coordination of unlikes

[Carl Pollard]

Hi Shuichi,

>
You wrote (several messages ago):

>I think in both LFG and HPSG the answer is roughly that the feature
>structure (f-structure for LFG, category for HPSG) of a coordinate
>structure is the set of categories of the conjuncts. Or at least
>the analog of `set' in some interpretation of higher-order logic.

I was wondering if it wouldn't be better to employ lists rather than
(the analog of) sets in this context.  More specifically, how about
saying
 (i)   that the value of the HEAD feature is not just a single "head"
       object but a list of "head" objects,
>>

Usually a list of length one, right?


>
 (ii)  that the HEAD value of a coordinate structure is the list
       obtained by concatenating the HEAD values of the conjuncts, and
>>

Could be right.

>
 (iii) that the VALENCE value of a coordinate structure is identical
       to the VALENCE value of each of the conjuncts?
>>

Too strong: what would be the VALENCE value of EITHER MEET OR BE in
the following example (due to Neal Whitman):

  He would like to either meet or be Michael Jordan.

assuming that MEET wants an NP[-PRD] complement.
>
If we do this, we will also have to assume that the MOD feature is not
a head feature but a valence feature,
>>

Then how would MOD values be transmitted from the lexical heads of
adjuncts to their maximal projections?

>
and so on, and I haven't really
thought through all the ramifications, but this analysis seems capable
not only of capturing coordination of unlikes but also of providing a
basis for developing an adequate account of sentences involving
"agreement with the nearest conjunct", namely sentences like "There
was a man and two women in the room." and "Either your brakes or your
eyesight is at fault."
>>

There are some bugs to work out, but your point is well-taken: simply
taking the category of a coordinate structure to be the set (or the
union, if the conjunct categories are already sets) of the categories
of the conjuncts, does not handle the `principled resolution' cases
(to use the terminology of Corbett 1983.

Carl