| |
Abstract:
Considerable discussion in recent psycholinguistic research
has centered on the issue of when and how contextual information is
used in parsing ambiguous sentences. Interactive, constraint-based
approaches to sentence processing propose that contextual pragmatic
and semantic information, among other factors, can influence the
parser's initial syntactic decision. In contrast, serial processing
theories such as the garden path model, maintain that only
structural principles such as Minimal Attachment and Late Closure
determine the initial analysis. If this analysis turns out to be
incorrect because of contextual or other information, a process of
reanalysis ensues.
This paper discusses the results of an eye-tracking experiment
of gapping sentences in context. Gapping sentences are a recent
addition to the inventory of ambiguous constructions investigated
by psycholinguists (see Carlson, 1997, for a questionnaire and
prosody study of these sentences in isolation). A gapping sentence
such as (1) below
(1) John visited Sarah at school and Michael at home.
is ambiguous between two possible interpretations, namely, (a)
John visited Sarah at school and (he) visited Michael at home (a
form of VP-conjunction), and (b) John visited Sarah at school and
Michael visited Sarah at home (a form of IP-conjunction). Intuitive
judgments and Carlson's results (Carlson, 1997) suggest that, at
least for sentences in isolation, the VP-conjunction reading is
much easier than the other reading. VP-conjunction is also
structurally simpler, as it involves fewer non-terminal nodes than
IP-conjunction.
The experimental items consisted of 24 ambiguous sentences such
as (1) above, preceded and followed by a context biased towards the
VP-conjunction or the IP-conjunction interpretation. Unreduced,
unambiguous versions of the two conditions served as controls.
The results provide support for the garden path model and for a
modular view of sentence processing. Despite the preceding biasing
context, the ambiguous region and the region following it took
significantly longer to read in the ambiguous IP-conjunction than
the VP-conjunction conditions. These significant effects were found
in second pass (not first pass) reading times, which suggests that
at first pass the parser treats the two structures as the same,
i.e., it only computes one, simpler structure. Also, ambiguity had
a significantly stronger effect on the structurally more complex
IP-conjunction than the VP-conjunction condition, a result expected
by the garden path model (ambiguity effects are stronger on the
unpreferred than the preferred interpretation) but not by an
interactive view of sentence processing.
Reference
Carlson, K. (1997). "Processing of gaps in a DP-PP frame."
Unpublished paper, University of Massachusetts, Amherst.
|
