Response (and thanks) to Ash [WARNING: Also Long and Definitely Boring]

[Seth Alfred Cable]

Hi Ash (and other LFG-Listers),

Your e-mail provided some excellent feedback, and I'm going to spend a
while studying your suggestions.  Actually, the idea you propose was what
I had roughly imagined, though your proposal is far more advanced than
anything I could have done on my own (I'm a novice at LFG, and still have
to turn back to Joan's book whenever I begin to think down LFG-lines).

There are, however, a couple things I want to say in response to the
points that you made.

The first is that, although I'm not very interested in the "resumptive
pronoun" theory of predicate clefts, this is more to do with its
cross-linguistic predictions and the status of the ECP than the nature of
resumptive pronouns.  The problem with Koopman's original proposal is that
it's too parochial to Vata.  It predicts that in languages where there is
no problem with government from the left periphery, one shouldn't have
predicate clefts.  That, however, is simply not the case (witness Yiddish,
Czech, Bulgarian).   Moreover, taking a further step back, the account
relies on the existence of "government" and "the Empty Catagory
Principle", both of which are dubious (both to myself and to the
Zeitgeist).

I say this because, contrary to your claim, there *is* a class of pronouns
in many West African Languages and West African Creoles which have,
perhaps wrongly, been given the name "resumptive pronouns," and which have
many (if not all) of the syntactic properties of traces.  For example, I
have been able to reproduce in my Yoruba informants all the judgments
which have been used by other authors to argue that resumptive pronoun
dependencies in Yoruba are island-sensitive and license parasitic
gaps.  Moreover, I have discovered that a Yoruba speaker tolerates the
Weak-Cross-Over configuration "Wh1...pronoun1...gap1" if and only if
"pronoun1" is in a position where resumptive pronouns are licensed (this
discovery is discussed in the paper I'm sending to you).  I argue in that
paper that this is further evidence that Yoruba resumptive pronouns behave
like Wh-traces.  Finally, there is the fact that, although Yoruba
resumptive pronouns cannot resume across islands, they do *not* freely
alternate with wh-trace (unlike the cases of "apparent resumption"
discussed in the work of Aoun and Li).

To my mind, this stands as strong evidence that the resumptive pronouns
seen in West African languages (though definitely NOT the ones found in
Irish, Hebrew and Arabic) are a kind of "allomorph" of wh-traces.  That
is, they are syntactically gaps, though they have phonological content.
Now, since this is clearly NOT the case for the resumptive pronouns found
in Irish, Hebrew, Arabic and (to a much lesser extent) English, I think
that the lesson this is teaching us is that our term "resumptive pronoun"
does not pick out a natural class.  We have to separate what's going on in
Western Africa from what's going on in Western Asia, and only after
looking at each individually ask "how might they be similar?"

The second thing I wanted to mention is the minor point that your proposal
might have to be tweaked so that the lexical process creating the
dummy cleft-verb *does* bear semantic information in some languages.  In
most languages, in fact, this "dummy verb" can be what is dominating the
argumental NPs of the sentence.  For example, in Brazilian Portuguese and
Yoruba, you can say things like "Read the book, Mary told me Bill read",
whereas one cannot say "Read the book, Mary told me Bill read the book".
Moreover, regarding the Hebrew predicate cleft, Landau 2003 makes the
following stunning observation.  Although in a Hebrew predicate cleft you
can "strand" some arguments of the verb within the main clause, producing
a structure like in (1), this has a very interesting restriction.

	(1) [Verb Arg1] Subject Verb Arg2.
	(2) EXAMPLE: "To submit the paper, I submitted to John"

The restriction is that [Verb Arg1] can be the focus field of the
predicate cleft IF AND ONLY IF Arg2 is optional.  In other words, if the
dummy-verb appears with one of its arguments, it has to appear with all of
them, up to optionality.  Nevertheless, the lower verb needn't appear with
any of its arguments; (3) is perfectly fine in Hebrew.

	(3) To submit the paper to John, I submitted.

Hence, if there is some weirdness to the equation
"(up PRED) = ((VFOCUS up) COMP* PRED)", it probably shouldn't be resolved
by appeal to the VFOCUS not having its own predicational information.  (To
be honest, I don't see the weirdness.  This was, in fact, exactly the
sort of constraint I had in mind.  I mean, I don't know LFG well enough to
see any strange or nasty consequences of this.)

I want to note here in passing that the facts regarding the Hebrew and
Yiddish predicate clefts raise interesting questions for the notions of
"Completeness" and "Coherence" found in LFG.  In both Yiddish and Hebrew,
the initial copy of the V bears infinitival morphology, suggesting that it
is an infinitive at some level (though there are some interesting
morphological complications in Yiddish).  Nevertheless, these "dummy-Vs"
can appear without *any* of their arguments, while "real infinitives" in
the language cannot.  Now, one might suppose then that these infinitives
don't supply any predicational information.  But the problem is that
although these dummy-Vs can always appear without their arguments, they
can also appear with some of their arguments.  In exactly these cases,
it's the "real", inflected verb down in the main clause which no longer
appears with its arguments.

I think these facts follow nicely from Ash's proffered constraint: since
the PRED feature of the focus V is identified with the PRED feature of the
lower V, if the higher V appears with ARG1, the lower V CAN'T appear with
that argument.  This was one of the consequences I was hoping to squeeze
out of the LFG account I was roughly imagining.

The last comment I wanted to make concerns Ash's "Residual Point #4".
Wouldn't your account rule out (5) because the inside-out constraint
requires the PRED of the dummy-V to be identical to the PRED of the lower
V (given the reasonable assumption that there are not perfect synonyms
in natural language)?  This is actually of sizeable importance to me,
because I thought that it would be a natural consequence of the LFG
approach I was imagining (and which you brought into existence) that no
language could have a predicate cleft in which the dummy-V can differ
lexically from the V in the main clause.

4)   X ... X
5) * X ... Y

(NOTE: actually, Hebrew does seem to allow clefts like in (5).  But, it
can be shown that these apparently "lexically distinct" verbs are actually
allomorphs of one another.)

I want to again thank Ash and you other LFG-Listers for the wonderful
support you've given me.  This is definitely the most stimulating ListServ
I've ever experienced.

Best,

Seth.