Reentrancy in feature structures

[Carl Pollard]

Hi Mike,

>
> Are there other cases where the strong notion of identity is required,
> besides (person) co-reference?

To which Carl Pollard replied:
> I hope not, since in my current research I have foresaken the
> graphical implementation of feature structures in favor of a treatment
> in terms of higher order logic. In the models this means feature
> structure types are indexed product types and the feature structures
> themselves are members of those types.  Thus there is only plain old
> equality (EQ), nothing corresponding to EQUAL. This makes some things
> (such as formalizing set values and distinguishing neutralization from
> ambiguity) much easier, but also means that coindexing cannot be
> handled the same way as in HPSG.

OK, I'll bite: how do you do coindexing?
>>

I haven't done anything about it yet. I suppose one could say there
is a syntactic feature INDEX with values of type _index_, which in
turn is the coproduct of all the agreement types (such as _3rdsng_),
and each of the agreement types would have a denumerable infinity of
members (so the logic would have to have a denumerable infinity of
constants of each of the agreement types). Then coindexing would just
be equality of indices.

>
With s.t. like Luis Casiallas'
indices (i, j...)?
>>

It does seem pretty much the same, doesn't it?

>
I used to think of those as a notation for
re-entrancy, but maybe not.
>>

It's different, because in the feature logics I'm familiar with there
are no constants correpsonding to re-entrancy tags -- in fact there
are no constants at all.


>
Is there a paper I can read that treats
(co)indexing under this treatment?
>>


Not unless somebody besides me wrote it, which is doubtful. If I were
to write such a paper, it would have what I said above as the main
idea. This idea in itself doesn't sound very exciting, but the work
would explaining what the connection is between the syntactic indices
and semantics (for example, why it is that if two referring NPs are
coindexed then they corefer, but coreferring NPs need not be
coindexed; and how you get coindexing of a pronoun with a quantified
NP to correspond to variable binding in the semantic logic). Many
people would probably say: this is a bad idea, coindexing is semantic,
and indices are therefore otiose. Perhaps such a concern underlay
Martin Jansche's query (to the effect "what's this got to do with
syntax") -- is that right, Martin?

Carl