belated response on pleonastics

[Mary Dalrymple]

I'd like to respond to Steven Schaufele's message about pleonastic
subjects and, more generally, arguments that don't bear a thematic
role.  

I think there are several assumptions that Steven makes that are at
odds with the usual assumptions of LFG.  The most central is the issue
about whether the f-structure represents the syntactic argument
structure of a verb, including the arguments that aren't assigned a
thematic role, or only the ones that bear thematic roles.  In fact,
the f-structure is a syntactic structure, and so syntactic arguments
that bear no thematic role are also represented there.  Of course,
there may also be other levels of semantic representation at which
pleonastic subjects and other nonthematic arguments are not
represented.

There is a standard notation for indicating that an argument is not
assigned a thematic rule by a verb: the argument is written outside
the angle brackets in the argument list.  I think the distinction
might come out most clearly if we look at the difference between a
raising verb like "believe" and an equi verb like "convince".
Both verbs have objects, but the object of "believe" is not assigned a
thematic role by the verb "believe" -- though it may be assigned a
thematic role by the complement verb.  Here is the f-structure (very
abbreviated!) for "Bill believed it to rain":

[ PRED `believe<SUBJ,XCOMP>OBJ'
  SUBJ  [PRED `Bill']
  OBJ   [FORM IT] ----------+
  XCOMP [PRED `rain<>SUBJ'  |
         SUBJ  -------------+ ]

The line between the OBJ of `believe' and the SUBJ of `rain' indicates
that the f-structure for `it', with IT as the value for FORM, fills
both positions: that is, "it" is both the SUBJ of "rain" and the OBJ
of "believe".

The f-structure for "Bill convinced it to rain" is incomplete and thus
ill-formed, since "convince" requires an OBJ with a PRED value, but
pleonastic "it" does not have a PRED:

INCOMPLETE/ILL-FORMED:

[ PRED `convince<SUBJ,OBJ,XCOMP>'
  SUBJ  [PRED `Bill']
  OBJ   [FORM IT] ----------+
  XCOMP [PRED `rain<>SUBJ'  |
         SUBJ  -------------+ ]

The f-structure for "Bill believed George to leave" is well-formed,
since "leave" requires a SUBJ with a PRED value, and "George" has a
PRED with value `George':

[ PRED `believe<SUBJ,XCOMP>OBJ'
  SUBJ  [PRED `Bill']
  OBJ   [PRED `George']-----+
  XCOMP [PRED `leave<SUBJ>' |
         SUBJ  -------------+ ]

(I hope these f-structures are at least moderately readable.)

Now I'll go on to a discussion of Steven's points.

    Steven:
    > In point of fact, the 
    > argument that pleonastic subjects of weather verbs must be represented in 
    > f-structure i find particularly unconvincing.  Nicolas Ruwet (1991, caps. 
    > 3-4) has already presented extensive, thought-provoking, and (to me) con- 
    > vincing arguments that such pleonastic subjects (in those languages which 
    > have them) are (1) devoid of semantic content and (2), within an REST 
    > framework, bear no theta-role -- i.e., the typical weather verb has no 
    > theta-role to assign.  That being the case, only the Extended Projection 
    > Principle licenses the presence of such pleonastic subjects in languages 
    > like English and French.  Within a framework like LFG, they would presuma- 
    > bly need to be licensed by a general principle that requires subjects 
    > even when they're not called for by the verb's lexical entry, possibly by 
    > means of a `default subject assignment' on which more below.  (For vari- 
    > ant views on this issue, cf. Bresnan & Kanerva 1989, Bresnan & Zaenen 
    > 1990, and Mohanan 1990.  I, personally, have tended to oppose suggestions 
    > that an across-the-board requirement for subjects is a linguistic univer- 
    > sal, but i am perfectly happy to assume that such a requirement exists in 
    > some languages; after all, it manifestly exists in English and French.)

    > Furthermore, Ruwet demonstrates that the grammars of some languages (e.g. 
    > French), even when they require pleonastic subjects in the absence of 
    > `real' ones, make a clear grammatical distinction between the pleonastic 
    > subjects of weather verbs and the more `normal' subjects of more `normal' 
    > verbs.  (In fact, his arguments suggest that weather verbs are at one end 
    > of a continuum, with archetypal verbs at or near the other end and verbs 
    > like `seem', `appear', etc. somewhere in the middle, with regard to the 
    > character of their subjects.)

I agree that pleonastic subjects are not assigned a thematic role, and
that there needs to be a distinction between pleonastic subjects and
thematic (`normal') subjects.  In LFG, the distinction is spelled out
in several ways: whether or not the subject has a PRED value, whether
the verb assigns a thematic role to the argument, and when it does not
what FORM value is required of its argument.

    Steven:
    > That the pleonastic subjects of weather verbs participate in passiviza- 
    > tion as in (1b) and are obligatory even in non-finite contexts as in (2), 
    > as pointed out by Hudson, are not probative, in my judgment, at least 
    > within an LFG framework.  Part of the strength (or at least appeal) of 
    > LFG's approach, after all, is that it makes feasible the `lexicalization' 
    > of most of what used to be done syntactically in the hypothetical `trans- 
    > formational component'.  Thus, `passivization' is not (essentially) a 
    > `relation-changing' operation, converting an object (whether `raised' 
    > from a subordinate subject or not) to a subject, but a licensing of an 
    > alternative lexical entry of a verb; the difference in mapping schemata 
    > between predicate-argument structure and GF is merely one manifestation 
    > of the difference between two related lexical entries.

    > (1)	a. We expected it to rain.
    > 	b. It was expected to rain.

    > (2)	a. *(Its) raining during the picnic was a nuisance.
    > 	b. *(For it) to rain in mid-summer is quite normal in London.

The issue of how raising verbs and verbs with pleonastic subjects
participate in linking theory is a really interesting question, and
I'm afraid I don't really have anything to suggest on this front.  Are
there suggestions from other researchers who have worked on these
issues?

    > Thus, we have a lexical entry for `rain' that may (at most) make the 
    > stipulation in (3), and a lexical entry for `expect' that stipulates that 
    > it takes a clausal complement, and furthermore, should that complement be 
    > non-finite, carries the stipulation in (4); `Passivization' licenses an 
    > alternative lexical entry for `expect' with the stipulation in (5).  But 
    > in a sentence like (1b), that SUBJ is represented only by constraint (3) 
    > in the lexical entry for the verb `rain'; there is no reason for it to be 
    > overtly represented in f-structure.

    > (3)	SUBJ =(c) `it'
    > (4)	OBJ =(c) SCOMP(-fin) SUBJ
    > (5)	SUBJ =(c) SCOMP(-fin) SUBJ

This is absolutely true.  To be really detailed about it, the
equations would be:

    (3)	SUBJ FORM =(c) `it'
    (4)	OBJ = XCOMP SUBJ
    (5)	SUBJ = XCOMP SUBJ

But note that we are talking about constraints on f-structures here.
In order to talk about the SUBJ in an equation like (3), you have to
actually have a SUBJ in the f-structure.

I think these are the main parts of Steven's message that I wanted to
respond to.  Any reactions or comments would be welcome.

 - Mary