Query on agreement

[Dan Everett]

I have a question about LFG that someone might be able to help me
with. Please be patient with me if the question only reveals my
ignorance of LFG. Here is my question: does LFG constrain possible
agreement relations configurationally or only functionally (i.e. based
only on grammatical function)? The empirical motivation for this
question comes from work I am doing on agreement in Arawan
languages. I will give an example of one of those here (English
tokens, though, to avoid all the glosses, etc. I can give real data to
anyone strongly interested).

In Banawa, the verb agrees in person and number with the subject and
in gender with the topic. Person and number surfaces as a fusional
verbal prefix. Gender surfaces as a verbal suffix.  The special
interest of Banawa verb agreement is seen when the topic NP is both
the subject of the clause and an inalienable possession
construction. In this case, the possessed NP governs person-number
agreement in the verbal prefix position, while the possessor NP
governs gender agreement in the verbal suffix position. That is, the
possessed NP behaves like the subject, the possessor as the
topic. This occurs, again, from subject position, not object
position. Moreover, it involves no overt movement. It is not,
therefore, possessor raising/ascension.

A hypothetical pair of examples:

(1) My[fem] brother[masc] 3sg-fell-masc.

(2) My[fem] hand[masc] 3sg-hurts-fem.

Ex. (2) shows that in an inalienable possession (and not in other NPs)
construction the possessed noun governs person-number agreement, while
the possessor governs gender agreement on the verb. Informally, it is
not hard to see that gender, which agrees with the topic, would agree
with the possessor since it is the possessor which is ultimately being
talked about or topicalized, not merely a part of his/her body or
inalienable possessions.

But formally, the situation is nontrivial. Simple feature percolation
won't get it. I have ideas on how to handle this in GB/Minimality theory,
but I am interested in knowing if any readers of this list find the 
structure problematic. I know that a rule in LFG format can be written
for this. That is not the question. I am interested in knowing whether
LFG constrains the kinds of rules that might be written so that writing
a rule for this construction is harder than writing a rule for
run-of-the-mill topic/subject agreement.

Thanks for any help you can provide.

Dan Everett