Relative clauses

[Helge Dyvik]

Dear colleagues,

Yehuda Falk asks about the analysis of relative clauses. Assuming, as
usual, that the relativized element is a TOPIC within the relative
clause, and that the clause itself is a member of the set of adjuncts
within the NP, what is then the nature of the coreference between
this TOPIC and the larger NP, he asks.

I reply not because I imagine I have any final answers, but because
we have reflected on the same problems in connection with the
Norwegian computational LFG grammar which we are developing in
Bergen. This is just a sketch of how we (and to some extent the
others within the ParGram project, although I take full
responsibility for what I am writing here) have been thinking so far.

In the Norwegian grammar we derive a separate semantic structure
(s-structure) alongside the f-structure. Assuming that an f-structure
represents grammatical properties of expressions, and an s-structure
semantic properties, it is not clear that the TOPIC and the
f-structure of the larger NP (f::NP) should share any (f-structure)
features, since it is not obvious that what they share can reasonably
be regarded as *grammatical* properties. A fortiori, they should not
be unified - not because this would lead to a circular structure
(what is theoretically wrong with them?), but because, as Yehuda
points out, grammatical features like CASE may not be the same. The
NP and the embedded TOPIC are not the same linguistic object. It is
arguably different in the case of control structures ("He wants me to
leave"), since we may claim that here the same linguistic object
("me") at once functions as the object of "want" and as the subject
of "leave" - without any case conflict, since there is no subject
case assignment in the infinitival complement.

Nor does it seem correct to let the TOPIC and the f::NP share the
PRED feature. What the two have in common is not so much
predicational properties as referential properties: we don't have the
same predicate, potentially modifiable in different ways in the two
structures to yield different referents; we simply have the same
referent, irrespective of the predicate used to pick it out. Or put
differently: shared PRED values are neither a necessary nor a
sufficient condition for coreference, and coreference is the issue
here.

Of course, one could interpret shared, i.e., token-identical, PREDs
as a way of expressing coreference (as opposed to just type-identical
PREDs), but that has the flavour of trying to do in a slightly
uncomfortable way in the f-structure what we really need s-structures
for. The representation of coreference should be independent of the
representation of shared predicational content, since those two
phenomena are independent.

What the NP and the TOPIC do have in common, then, is coreference -
and selectional properties. One could argue about the status of the
latter; we don't see them as grammatical properties and represent
them in s-structure under the attribute 'sel'. Furthermore, we
analyze selectional violations as suboptimal rather than forbidden:
an analysis which violates selectional restrictions comes out on top
unless there is an alternative analysis which does not violate them.
Referential indices are captured in the s-structure as values of
'ind', e.g., [type individual], and coreference is expressed by the
unification of 'ind' values. Thus, s::NP and s::TOPIC share values
for the attributes 'sel' and 'ind'. There is no problem expressing
these unifications - I don't quite understand Yehuda's last point
about the difficulty of inside-out designators from elements in sets.
We don't need functional uncertainty at all in this case, since there
is no long-distance dependency between the mother NP and the embedded
TOPIC - the long-distance dependency starts only there. But I may
have misunderstood something?


Helge Dyvik






--------------------------------------------------------------------
Helge J. Jakhelln Dyvik
Department of Linguistics and Comparative Literature
Section for Linguistic Studies
University of Bergen    	   Phone: 	  +47 55582261
Sydnesplass 7   	           Fax:   	  +47 55589354
N-5007 Bergen, Norway              E-mail: helge.dyvik at lili.uib.no
--------------------------------------------------------------------