Movement's just another word for nothing left to merge

Carl Pollard pollard at ling.ohio-state.edu
Wed Jun 1 21:27:55 UTC 2005


Thanks to Paul Kay and Tibor Kiss for pointing out some
errors and obscurities in my previous message. Let me
try again.

First, in the following passage, the fifth word "local" is a typo for
"nonlocal".


> If SLASH values inlcuded local information, then we might expect
> to find, among those languages that mark extraction paths, some
> languages that have different marking for different kinds of
> extractions,
> e.g. a language where a verb inflects one way if its complement
> contains an interrogative gap, another way for a relative gap,
> and yet another way for a topicalization gap. Thus this architecture
> (SLASH values having type local) predicts that this is not a possible
> dimension of variation for gaps. Is this prediction right?

What I meant was that if SLASH values contained NONLOCAL information,
then you might have a language like Chamorro but instead of just
registering on the verb that the complement has a non-null SLASH value,
you could have two different inflected forms of the verb, one that
selects an XP[SLASH YP[QUE ZP]} and another that selects an
XP[SLASH YP[QUE \empty]].


My other point was about parasitic gaps:

> Also, if SLASH contained SLASH information, we should expect
> examples like
> this to be good:
>
> (1)  * Without reading __, I don't know how many reports Kim filed __.
>        [my judgment]
>
> where the gap after READING is supposed to be parasitic on the gap
> after FILED. Cf.:
>
> (2) Without using the Axiom of Choice, I don't know how many cool
>     theorems Kim will be able to prove.
>
> The point of (2) is just to show that what makes (1) bad is not the
> long adjunct extraction, but rather the fact that it contains
> parasitic gap.

To this Tibor replied:

>
Let's assume that you are right, how does HPSG (1994) cover this issue? I
bet it doesn't. HPSG (1994) simply says that SLASH information is not
projected, it does not say anything about SLASHes appearing in phrases which
are linked through SLASH. In the present example, I guess you would have to
propose that a filler/topic has an empty INH|SLASH value to block it. Do you
agree?
>>


In fact, Fillers CAN have nonempty INH|SLASH values. But what is ruled
out is getting those values because the SLASH value on the head sister
of the filler says they have to be there.

E.g. these are good:

(3)a  This is the document that I can't remember

    [[how many photocopies of __]_i we need to send __i to the registrar].

(3)b  This is the abstract that I still haven't decided

     [[which version of _]_i to submit _i for the workshop.]


The point is that the gap in the filler (in (3)a) is not there because
the SLASH value of

   we need to send __i to the registrar

is asking for a filler with a gap in it, but is independent of it.

By contrast, the point of

(2) *[Without even reading __i]_j,  [I don't know

    [[how many reports]-i [Kim [[filed __i] __j]]]]

is that the gap in the VP [filed _i] is supposed to have licensed a
parasitic gap in the adjunct [without even reading _i] and then that
adjunct is long-extracted (so that the SLASH sent up from the trace _j
bears the specification [SLASH i]). If SLASH values are of type local,
there is no way for the gap in the upper filler to get linked with the
object gap.

[It would still be posssible for the gap in the upper filler to be
"accidentally" identical to the object gap, as an instance of
Hoehle's Problem, but this is a general unintended consequence
of the HPSG formalism, not part of the theory.]

Carl



More information about the HPSG-L mailing list