Gerunds (was Re: Filler-gap mismatches)

[Malouf R.]

Hi,

Dan Parvaz writes:
 > > Would someone kindly explain the meaning of the terms `lexical
 > > coherence' and `phrasal coherence'?
 >
 > Phew -- I thought it was just me!

No, don't worry, it's not just you.  In fact, I'd bet it's pretty much
everyone except Yehuda and me!

Here's the basic idea.  Take gerunds.  Gerunds are constructions which are
headed by some kind of non-finite verbal form plus a subject and object(s),
and which have the distribution of noun phrases. Inside, the various
dependents of the head can combine with it either like you'd expect in a noun
phrase, or like you'd expect in a non-finite verbal projection, like so:

	 Pat's visiting of Amsterdam surprised me.
	 Pat's visiting Amsterdam surprised me.
	 Pat visiting Amsterdam surprised me.

Interestingly, what you don't find (in English or in any other language that I
am aware of) is things like this:

  	*Pat visiting of Amsterdam surprised me.

It looks like you can get both the subject and the object combining with the
head in a verb-like way, or both in a noun-like way, or you can get the object
combining like a verb and the subject combining like a noun.  You can't get
the object combining like a noun and the subject combining like a verb.

So, why this asymmetry?  Well, if you adopt the Abney/Bresnan view of gerunds
as noun phrases headed by verb phrases (i.e., some kind of mixed projection),
then all the attested gerund types involve a single point of conversion:

	 NP[Pat's VP[visiting] of Amsterdam] surprised me.
	 NP[Pat's VP[visiting Amsterdam]] surprised me.
	 NP[VP[Pat visiting Amsterdam]] surprised me.

while in the unattested type the nominal and verbal parts of the projection
are not continuous:

  	*NP[VP[Pat NP[VP[visiting] of Amsterdam]]] surprised me.

Under the mixed projection approach, you can account for the cross-linguistic
pattern by what I called a _phrasal coherence constraint_ (PCC) ruling out
multiple points of categorial conversion.

On the other hand, in my book I argue for a different way of looking a
gerunds. Instead of thinking of them as mixtures of categories in one module
of grammar (c-structure), my claim is that you want to think of them as
mixtures of categories in different modules.  In particular, gerunds are
simply nouns in the phrase structure which project NPs in the usual way. But,
they are strange nouns in that they inherit at least some of their
subcategorization properties from verbs and so combine with some of their
arguments in a verb-like way.  To account for the cross-linguistic patterns, I
propose a _lexical coherence constraint_ (LCC), along the lines of:

	If a word inherits its SUBJ value from a verbal type, then it
	also inherits is COMPS value from a verbal type.

There's more to than that, but the details aren't important here.

Where do Arabic masdars fit in to this?  Well, in most languages the ways that
subcategorization properties get expressed phrasally cause the PCC and the LCC
to make exactly the same predictions.  But, in at least some varieties of
Arabic (according to Fassi Fehri (1993)) apparently you get constructions like
this (I apologize for the weird transcription -- it's been a few generations
since anyone in my family spoke Arabic):

	Paqlaqa-nii ntiqaadu        zaydin   'amran
	annoyed-me  criticizing.NOM Zayd.GEN Amr.ACC
	`Zayd's criticizing Amr annoyed me.'

This looks to be analogous to the English verbal gerund in that you have a
deverbal form which combines with its subject in a noun-like way and its
object in a verb-like way, and which heads something with the distribution of
a noun phrase.  But, there's a problem.  The subject, part of the nominal
portion of the projection, is stuck right in the middle of the verbal portion:

	annoyed.me NP[VP[NP[VP[criticizing.NOM] Zayd.GEN] Amr.ACC]]

This is exactly what the PCC predicts should never happen.  Under the LCC,
though, this is what you would expect to find.  In terms of its lexical
subcategorization properties, the nominalization here is no different from its
English counterpart.  Those lexical properties get expressed rather
differently in the two languages, but the differences follow from general
differences in the syntax of the two languages.  Nothing special needs to be
said about the nominalizations.

Now, of course, this argument all depends on what kind of phrase structure you
assume for Arabic NPs. One could (and many have) adopted a more abstract view
of the structure of these constructions which would make the predictions of
the PCC more difficult to interpret.  So, it's not at all clear that this
constitutes a real empirical argument for the LCC over the PCC.  But, at the
very least, it looks like an approach based on the PCC needs to say something
extra about constructions like Arabic masdars (and other similar
constructions).

Rob Malouf
malouf at let.rug.nl