Predicate Clefts in LFG [Warning: long and possibly boring]

[Ash Asudeh]

Hello Seth and LFG-listers,

I'm not sure that I have anything terribly enlightening to contribute
and I'm not very familiar with the kind of predicate clefts that you
asked about, but I have a few things I wanted to mention.

First, I think the "resumptive pronoun" theory of predicate clefts is
not too promising. The only similarity between resumptive pronouns and
the kinds of clefts that Koopman (1983) has in mind is that there is
something at the top of an unbounded dependency and rather than a gap at
the bottom there's also "something".

Here's a couple of  examples from Koopman (1983). I can't reproduce the
tone accents.

1) ngOnU n   wa   na n   ka    ngOnU a
   sleep you want NA you FUT-A sleep-Q
   'Do you want to SLEEP'
   (Koopman 1983:154,(2a))

2) mlI   wa   mlI
   leave they leave
   They LEFT.
   (Koopman 1983

The reason that Koopman relates this verb doubling to WH-movement is
that a) long movement is possible across a bridge verb, 2) the movement
is island-sensitive and cannot get out of WH-islands or Complex NP
islands (Koopman 1983:161-162). However, the hallmark of resumptive
pronoun dependencies is that they are *not* island sensitive. This is a
general claim about resumptives and it certainly seems to be true in the
majority of languages that have been investigated. There have been
attempts to derive resumptives by movement (the most recent is probably
Boeckx's dissertation), but these always have problems with the island
facts (and would presumably have problems with the very robust fact that
filler-gap dependencies show form-identity effects between the filler
and the gap site, e.g. "whom", whereas the binder of a resumptive
pronoun never bears the case features of the resumptive; see Merchant
2001).

So I guess we're in agreement that the resumptive pronoun explanation is
not promising, since you didn't try to take it up.

Second, although I think I understand the gist of your sketch of an
analysis, this kind of predicate clefting will be just as hard to
accommodate in LFG's theory of unbounded dependencies as it was in the
GB theory Koopman was working in and as it is on just about anyone's
current theory, I think (who knows what's out there, though!). If we were
trying to use outside-in functional uncertainty like in Kaplan and Zaenen
(1989), then the problem is that there's nothing to grab onto. If the
downstairs verb is contributing ^ = v as usual, then it bears no GF for
the outside-in uncertainty to terminate in. If we're using inside-out
uncertainty, as used in versions of LFG that have (some) traces (Bresnan
1995, 2001, Falk 2001), then we can find the predicate cleft by looking up
the tree, as in your sketch. However, I don't think that the standard
unbounded dependency function FOCUS is what we should be looking for.

This brings me to the third problem. If the predicate cleft bears the
FOCUS function or any function you might want to use in a standard
unbounded dependency in LFG, then there has to be some way of integrating
it into the grammatical representation in order to satisfy the Extended
Coherence Condition. The two standard ways to do this are by functional
equality of TOPIC/FOCUS with some GF (the method used in standard
filler-gap dependencies) or by anaphoric binding of some GF by the
TOPIC/FOCUS (the method that I argue in my thesis should be used for
resumptive pronouns). But then the problem of identifying the GF presents
itself again: the main verb is ^ = v, not (^ GF) = v.


Here's a sketch of how I would start to think about this. I guess it bears
the most similarity to the "do support" theory that you allude to,
although it is not literally do support at all.

Koopman (1983:158) observes that the basic generalization about which
verbs in Vata can be clefted is that "any verb with a base form may
occur in the predicate cleft construction". In particular, verbs that
lack a base form *cannot* be predicate-clefted. Notice that by "base
form" Koopman means that the root of the clefted verb can be the input
to morphological processes and that the clefted verb can therefore have
all kinds of morphology (including causative, applicative, etc.).
Furthermore, the clefted verb bears the segmental form of the cleft
target, but does not bear its tonal specification, taking only mid tone
(Koopman 1983:155). These facts strongly indicate that formation of the
predicate clefted verb is a lexical process, rather than any kind of
syntactic "copying".

Suppose the lexical process did the following:
1. Takes a base form and copies its segmental make up
2. Copies none of the syntactic or semantic information of the base form

So the gist is that there is a lexical process that takes verbs that can
form bases for regular morphology, etc.,  and copies their
morphophonological information (except the tones) and makes a set of
matching "dummy" verbs.

These verbs are then inserted into the f-structure in some discourse
function, call it VFOCUS. In order to integrate VFOCUS we identify the
PRED of the predicate cleft target with that of the predicate cleft via
functional uncertainty. So the following kind of equation sits on some
appropriate part of the c-structure for the predicate cleft:

(^ PRED) = ((VFOCUS ^) COMP* PRED)

So the VFOCUS dummy takes the PRED of the f-structure that contains the
VFOCUS or, again starting with the f-structure that takes the VFOCUS, the
PRED of an arbitrarily deeply embedded COMP.

Now this will look really weird to LFG people. However, recall that the
dummy verb that was the output of the proposed lexical process has no
syntactic or semantic information of its own.

So how does the island stuff follow? Well the standard ways of capturing
island effects in LFG are either by tweaking the functional uncertainty
path or by using off-path constraints. The equation above already captures
the Complex NP island facts, since NPs cannot be COMPS (presumably this is
true in the relevant predicate-cleft languages, too).  As for the
wh-islands, we would need an off-path equation. We could grab the solution
that Dalrymple (2001:148ff.) proposes for capturing lack of movement
across non-bridge verbs, with a feature LDD (long distance dependency):

(^ PRED) = ((VFOCUS ^)    COMP*       PRED)
			           (-> LDD) ~= -


Perhaps the wh-islands should really be captured in terms of FOCUS. The
details are not that important, given that I'm just brazenly throwing
ideas around. The overarching point is that by positing the lexical
process that creates "dummy" copies and by making this copy take the a
PRED value using functional uncertainty we have related the mechanism
governing the predicate cleft to the mechanism governing unbounded
dependencies of the filler-gap variety. Namely, the ones that are
island-sensitive.

Of course, there's a whole host of residual issues:

1. If a PRED can be shared, then Completeness may have to be tweaked so
that they are satisfied as long as the subcategorized grammatical
functions are found in some f-structure that has the PRED in question
(in this case the f-structure of the main verb). This would be
interesting, since it would be the analogous case for Completeness as
that raised by functional control for Coherence with respect to, e.g.,
reflexive binding. I personally advocate reducing Completeness and
Coherence to the resource-sensitivity of Glue Semantics, in which case
this is not an issue (although a lot of other cases would have to be
checked!)

2. Speaking of semantics, we need to figure out the semantics of the
predicate cleft. Unlike a resumptive pronoun which works more or less like
a bound variable, the predicate cleft seems to have "fancier" semantics.

3. Staying on semantics, Koopman shows that the predicate cleft can take
adverbial modification:

3)  ye    kpe    lagO  ye
    come  really rain  comes
    'It is really RAINING, isn't it'
    (Koopman 1983:156,(7))

So for a start we might want to identify the s-structure resource of the
predicate clefted "dummy" with that of the copied verb:

sigma(^) = sigma((VFOCUS ^) COMP*)

This would allow the adverbial inside the predicate cleft to take scope
over the cleft target. It also fits well with the fact that their PREDs
are shared.

4. What ensures that the inserted dummy is the lexically produced dummy
that matches the cleft target? How do we get (4) and avoid (5):

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

I think it would be reasonable to appeal to general recoverability
conditions here. Otherwise we could use some kind of feature to control
things


[ADD MORE RESIDUAL ISSUES TO TASTE]

Best,
Ash


REFERENCES

@PhdThesis{boeckx01,
  author = 	 {Cedric A. Boeckx},
  title = 	 {Mechanisms of Chain Formation},
  school = 	 {University of Connecticut},
  year = 	 2001,
  address =	 {Storrs, CT}
}

@InCollection{bresnan95,
  author = 	 {Joan Bresnan},
  title = 	 {Linear Order, Syntactic Rank, and Empty Categories: On
Weak Crossover},
  booktitle = 	 {Formal Issues in {Lexical-Functional Grammar}},
  crossref =	 {dalrymple;ea95},
  pages =	 {241--274}
}

@Book{bresnan01,
  author =	 {Joan Bresnan},
  title = 	 {{Lexical-Functional Syntax}},
  publisher = 	 {Blackwell},
  year = 	 2001,
  address =	 {Oxford}
}

@Book{dalrymple;ea95,
  editor =	 {Mary Dalrymple and Ronald M. Kaplan and John T. Maxwell
and Annie Zaenen},
  title = 	 {Formal issues in {L}exical-{F}unctional {G}rammar},
  publisher = 	 {CSLI},
  year = 	 1995,
  address =	 {Stanford, CA}
}

@Book{dalrymple01,
  author =	 {Mary Dalrymple},
  title = 	 {{Lexical Functional Grammar}},
  publisher = 	 {Academic Press},
  year = 	 2001,
  address =	 {San Diego, CA}
}

@Book{falk01,
  author =	 {Yehuda Falk},
  title = 	 {{Lexical-Functional Grammar}: An Introduction to Parllel
Constraint-Based Syntax},
  publisher = 	 {CSLI Publications},
  year = 	 {2001},
  address =	 {Stanford, CA}
}

@InCollection{kaplan;zaenen89,
  author = 	 {Ronald M. Kaplan and Annie Zaenen},
  title = 	 {Long-distance dependencies, constituent structure, and
functional uncertainty},
  booktitle = 	 {Alternative Conceptions of Phrase Structure},
  crossref =	 {baltin;kroch89},
  note =	 {Reprinted in \citet{dalrymple;ea95}}
}

@Book{koopman83,
  author =	 {Hilda Koopman},
  title = 	 {The Syntax of Verbs: From Verb Movement Rules in the
{Kru} Languages to {Universal Grammar}},
  publisher = 	 {Foris},
  year = 	 1983,
  address =	 {Dordrecht}
}

@Book{merchant01,
  author =	 {Jason Merchant},
  title = 	 {The Syntax of Silence},
  publisher = 	 {Oxford University Press},
  year = 	 2001,
  address =	 {Oxford}
}