|
Abstract:
In this paper we present a systematic explanation for the
apparent variability in results of experiments on the role of
island constraints in parsing, and argue that the parser's limited
ability to posit gaps inside syntactic islands is precisely what is
necessary for parsing parasitic gap constructions incrementally and
accurately.
Notwithstanding numerous theoretical attempts to unify islands
under a single grammatical constraint, differences among islands in
structure and severity are important in several respects. We show
at least three related differences between two classes of islands:
those in which just one offending category (e.g. subject, adjunct,
filled-comp) is crossed (mild islands), and those in which more
than one offending category is crossed (severe islands).
(A) Previous experiments which show evidence for violation of
island constraints in parsing have focused on mild islands (e.g.
Freedman & Forster 1985, Kurtzman et al. 1990, B. Stevenson
1999), whereas experiments which show evidence for respect of
island constraints in parsing have focused on more severe islands
(e.g. Stowe 1986, Kluender & Kutas 1993, Traxler &
Pickering 1996, McKinnon & Osterhout 1996). We show results
from a grammaticality rating study that support this division.
(B) Furthermore, the partition of island types in (A) makes
sense in the light of parasitic gap constructions such as (3)
below, in which the perfectly-acceptable main clause object gap in
(1) is combined with the "bad" gap in (2) (unacceptable since the
wh-dependency crosses a subject island, a single offending
category), rendering the sentence acceptable - the offending gap is
'parasitic' upon the well-formed gap.
However, only mild gaps can be rescued- the severe first gap in
(4), inside two island-forming categories (a subject NP and a
finite relative clause) cannot be saved by the well-formed gap.
[These judgments have been confirmed by a grammaticality rating
study using 24 sets of sentences modeled on (1-4).]
(1) What did the plan to enlarge the building ultimately damage
___?
(2) *What did the plan to enlarge ___ ultimately damage the
building?
(3) What did the plan to enlarge ___ ultimately damage ___?
(4) *What did the plan that enlarged ___ ultimately damage
___?
(C) Parasitic gap constructions like (3) pose a challenge for
incremental grammatical analysis, because they involve dependencies
whose well-formedness cannot be determined until later in the
sentence. Fully incremental analysis requires that the parser posit
a gap inside a subject island, then search for a licensing gap
later in the sentence. The parser should limit such island
violations to the mild islands which can support parasitic gaps, to
avoid positing impossible gaps as in (4). Results of a self-paced
reading experiment suggest that this is precisely what the parser
does.
Experimental materials were constructed in a 2 x 2 design, which
varied the finiteness of the embedded clause (infinitive vs. finite
relative clause) and the presence or absence of a wh-dependency
(what vs. whether), as illustrated in (5a-d). The critical embedded
verb (e.g. 'preserve') was always transitive. 24 sets of
experimental items were interspersed with 96 fillers.
(5) a. The outspoken environmentalist worked to investigate what
the local campaign to preserve the important habitats had actually
harmed in the area that the birds once used as a place for resting
while flying south. [infinitive, gap]
b. ...whether the local campaign to preserve... [infinitive, no
gap]
c. ...what the local campaign that preserved... [finite, gap]
d. ...whether the local campaign that preserved ... [finite, no
gap]
At the verb inside the complex subject NP ('preserve' in (5))
the infinitival conditions showed a significant slowdown in the gap
condition relative to the no-gap condition [F1(1,25)=5.18,
p<.05; F2(1,23)=5.58, p<.05].
|
