| |
Abstract:
Locality (or distance) of linguistic dependents has been
identified as a major source of processing variability in both
ambiguous and unambiguous constructions. Gibson's Syntactic
Prediction Locality Theory (SPLT, in press) and Hawkin's (1994)
theory, among others, provide metrics of processing cost that can
be used to derive detailed predictions for off-line acceptability,
as well as for on-line processing characteristics.
In a series experiments, we investigated the processing
characteristics of extraposed and non-extraposed German relative
clauses, dependent on the "length" of the (potential) NP-RC
distance and the "length" of the RC.
According to Gibson's (in press) SPLT, the cost of integrating
new items into sentence structures depends on the number of new
referential entities (NRE) intervening between the item and its
dependent, i.e., between (a) the verb (participle) and its
arguments, and (b) the RC (pronoun) and its host. In an off-line
magnitude estimation study, we found out that the number of words,
both in the RC and in the material intervening between the RC and
its host, but not the number of NRE determines the acceptability of
sentences with extraposed and non-extraposed RCs. This result
roughly confirms Hawkin's (1994) account and is further supported
by recent corpus studies (Uszkoreit et al., 1998). However, it is
still an unresolved issue whether or not the extraposition
preferences obtained in corpora and off-line rating studies can be
attributed to incremental on-line processing characteristics, as
suggested by Gibson and Hawkins.
We are currently running an eye-tracking study, (1) to
establish whether corpus data and off-line judgments correspond to
on-line processing characteristics, and (2) to determine the
precise location of processing (memory and integration) cost.
|
