[Prev][Next][Index]

expanded PPs

This week at the XTAG meeting we will talk about the expanded PP question
again. We decided a couple weeks ago that the argument PP that goes
with "put" type verbs would not be expanded out (P NP), but rather
would just have a PP substitution site (except in the wh tree where
the NP object of the preposition was extracted). After I sent out the 
meeting summary, Srini wrote and had some concerns about our decision.
Apparently, it had been decided in October or November of last year to
expand these PPs (where they weren't expanded previously). 

Carlos suggested that I sent out the messages that passed between me and
Srini to let people know what the issues are. So here it is:

NOTE: I ADDED A SECTION JUST NOW (THE PART IN CAPS) THAT WASN'T IN THE
ORIGINAL MESSAGE. I THINK THE PART I SAID IN THE ORIGINAL MESSAGE ABOUT
EXTRAPOSITION IS THE WRONG CASE, SO I ADDED WHAT I THINK THE RIGHT CASE IS
TO MAKE THE SAME POINT.

tonia

---------- Forwarded message ----------
Date: Tue, 3 Nov 1998 11:17:16 -0500 (EST)
From: Tonia Bleam <tbleam@linc.cis.upenn.edu>
To: Bangalore Srinivas <srini@research.att.com>

from Srinivas:
> The thing that bothers me is more to do with my understanding of the
> relationship among the members of a tree family. If you stop at the PP
> node in the A_nx0Vnx1pnx2 tree then how is it that you can get at the
> NP_2 in the A_W2nx0Vnx1pnx2 tree (for the extraction tree). I am aware
> of the point of an exhaustive PP not being able to subsitute if the PP
> is expanded out, but this can be gotten around by other ways (possibly
> creating a new tree fam, in the worst case). 
> 
> The domain of locality for the "parent" elementary tree of a tree
> family should certainly include the complement-NP of the PP, since you
> need to expand it out for the extraction case.
> 
> I know this is a sticky point, but we had gone back and forth several
> times before. I am wondering if this time around, the arguments for
> the change were more conclusive.
> 

hmmm. i don't know that are arguments were more conclusive or we just had
fewer people who objected. :) thanks for your comments. this merits more
discussion. (luckily jason has to finish is master's thesis this week and
so won't be implementing anything right away!)

i'm not sure i see what the argument is. so let me see if i can say it
back to you. because the NP can be extracted, we want there to be access
to it in the elementary tree. allowing access just in the actual case of
the extraction makes our representations inconsistent? because in
principle, anything extractable should be present in the elementary tree.
is this approximately it?

the thing is that having all the PPs articulated but then adding one tree
where the PP is not articulated (for the "there", "somewhere", etc cases)
seems just as inconsistent. I feel like it's a matter of what linguistic
reality you are trying to capture most generally: you either capture the
extraction fact generally and make an exception about the "there" case, or
you capture the fact that the PP acts as a unit and make an exception for
the extraction case. (although i don't really see the extraction case as
much of an exception, since we need a separate tree for extraction anyway,
it doesn't matter if it looks slightly different. on the other hand, the
alternative, which is adding another tree just for the "there" case, does
seem like a big exception).  

if you allow the articulated structure in general, you are also predicting
that the NP can "escape" in other ways. So it seems like you predict that
the NP could extrapose (heavy NP shift) without the whole PP. But this is
not the right prediction. ACTUALLY, I DON'T THINK THAT THESE ARGUMENT PPS
CAN EXTRAPOSE ANYWAY, SO THIS POINT DOESN'T REALLY GO THROUGH. BUT A 
RELEVANT CASE HERE IS TOPICALIZATION. SO I THINK IT'S THE CASE THAT
TOPICALIZATION IS POSSIBLE OF THE WHOLE PP BUT NOT OF THE NP. 

ON THE TABLE JOHN PUT THE BOOK
UNDER THE BED MARY PUT THE SUITCASE
*THE TABLE JOHN PUT THE BOOK ON
*THE BED MARY PUT THE BOOK UNDER

IF THESE FACTS ARE RIGHT, THEN IT SUGGESTS THAT THE PP ACTS AS A UNIT IN
THE SYNTAX. TOPICALIZATION HAS ACCESS TO THE PP BUT NOT TO THE INTERNAL
NP. (I DON'T UNDERSTAND WHY TOPICALIZATION AND WH-MOVEMENT WOULD BE
DIFFERENT THOUGH).

also -- it seems like we need something more general for extraction
besides just articulated structure given the fact that elements from
inside NPs can extract as well. (unless we want to say that the NPs should
have articulated structure in the elementary tree also). <- this maybe
seems a bit extreme. I do have an intuition that the internal structure of
PPs and NPs have a different status (maybe depending on the NP), but I
also see the PPs as being more of a unit just like the NP. 

Maybe i should put "more of a unit" into more formal terms. depending how
you see the PP, either the P is the lexical head, and the whole PP is the
theta domain and case marking domain of the P; or the N is the lexical
head and the PP is a functional projection of the N. Either way the PP is
the same kind of structure as an S or NP -- it should be an elementary
tree of its own. 

i guess it depends how you see these argument PPs -- is the NP really the
argument of the verb, and then the P is just some reflex of the
relationship between the verb and the noun (a dummy in some sense). or is
the whole PP the argument of the verb. i don't think there's an easy
answer to this question. my conception of it is that the whole PP is the
argument of the verb in cases like "john put the basket on the table" --
since the P is not fixed, it can be any locative P. Dative "to" seems more
like the first option -- just a reflex of the relationship, a dummy. In
exactly this case, we are leaving the "to" in the tree (the articulated
structure). 

This is all I can think of for now. Sorry if this is a bit rambly and
disorganized (it's off the top of my head).  Let me know what you think.

Tonia