empty heads in HPSG (was: status of words; HPSG and CG)

[bender at csli.stanford.edu]

Berthold Crysmann wrote
>
> Emily Bender wrote:
>
> [...]
> > In my study of the syntax of AAVE copula absence (see Bender 2001,
> > ch 3), I was grudgingly led to the conclusion that the facts require one
> > to posit either an (i.e., one particular) empty verb or a construction
> > that in some contexts is phonologically empty.
>
>
> At first blush, that sounds a bit strange to me: isn't it one of the
> advantages of constructiuons that you can assign properties to
> configurations, let's say a non-headed structure, whithout actually
> the need of assigning these properties to some empty head? Can you
> summarise the main reason for that move even within the
> constructional context?

I started the project with the conviction that the availability of
constructions as analytical tool would obviate the need for any
empty elements.  This is why I say I was led to the conclusion
grudgingly.

The argument depends on a constellation of facts, but here are
the highlights:

The first obvious construction to write is a subject-head construction
that allows predicative elements as the head.  This construction quickly
proves inadequate, as its possible to extract the subject of copulaless
clauses:

(1) Tha's the man they say in love.

Since I'm assuming no traces, there would be no subject for the
proposed subject-head construction.

The next idea I considered is a non-branching construction that
makes a finite VP out of a predicative phrase.  This one doesn't
have any problems with subject extraction, but it does with existentials
(2), the combination of ellipsis and not (3), and extraction of the
predicate (all of which would also pose problems for the first
constructional analysis):

(2) I know there at least SOMEBODY happy about this.
(3) They say they're best friends and shit, but they not. (attested)
(4) How old you think his baby?

To bring the existential sentence in (2) under the second
constructional analysis, we'd have to treat "at least SOMEBODY happy
about this" as a predicative phrase, against all the evidence for
treating NP and XP as two separate complements of be (when it's there)
in the existential schema "there is NP XP".

To bring (3) under the constructional analysis, "not" would have to be
a predicative element that can licenses ellipsis, which in turn means
that the "not" would take a predicative complement.  This would lead
to spurious ambiguity every time "not" occurs outside examples like (3)
(copulaless elliptical clauses).

Finally (4) is simply intractable on a traceless analysis of extraction.
There is no selecting head of the predicative phrase "how old" on
the constructional analyses, and the available traceless analyses of
extraction localize the bottom of the SLASH dependency in the selecting
head.

Emily