| |
Abstract:
We investigated attachment preferences of adjacent and
extraposed relative clauses (RCs) following three-site NPs of the
form [Det-NP1-Det-NP2-Det-NP3] in German. Hemforth et al. (1997)
established an off-line NP3 over NP1 over NP2-preference for
adjacent RCs. This is compatible with Gibson et al.'s (1996)
results from Spanish and English. In the extraposed cases,
however, the preference pattern changes: readers chose more NP1-
than NP3-attachments; as with the adjacent RCs, NP2 is strongly
dispreferred.
The different preference patterns for adjacent and extraposed
RCs can be explained by assuming that RC-attachment is determined
by a combination of anaphoric and syntactic processes (Hemforth
et al., in press) that take place in independent modules. The
module which provides a solution first determines the analysis of
the ambiguous input. Whereas recency-based (syntactic) processes
attach the RC to the nearest site, anaphoric processes
preferentially bind the relative pronoun to the most salient
discourse referent (Garrod and Sanford, 1985). We assume that the
salience of referents maps onto activation patterns: the most
salient referent is also the most active one. We also propose
that, in addition to the natural decay of the activation of all
referents, non-central discourse referents (NP2, NP3) are
deactivated at clause-boundaries to free resources for the
processing of the embedded clause. This deactivation is assumed
to be even stronger before extraposed RCs (Walter and Hemforth,
1997). Given these and additional assumptions on parameters of
initial activation and decay (cf. Hemforth et al., 1997), the
following activation patterns for the three referents can be
computed:
(1) adjacent RCs: a(NP1)$\ge$a(NP3)>a(NP2). With the
activation of NP1 and NP3 being similar, anaphoric processes
cannot decide between the two hosts. Therefore recency-based
processes win the race resulting in an N3-preference. NP1, which
is more active than NP2, is chosen next.
(2) extraposed RCs: a(NP1)>a(NP3)>a(NP2). If extraposed
RCs are base-generated in an IP-adjoined position, recency is not
involved and the preference pattern is determined by the
activation of the referents only. If extraposed RCs are
IP-adjoined after movement, recency is involved, too. In this
case the stronger deactivation of NP2 and NP3 at the
clause-boundary renders the difference between a(NP1) and a(NP3)
large enough for anaphoric processes to override recency.
Our predictions are compatible with the off-line data found by
Hemforth et al. (1997). To test whether on-line preferences
follow the same pattern, we are currently conducting an
eyetracking experiment.
Garrod, S., and Sanford, A.J. (1985). "On the real-time
character of interpretation during reading."
Language and Cognitive Processes,
1, 53-59.
Gibson, E., Pearlmutter, N., Canseco-Gonzalez, E., and Hickok,
G. (1996). "Recency preference in the human sentence processing
mechanism."
Cognition,
59, 23-59.
Hemforth, B., Konieczny, L., and Scheepers, C. (1997).
"Modifier attachment: relative clauses and coordinations."
Unpublished Manuscript, University of Freiburg.
Hemforth, B., Konieczny, L., and Scheepers, C. (in press).
"Syntactic attachment and anaphor resolution: two sides of
relative clause attachment." In: M. Crocker, M. Pickering, and C.
Clifton, jr. (Eds.),
Architectures and mechanisms of language processing.
Cambridge: Cambridge University Press.
Walter, M., and Hemforth, B. (1997, September). "Relative
clause attachment and syntactic boundaries." Poster presented at
the 3rd conference on Architectures and Mechanisms of Language
Processing (AMLaP) in Edinburgh.
|
