Comments: To: lfg at list.stanford.edu

[Tara W Mohanan]

On our long trip back home from the LFG colloquium in Grenoble, we (TM,
KPM, & AA) had fruitful discussions about linguistics, triggered by the
talks we attended. The value of a conference should be measured not only by
the talks presented but also by the discussions that they stimulate. By
this, the Grenoble conference was for us most valuable.

There is one particular discussion that we had which we would like to share
with the LFG group. This has to do with the status of the function COMP in
LFG. Our discussion was triggered by Judith Berman's paper on sentential
arguments in German, which proposes that what was previously thought of as
a sentential subject should be analysed as a COMP. The effect of this
proposal is that is that an NP argument can be a SUBJ, OBJ, or OBJ-th, but
its sentential counterpart is always a COMP.

We are persuaded that, while COMP might have been a necessary notion in
classical LFG, in current LFG with its further factorization of information
and enriched levels of representation, COMP is (a) unnecessary, and (b)
inconsistent with the Lexical Mapping Theory. In this message, we would
like to spell out these problems, and suggest a solution that we believe
will contribute to the simplicity and elegance to the framework of LFG.

PROBLEM 1

In recent years, many researchers within LFG have pursued and adopted the
idea that the stipulation of information about the syntactic
subcategorization of predicates is best stated in terms of an argument
structure, rather than in terms of grammatical functions as was done in
earlier versions of LFG. This has motivated proposals for linking theories
that relate argument structure to grammatical functions, so that
grammatical functions are to a large extent predictable from argument
structure. The most widely accepted of these theories is the Lexical
Mapping Theory (LMT).

This theory proposes that grammatical functions are decomposed into (or
classified by) two binary features, [=B1o] and [=B1r]. This feature system
divides the map of grammatical functions into four categories:  [+o, -r],
[+o, +r], [-o, -r], and [-o, +r], which correspond to OBJ, OBJ-th, SUBJ,
and OBL respectively.

Given LMT, then, the problem is the following: How does the grammatical
function COMP fit into the four-way classification of grammatical
functions? Should we complicate the feature system so that we can define
COMP as a grammatical function distinct from the other four in its
classification? This can be done, for instance, by adding a feature [=B1c] t=
o
the existing set of features. This is clearly not a desirable move.


PROBLEM 2

The distinction between OBJ and COMP at f-structure is systematically
duplicated at c-structure. For instance, the second argument in "Xu knows
the story" is an OBJ, but the corresponding argument in "Xu knows that the
earth is flat" is a COMP. Similarly, the third argument in "Xu told Zi the
story" is an OB-th, but the corresponding argument in "Xu told Zi that the
earth is flat" is a COMP.

The generalisation is that a non-subject non-oblique argument ( i.e., a
[+o] ) is an OBJ or OBJ-th if it is an NP, and a COMP if it is an S or
S-bar. This results in a redundancy; the same theoretical distinction is
represented at two levels of representation, that between OBJ and COMP at
f-structure and NP and S at c-structure. This duplication of information is
undesirable in any theory.

PROBLEM 3

A generalization about passivization in English in the 1982 LFG was that
only verbs that take an OBJ in the active form can passivize. However,
verbs that take a sentential complement can also passivize. An example
would be the following active-passive pair:

(1)     a.      Many aliens believed that the earth was round.
        b.      That the earth was round was believed by many aliens.

Given this alternation, if we analyze the sentential complement in (1a) as
a COMP, the generalization is re-stated as: only verbs that take an
argument realised as an OBJ or [-r, +o] COMP in the active form can
passivize. This is a strange statement.

A SOLUTION

The solution we would like to suggest is that we abandon the grammatical
function COMP, and treat sentential complements as bearing the grammatical
functions OBJ or OBJ-th. This way, there is no need to complicate the
feature system, because a four-way distinction provided by the current
system is sufficient.

If we make this move, sentential complements are simply OBJ/OBJ-th linked S
or S-bar at c-structure. The distinction between sentential complements and
NP objects, then, is not OBJ/OBJ-th vs. COMP, but NP vs. S. This solves
problem 2 as well. The distinction between sentential complements and NP
objects is made only at c-structure, and hence, there is no redundancy.

If we abandon the distinction between OBJ and COMP as distinct GF's, all
that we need to say about the generalisation in the third problem is that
only verbs that take an argument linked to OBJ in the active can passivize.
(We ignore here the problems of examples like "The bed has not been slept
in.")

RESPONSE TO POSSIBLE ARGUMENTS FOR COMP

What we have said above is that unless there is a strong reason to include
COMP in the inventory of GF's, it would be best to do away with this GF.
Now, a possible argument that one might advance in support of COMP is the
observation that clauses cannot appear as complements of prepositions:

(2)     a.   *  Xu talked about that the earth was round.
        b.      Xu talked about that story.

The contrast between (2a) and (2b) is occasionally cited as evidence for
COMP. We can account for (2a) by assuming that a preposition does not take
a COMP.

However, this assumption cannot be correct: a clause in topic position can
satisfy the grammatical function of the object of a preposition, as
illustrated by (3):

(3)     a.   *  The aliens talked about that the earth was round.
        b.      That the earth was round the aliens talked about.

If we assume that sentential complements bear the grammatical function
COMP, we can explain the unacceptability of (3a) by stating that
prepositions take only OBJ, not COMP. However, this makes the wrong
prediction for (3b), because the topicalized that-sentence which can only
fill a COMP function should be inconsistent with the grammatical function
required by the preposition.

=46urthermore, there are passives with sentential subjects, which don't have
an active counterpart, as illustated in (4):

(4)     a.   *  The aliens argued for that the earth was round.
        b.      That the earth was round was argued for by the aliens.

The generalisation, in other words, is not that prepositions do not take a
COMP at f-structure, but that they cannot take a sentential sister at
c-structure. Thus, the ungrammaticality of (2a) and (3a) has nothing to do
with COMPs.

Another potential argument for COMP are asymmetries of the kind illustrated
in (5) - (7):

(5)     a.      Zi knows the claim
        b.      Zi knows that the earth is round.

(6)     a.  *   Zi insisted the claim.
        b.      Zi insisted on the claim.
        c.      Zi insisted that the earth is round.

(7)     a.      Zi suppressed the claim.
        b.  *   Zi suppressed that the earth is round.

In classical LFG, one can account for such data by stipulating that "know"
subcategorizes for OBJ or COMP, "insist" does not subcategorize for an OBJ,
and "suppress" does not subcategorize for a COMP.

Now, this is a legitimate motivation within classical LFG, but not within
current LFG where the GF's associated with arguments are derived, not
stipulated. The subcategorisation frames of "know", "insist" and "suppress"
carry information about argument structure, but not about GF's. Hence the
solution in terms of stipulations that distinguish between OBJ and COMP is
no longer valid.

A possible strategy to account for (5)-(7) is to stipulate the grammatical
categories (NP, PP, S) as part of subcategorization. Another strategy would
be to derive this result from the lexical semantics of these verbs.
Whichever move we make, COMP is not the solution.


Alex Alsina, KP Mohanan, Tara Mohanan

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