on COMP

[Alex Alsina]

The day to day work has prevented us from getting back to the issue of
COMP.  It is very encouraging to see that most of the messages on this
topic in one way or another share our conclusion that COMP should be
abandoned as a theoretical construct.  This is the essence of Avery
Andrews's message, although the way he gets rid of COMP is different
from the way we suggested we should get rid of it.  Even Chris Culy's
message and Judith Berman's message indicate that they are not
completely opposed to eliminating COMP, but that they retain it in
their analyses for practical reasons.  Let us consider here how valid
these reasons are and whether there is a way around them.

Chris's argument for retaining COMP is that it is needed in order to
explain the fact that the that-clause argument of verbs like "find"
cannot be the subject of the passive form of this verb, as in the
following example.

(1) *That they could reach the cookies was found (by the children) 

Although Chris does not spell out the details of his analysis, he
indicates that there would be a feature, probably associated with the
clausal argument of "find" that would prevent this argument from
appearing as the subject.  In our view, although this feature might
appear to be the a-structure counterpart of COMP, in fact it is
independent of the notion of COMP.

An analysis incorporating such a feature might be as follows.  "Find"
would have an a-structure with two arguments, the second one of which
could be either [+c] (link to COMP) or [-c] (don't link to COMP).  The
former feature value would be chosen when the argument is an S at
c-structure; since it would be a COMP, it would not be a SUBJ,
explaining the unacceptability of (1).  The latter feature value would
be chosen when the argument is an NP at c-structure and it would allow
the argument to map onto either OBJ or SUBJ.  In contrast with "find",
verbs like "discover", "realize", "find out", etc. allow their
sentential argument to be the passive subject, so we have to assume
that at a-structure this argument is [-c].  If it were [+c], it would
link to COMP and would be prevented from being a passive subject.
Therefore, the sentential complement of "discover" and the like, even
though it is what would normally be called a COMP, would have to be an
OBJ.

So, the notion of COMP as is normally understood plays no role in
explaining the unacceptability of (1).  The effect of the diacritic
[+c] is just to ensure that the argument that carries this feature
does not link to the subject function.  It has nothing to do
with COMP.  In fact, it would be much simpler to say that the feature
[+c] is a feature stipulating "don't link to subject".  In this way we
would account for the ungrammaticality of (1) and we would dispense
with the grammatical function COMP. 

But before adopting the use of any such diacritic, we must explore
other alternatives.  In our previous message, we made a suggestion
that related the ungrammaticality of (1) not to the claim that the
that-clause is the subject of the sentence, but that it is the topic
of the sentence.  This assumes the proposal that in English a clause
bearing the subject function is a topic.  The reason why the clausal
argument of a verb like "find" (but not of verbs like "realize" or
"discover") cannot be a topic would follow from the idea that that
argument carries certain presuppositions that are inconsistent with
the presuppositions carried by the topic function.  "Find" with a
clausal argument carries the presupposition that the speaker did not
know that the proposition expressed by this clausal argument is true
prior to the event of finding.  The contrast in the following examples
supports this claim:

(2) a.	I have known it all along, but John realized that Bill was a
	kleptomaniac only last week.
    b. *I have known it all along, but John found that Bill was a
	kleptomaniac only last week.

Let us further assume that the topic function carries certain
presuppositions like knowledge on the part of the speaker that make
the topic function incompatible with the clausal argument of "find".
What the exact connection between presuppositions and the discourse
function of topic may be need not concern us here; the relevant claim
is that the clausal argument of "find" cannot be a topic.

We are comparing an analysis that claims that the clausal argument of
"find" cannot be a subject with an analysis that claims that the
clausal argument of "find" cannot be a topic.  They both seem to
achieve the same result.  The latter does it in a more roundabout way
by appealing to the assumption that a clausal argument can only be a
subject if it also is a topic.  So why should we prefer the latter?
For two reasons (maybe more): first, the former analysis resorts to a
syntactic diacritic, whereas the latter analysis tries to explain the
fact by appealing to (hopefully) independently motivated assumptions
about the structures involved, i.e., without diacritics; second, the
latter explains more data than the former.  How so?  Well, the latter
analysis says that the clausal argument of "find" cannot be a topic.
This includes the case in which this argument is a subject, because
then it would also have to be a topic, but it also includes cases in
which it is not a subject.  In other words, the latter analysis, but
not the former, predicts that the following sentences are
unacceptable, and we believe the prediction is correct:

(3)  a.   That he has shrunk, he just realized.
     b.  *That he has shrunk, he just found.

(4)  He could not reach the cookie jar, as he had shrunk,
     a.   which he had just realized.
     b.  *which he had just found.

In (3b), the clausal argument of "find" is a topic and, even though it
is the object of "find", the sentence is ungrammatical.  (4b) shows
the same with the clausal argument of "find" expressed as a relative
pronoun, and therefore bearing the topic function.

Thus, the analysis that relates the anomalies of "find" to the notion
of topichood is more general than the analysis that relates them to
the notion of subjecthood.  The ungrammaticality of (3b) and (4b)
cannot be explained by stipulating that the clausal argument of "find"
cannot be a subject, because it is not a subject in these examples.
Therefore, the analysis proposed here is not only more general, but
does away with a syntactic diacritic--a diacritic which, in any case,
has nothing to do with COMP.

Judith Berman notes:

> However working on extraction in German showed that the distinction
> between a nominal and a sentential argument is relevant at the level
> of f-structure and the function COMP/XCOMP is one way to make this
> distinction.

It is significant that Judith notes that we are dealing with a
distinction between a nominal and a sentential argument, that is, a
distinction that is formally expressed at the level of c-structure.
This distinction is relevant at the level of f-structure because it
has an effect on extraction, which is analyzed as a relation between
grammatical functions, or f-structures.  However, the fact that a
distinction at level of representation A has effects on another level
of representation, B, does it require copying the distinction from A
into B?  We believe we all agree that, in a framework such as LFG that
allows simultaneous accessibility of levels of information, this is
unnecessary and goes against the very spirit of the framework.  We
should exploit this unique feature of LFG and avoid duplicating
information at different levels.

We hope to have shown that dispensing with COMP is not only
conceptually desirable, but it is perfectly feasible, and may well
lead us to new and interesting insights.

Alex Alsina
K. P. Mohanan
Tara Mohanan