No Subject

It was proposed in the last two meetings that the structure that we currently
use for passive by-phrases does not capture relevant data. 

1. Athens was attacked by Sparta.

Passives as in (1) currently get the tree shown in (2):

2. 			  S
		        /  \
		      NP    VP
			   /  \
			  V    PP
			      /  \
			     P	  NP	

The problematic data are:

3. Athens was attacked quite often by Sparta.
4. Athens was attacked in the summer by Sparta.
5. Athens was attacked more often than Corinth by Sparta.

These sentences are problematic for our current analysis of the passive
by-phrase (or vice versa) because our tree does not allow for anything to
adjoin between the V and the by-phrase, and yet obviously things (PPs, adverbs,
etc.) must be allowed to do so.

Also, notice that extraction is allowable even when other elements intervene
between V and the by-phrase.

6. Who was Athens attacked quite often by ?
7. Who was Athens attacked in the summer by ?
8. Who was Athens attacked more often than Corinth by ?

9. Who was Athens attacked by ?

Assuming that (6-8) have the same level of grammaticality as (9), then this
suggests that both kinds of sentences should get a parallel analysis. Our
current grammar analyzes these in two different ways. In order to derive (9),
the by-phrase is present in the elementary tree. To derive (6-8), the by-phrase
is adjoined -- and then simulated multi-component TAG is needed to do the
extraction (something we need independently of just these examples at this

The idea is that in both cases (where material intervenes and where no material
intervenes) the by-phrase represents a (demoted) argument of the verb (and in
both cases extraction is possible), and so the two cases should get a
consistent analysis. 

So a major question is whether the by-phrase should be part of the elementary
tree or not. 

Here are arguments for both sides.

-The by-phrase is optional. This is the criterion that we use in other cases to
say that a PP should be adjoined rather than present in the elementary tree. 

-The by-phrase can be extracted from. Hence should be present in the
elementary tree. This is a good argument/test in theory, but we find cases of
adjuncts that allow extraction, and thus must allow a mechanism to extract out
of adjuncts. {However, I have a question about the nature of these so-called
adjuncts that allow extraction. Is there a list of these somewhere? are they
implemented in the grammar? Are they all examples of sources and goals -- which
have argument properties anyway? In the end we may decide that extraction
really is a good test and we'll add elementary trees which treat as arguments
all elements (PPs) that can be extracted from. }

-The by-phrase constitutes a semantic argument of the verb. 
If the passive is considered to be derived through a syntactic transformation,
then the by-phrase should be present in the elementary tree (to be consistent
with the way we represent transformations).

Even if passive is considered to be a lexical operation rather than a syntactic
one, the meaning or thematic role of the by-phrase seems to be present even
when the by-phrase is dropped. Tests for agentivity (purpose clauses like "to
collect the insurance") show that the passive sentence still has an agentive
subject of the action (represented at some level). Since the by-phrase realizes
this agent, then it is an argument and should be present in the elementary

I think this same view was expressed in a number of different ways. A similar
point was that if we want semantics to be reflected in the tree, then we should
have the by-phrase present in the el tree. (i don't think i'm doing this
justice). BUT a response to this is that the fact that the event has a subject
could be represented by features or variables in the tree nodes (see Matthew
Stone's TAG trees for SPUD). Then we could adjoin on the by-phrase, whose
referent could bind the variable. So semantics could be represented in a number
of different ways -- not just by nodes in the trees, but by features or
semantic formulae in the trees. 

I. One proposed analysis is the following:

10. 			  S
		        /  \
		      NP     VP
			   /    \
			  VP	 PP
			  |     /  \
			  V    P    NP	

Here, there is an extra VP node, thereby allowing adjunction of PPs, adverbs,
etc. between the V and the by-phrase. 

This structure looks like an adjunction structure (i.e., like the by-phrase has
been adjoined in) but would be an elementary tree. The question was whether we
want to introduce this new-fangled object into the grammar. 

Obviously we have to figure out what the theoretical import of introducing this
tree is. We would be saying that there are some things which we think are
"arguments" -- in this case because it was once the subject or _is_ the
underlying/semantic subject, and so we think it should be part of the
elementary tree, but at the same time we want to distinguish this PP from a
complement PP. Complements are sister to V. The by-phrase is not a complement
and therefore not sister to V. So we would be introducing a new structure to
deal with this single case (?) where there is a PP that is an argument but not
a complement of V  (nor a subject of V). 

II. Another way of solving the problem of elements adjoining between the V and
by-phrase is to allow adjunction to V. We may want this independently to
account for elements that can appear between V and complement PPs. (However,
the class of things that can appear between V and complements may be more
restricted than the things that can appear between V and by-phrase.) 

III. A third solution is just to always adjoin the by-phrase, and do
multi-component extraction when extraction is possible.

Summary from XTAG meeting on 10-29-98

I. We decided that the articulated structure of PP that now exists in the
Tnx0Vnx1pnx2 is unnecessary. So rather than having a P substitution site and an
NP substitution site, we will have a PP substitution site. 

The motivation for this is that we want exhaustive PPs like "there" to be able
to substitute into this slot. With the articulated PP structure in the
elementary tree, this was not possible.

The only place where the articulated structure is needed and thus will remain
is in wh-trees where the NP is extracted from the PP (this is so we can
co-index the wh NP and the NP position from which it was "extracted"). 

The Tnx0Vnx1tonx2 family will remain as it is with the articulated structure.

II. We need to implement a featural way to restrict ambiguities involving the
ordering of right-adjoining adverbials and auxiliary verbs (and their traces)
-- which both adjoin to VP.

III. The question arose whether "ahead of" is a multi-word P versus an adverb
(ahead) modifying a PP (of NP). I will look to see what past documentation says
about these -- what the tests are, etc.

IV. Can PPs left-adjoin to questions? Native English-speakers' judgments would
be helpful here:

1.	Instead of flowers, did John buy Mary chocolate?
2.	Instead of Sue, did John buy Mary chocolate?
3. 	Instead of flowers, what did John buy Mary.
4.	Ahead of Mary, did John finish the race?

other examples? do people know of conditions here that would determine