Responses to the responses on COMP in LFG

[Tara W Mohanan]

We would like to respond to the responses on our posting on COMP.

Point A:
        Chris points out that COMP might be useful to prevent
passivization with certain predicates, as in (2):

(1)  The children found that they could reach the cookies.
(2) *?That they could reach the cookies was found by the children.
(3)  ? That they could reach the cookies was realised by children.

Even if we had the function COMP, we will not be able to use it in
current LFG because the function is available only as the output of
linking rules. The information provided by argument structure, which
is the input to passivisation, does not distinguish between objects
and comps. Hence the ungrammaticality of (2) will have to be sought
in some other distinction.

One might speculate that what prevents passivisation in (2) is that
the second argument of "find" in (2) is a proposition. This move faces
problems when it comes to the contrast between (2) and (4):

(4)  It was found (by the children) that they could reach the cookies.

If, following the received wisdom, we treat the clause "that they
could reach the cookies" in (4) as an extraposed subject, then we
should not block passivisation for clausal arguments in "found".

The interesting question is, does the contrast between (2) on the
one  hand and (3) and (4) on the other stem from the same source?
It probably does. Chris claims that the distinction between (2)
and (3) cannot be captured in terms of semantics since "find" and
"realise" are near synonymous. Notice, however, that "realise"
carries the meaning that the speaker knew prior to the act of
realisation that the proposition is true. That is, the children
discovered something that the speaker already knew.  In contrast,
"find" means the opposite. The children discovered something that
the speaker didn't know prior to the discovery.  Notice what happens
to "found out", which contains the inference that the speaker knew
it before the children discovered it:

(5)  The children found out that they could reach the cookies.
(6)  That they could reach the cookies was found out by the children.

A possible way of accounting for these facts is to adopt Bresnan's
proposal that clausal subjects are not in the c-structure position
occupied by subjects, but in the c-structure position occupied by
topics, and this topic is coindexed with the subject in f-structure.
The claim that sentential subjects are subject-topics is in harmony
with the semantic differences we have hinted at above, though we do
not know precisely how to exploit this distinction to construct an
explanation. However, our gut feeling that the solution lies in
semantics remains. (We are using the term "semantics" to include
discourse effects, presuppositions, etc.) Even if our gut feeling
turns out to be incorrect, the solution could not be located in
the COMP/OBJ distinction, for reasons stated above.

Point B:
        Chris points out that Bresnan and Grimshaw have demonstrated
that contrary to popular belief, prepositions do take sentential
complements:

(7)     We didn't talk much about whether or not our shoes were muddy.

We do not remember what the Bresnan-Grimshaw solution is, but the
contrast between that-clauses and wh-clauses as in (7) can be accounted
for if we assume that the wh-clauses in (7) are nominal (e.g. as in
headless relative clauses.). This suggestion ties in with point D below.

Point C.
        Helge makes a plea for COMP on the grounds that there are
well known differences between complement clauses and NP objects.
That there are differences between NP's and S's does not mean that
these differences should be captured in terms of grammatical functions
rather than grammatical categories.

Point D.
        Helge points out that in languages like Icelandic, Norwegian
and Chichewa, sentential complements have the same distribution as
objects, unlike those in English. This difference can be made to
follow from the assumption that English makes a distinction between
NP's and S's, while Icelandic, Norwegian and Chichewa do not make
this distinction. Once again, the difference between the two groups
of languages can be attributed to categories rather than functions.

        An alternative would be to assume a categorial differences
between the sentential arguments in the two types of languages. In
her paper "Category Mismatches", Bresnan examines the differences
in the behaviour of sentential complements in English and Chichewa,
and accounts for them in terms of categorial properties. She proposes
that sentential arguments are categorised as CP's in English but as
NP's in Chichewa. We could appeal to the same categorial distinction
in Icelandic and Norwegian as well. (Languages like Malayalam make an
 overt distinction between nominal and non-nominal clauses by incorporating
a pronominal suffix into the complementiser. A postposition can take
as its complement a clause with the promoninal affix, but not one
without it. May be what happens in Icelandic, Norwegian and Chichewa
is that this overt distinction between two categories of clauses
is lost.)

Point E.
        Helge claims that the problem of redundancy (the redundancy
involved in representing the same distinction at two different levels)
is not relevant. We agree that it is not relevant for establishing
descriptive entities, but it is very relevant for establishing
theoretical entities. A theoretical entity should be accepted only
if we cannot do without it, or if dispensing with it complicates
the grammar excessively.

        We (TMA) have shown that (i) COMP is a theoretical entity
is unnecessary, and does not simplify the grammar in any way, and
(ii) is inconsistent with LMT. We are not objecting to using it as
a descriptive label, although we would recommend "sentential complement"
or "sentential argument" which is more transparent and does not look
like a theoretical entity.

Alex, Tara, Mohanan

----
Tara W Mohanan, Dept of English & Literature, NUS