Given Dick & Chet's paper on null words, I wonder whether traces/
chains are thereby not necessarily beyond the pale any longer.
They would be fundamentally different in kind from the 'deletions'
(as I would analyse them) discussed in the paper, since they
would involve something like a dependency between two positions,
the import of the dependency being that the filler of the
one position is the filler of the other. But I wonder whether
now that the pandora's box of nonconcreteness has been opened,
the old arguments against traces/chains still hold the same force.
You might reply "But WG has a perfectly good mechanism for
handling extraction phenomena, so why add the complication of
traces?". To this reply, I suggest we compare the following
two models:
A. Standard current WG: dependencies may be 'above the line'
or 'below the line'. Formally this can be modelled by
treating 'above the line' as a relational type in its own
right (called 's-dependency' and 'subordinate' by me in
past work).
The essence of this view is that the grammar creates
two structures, a hierarchical subordination structure, and
a tangling dependency structure in which words' dependency
requirements are satisfied by other words, and imposes
constraints on the correspondence between these two
structures.
B. Alternative: standard dependencies are all 'above the
line' relations between positions (in which case there is
no need to distinguish dependency from subordination).
Positions can be directly associated with a filler. Or
they can be linked by a 'cofilled' relation to another
position -- handling extraction and control/raising. Or,
arguably, they can simply be unfilled -- handling such
things as "He ate __ (off the plate)".
The essence of this view is that a word's dependency
requirements create a set of positions, with hierarchical
structure falling out naturally from this. A separate
body of rules and representations governs how these
positions are 'filled'.
The advantage of A is that it requires no distinction between
position and filler. The advantages of B (besides being more
efficacious expositively and pedagogically) are that it
allows for the possibility of unfilled positions, that it
is conceptually more 'explanatory' (i.e. begs fewer questions),
and that by allowing 'traces', affords a way of representing
the nested dependency constraint on multiple extractions.
It's not clear to me that A is cognitively superior in the
sense of there being better extralinguistic analogues. What
are the analogues of the dependency--subordination distinction
(or above/below the line)? Contrastingly, extralinguistically
the distinction between a space/place and the occupancy of
the space is pretty straightforward: we can distinguish
between places that are occupied, places that are unoccupied
but owned, and places that are neither occupied nor owned.
--And.
|