prepositional adjuncts

[Van Eynde]

At 10.58 29/03/01 +0100, Colin Matheson wrote:
>> If semantically vacous prepositions are (marked) NPs they
>> should be linearized as NPs but they aren't. Of course one could
>> distinguish between marked and unmarked NPs but then one had three
>> classes instead of two.
>
>That always seems to me me to be an odd way to argue about linguistics
>- it may be formally more parsimonious, but to me there's plenty of
>evidence that `real' grammars make dozens of low-level distinctions
>like this.
>I analysed "of" in English partitives as a marker in my PhD thesis and
>in fact treated the definite article the same way.  You do
>subsequently need to make slightly different assumptions about the
>distribution of marked and unmarked NPs, but that seems absolutely
>fine to me, and *much* nicer than a class of PP into which loads of
>strange bedfellows are shoehorned (is that enough mixed metaphors? :)

Hi,

Judgements about nicety have to be handled with care, esp. when bedfellows
are concerned.

So, instead of reducing this to a yes-no matter, let me
introduce into the discussion
the distinction between syntactic and semantic heads.
That the two need not coincide has been argued by many.
Just to give an obvious example: the English auxiliary 'do'
is the syntactic head of a finite verbal projection, but
hardly qualifies as a semantic head, since it does not
contribute any content.

Similarly, a case-marking preposition may be
vacuous semantically, but still be the syntactic head of a PP.
Whether it is a syntactic head is not a matter of semantics, but
of syntax (of course). Hence, the relevance of strandability and
of extraposability. If one has the constraint that head daughters cannot be
extracted (as in HPSG), then strandability is a counter-indication for
a marker analysis. Similarly, if you have a language in which PPs but not
NPs are extraposed, then the extraposibility of a PP complement is a
counter-indication for a marker analysis.

Does this mean that in a language which allows preposition stranding
all prepositions have to be treated as heads of PPs? Certainly not,
for the conditions under which stranding is allowed tend to be quite specific
(depending on the language) and hence leave room for a non-head
analysis of combinations in which the preposition cannot be stranded.
Dutch, for instance, is a language which allows adposition stranding,
but the strandability is constrained by such specific conditions, that
there are many combinations in which the prep
could---at least in principle---qualify for a non-head treatment.
Similarly, Dutch is like German in that it allows extraposition of PPs
much more readily than the extraposition of NPs, but there are
combinations of a prep with an NP (in Dutch), which resist extraposition just
as much as 'normal' NPs, and in which the prep
could hence qualify for a non-head treatment.

In sum, whether the prep is the head or not should be determined
on the basis of syntactic (not semantic) arguments, and these are bound to
give
different results for different languages.
What is needed at this point is a set of criteria which---together
with strandability and extraposability---provide tests for deciding on
the proper analysis of particular cases. In 'On the notion 'minor
preposition'' (included in the online proceedings of HPSG-2000)
I have proposed four such criteria and applied them to Dutch.
I'd be happy to get some comments about them, and to discuss
applications to other languages.

Friendly greetings,

Frank Van Eynde