the intuition we had in packaging it all up in the single formal
device, although this intuition didn't get discussed as a separate issue
among all the other ideas we were trying to put forward.  So even if all or
some of these properties are provided by alternative semantic and syntactic
mechanisms, there would still be a theoretical claim that all these things
"go together".

As an aside, there is some debate as to whether it is a "bug" or a
"feature" of the linear logic approach that it reduces the two completeness
and coherence requirements to a single formal condition.  It achieves this
by converting the syntactic subcategorization requirements to corresponding
conditions of semantic saturation.  We considered and rejected this option
when we were first defining the formalism, because the correlation between
subcategorization and saturation, although quite high, is not perfect.  In
order to get this to work, you might have to assign otherwise unmotivated
dummy meanings to grammatical formatives such as "it" and "there", and also
make sure that nonthematic subcategorized functions are also properly
treated.  It seemed more appropriate to recognize and formalize a somewhat
independent notion of syntactic subcategorization (and thus avoid another
instance of Occam's seduction).  By the time you augment the single linear
logic condition with the extra meaning specifications, the linear logic
reconstruction of subcategorization looses some of its appeal, at least to
me.

Avery continued:

>I'd reject this on the basis that it is better to have several different
>variants of the theory, expressing apparently different ideas, than one
>theory where the different ideas are blended together.

I'm not sure why this is "better" if the various ideas form a natural,
completely correlated class.  Keeping them separate is a good strategy if
you aren't sure they go together, and maybe that's Avery's point--he
doesn't (yet) believe the intuition that led to the original formulation.

>
> >(2)  Knowing what devices a particular semantic theory uses to characterize
> >items 2,3,4 above, we can retain the semantic forms in the lexicon and
> >grammar but treat them as an abbreviatory notation for the more detailed
> >specifications that the semantic theory requires.  Thus, in the example
> >above, we would think of the PRED semantic form as a short-hand for the 3
> >sigma equations.  Or, on the linear logic view, we would think of the
> >semantic form as short-hand for premises such as "if (^ SUBJ) means X and (^
> >OBJ) means Y, then ^ means kick(X,Y)".
>
>Yes, I'd agree with this, but then the architecture, or at least its
>presentation, should be altered to suit.

I agree that if we are regarding semantic forms as in (2), then we ought to
change the presentation to make this more explicit.  I don't see that this
is a change to the "architecture", since all the other formal concepts
(correspondence, description, etc.) remain the same.

--Ron