| |
Abstract:
This paper presents an argument against an approach to
sentence reanalysis frequently advocated by researchers assuming a
serial processing framework (e.g., Frazier & Rayner, 1982;
Frazier, 1987; Pritchett, 1988, 1992; Ferreira & Henderson,
1991; Lewis, 1993; Fodor & Inoue, 1994; and many of the papers
in Fodor & Ferreira, forthcoming). According to this approach,
in order to arrive at a new analysis, the parser manipulates the
structure which was built during the first analysis. We refer to
this type of reanalysis as "destructive reanalysis". We suggest an
alternative approach to sentence reanalysis, Generalized
Incremental Processing (GIP), which is couched in a parallel
processing activation-based framework (cf. Stevenson, 1994,
forthcoming) and which differs from destructive reanalysis in
assuming that reanalysis is identical to structure-building in the
first analysis. We also show that the arguments that have been made
against parallel-processing-based approaches to reanalysis do not
apply to the GIP proposal.
The general form of the argument is this: given the destructive
reanalysis hypothesis, the parser should be able to process some
locally ambiguous structures whose unambiguous counterparts are
effectively unprocessable. Specifically, the parser should be able
to bypass the high-processing-cost parse states within the target
structure by first pursuing a less costly structural analysis and
then destructively manipulating the structure it has built to
arrive at the target interpretation. To our knowledge, there exist
no examples of ambiguous sentences that are easier to process than
their unambiguous counterparts, suggesting that the destructive
reanalysis hypothesis is wrong.
Consider this general argument applied to a specific destructive
reanalysis theory. Fodor & Inoue (1994, forthcoming) assume
that reanalysis is cue-based, so that the nature of the reanalysis
cue completely determines the ease or difficulty of a particular
instance of reanalysis.
(1) The report that the employee stole ...
a. some office supplies amused the competition. (sentential
complement)
b. amused the competition. (relative clause)
For example, within the locally ambiguous structure illustrated in
(1), the sentential complement interpretation shown in (1a) is
initially preferred, but both interpretations are easy to obtain
(Fodor, 1985; Gibson, 1991; Pearlmutter & Mendelsohn, 1997).
Under F & I's approach, this shows that the reanalysis cue in
(1b) - the lack of a direct object following the verb "stole" - is
a good reanalysis cue, one that makes it easy for the parser to
reanalyze the initial component of the clause to the relative
clause reading of (1b). The ambiguous sentence (2) is identical to
(1b), except that it contains an extra embedded relative clause.
(2) The report that the employee who the company just hired stole
amused the competition.
Under F& I's approach, the reanalysis in (2) is predicted to
be as easy as the reanalysis in (1b): the reanalysis cue is the
same in the two sentences, as is the way in which the initial
structure is manipulated during reanalysis. (Within this theory,
reanalysis difficulty is not affected by the distance over which
the reanalysis occurs. Other destructive reanalysis theories
propose that distance affects reanalysis difficulty. It can be
shown that these theories make related incorrect reanalysis
difficulty predictions.)
|
