COMP and OBJ

[Joan Bresnan]

Thanks to Mohanan, Tara, and Alex for their excellent clarification of
the issues, which I agree with.  It helps me to see exactly where I
differ from their position:

[snip]

> QUESTION 3: Granted that (a) we need to distinguish between grammatical 
> functions and grammatical categories, and (b) we do not stipulate 
> grammatical functions in the subcategorisation of verbs, is it necessary 
> to distinguish between OBJ and COMP? 

[snip]

> 	Our own answer to question 3 is No. The primary argument 
> (Argument 1) that we advanced in our first message was this. If we agree 
> that (a) grammatical functions are not stipulated but are 
> inferred/calculated from argument structure, and (b) LMT is good theory 
> for calculating grammatical functions from argument structure, then the 
> distinction between object and complement is illegitimate, because the 
> feature system in LMT ([r], [o]) does not provide for this distinction. 

I do not agree with (b).  LMT may be an inadequate theory for calculating
grammatical functions from argument structure, precisely because its
feature system does not yet provide for complement functions.  For
example, entirely aside from the question of sentential complements,
LMT (defined in terms of +/-o, +/-r or near equivalents) does not
provide for predicative complements (XCOMPs in early lfg notation):

1a) Mary looks [proud of herself].
 b) Jogging kept Susan [in a bad mood].
 c) Henry became [his cook].
 etc.

I have argued (in lecture notes, so far) that this predicative
complement is distinguishable from other functions (such as OBL) in
its anaphoric patterning and other grammatical properties.  (However,
see Zaenen and Engdahl 1994 [Descriptive and theoretical syntax in the
lexicon. COMPUTATIONAL APPROACHES TO THE LEXICON, ed. by T.T.S. Atkins
and A. Zampolli, pp. 181-212.  Oxford UP] for an interesting extension
of LMT to XCOMPs.  The latter still differ from OBL, in their theory,
in being functionally controlled.)

> None of the supporters of COMP have refuted this argument. 

I think that the argument is refuted by showing the falseness of its
assumption (b).

>The seconary 
> argument (Argument 2) that we advanced was that if all COMPs are Ss and 
> all OBJs are NPs, then the distinction betwen OBJ and COMP is 
> redundant: whatever we need to say about COMP and OBJ can be stated as 
> S-OBJ and NP-OBJ. 
> 

This argument is also invalid.  Not all COMPs are S's.  There are COMPs
(by which I mean complements other than OBJ or OBL) which are APs,
PPs, NPs, and VPs--the predicative complements illustrated above in (1).
Of course, I am hopeful that a simple extension of LMT, perhaps along
the lines proposed by Zaenen and Engdahl or by Avery Andrews, could
prove adequate for sentential complements as well.

I share Mohanan, Tara, and Alex's concern about a proliferation of
functions and the desirability of a more unified theory of
complements.  But I believe that they have oversimplified the
empirical picture too much to be fully convincing.

[snip]