| |
Abstract:
MacDonald et al. (1994) argue that frequency biases
associated with preterite versus past-participle uses of
morphologically ambiguous verbs, e.g. "refused" in (A), play a role
in resolving structural ambiguities. Trueswell (1997) provides
experimental evidence from a word-by-word reading task supporting
the claim. This eye-movement study with reduced-relative garden
path sentences further explores the effect. Bias and sentence type
were manipulated. Three levels of BIAS (contrasting with two in
Trueswell) reflect relative frequencies (Francis & Kucera,
1982) of preterite versus past-participle uses of ambiguous verbs
in "-ed": e.g., the preterite is more frequent for "refused"; the
participle is more frequent for "presented"; and frequencies are
equivalent for "offered." 12 such triplets were alternated in 24
frames like (A).
(A) The employees refused plaques (but) were unhappy with
management.
presented
offered
Two levels of sentence TYPE, differentiated through the presence or
absence of "but", correspond to whether the ambiguous verb is a
main-verb or heads a reduced-relative. The six conditions were
incorporated in a counter-balanced design.
First-pass reading times from 24 subjects show an expected
significant effect of TYPE (garden path effect), with longer
reading times in disambiguating regions of sentences containing
reduced-relatives. There is a marginal interaction of BIAS and TYPE
(F1(2,46)=2.90, p=.065; F2 ns). Analyses of simple effects indicate
that garden paths are severe when verbs are biased toward preterite
use. However, degree of garden pathing does not differ for the
"participle" and "equi-bias" conditions. This is unexpected under
the MacDonald et al. hypothesis which predicts that processing
difficulty for reduced relative garden paths will decrease as
participle frequency increases.
The result can be explained by a weakly interactive model based on
the Referential Theory (Crain & Steedman, 1985; Ni, Crain,
\& Shankweiler, 1996) modified so that (a) BIAS determines
latencies at which syntactic structures become available for
evaluation and (b) structures requiring extensions to the mental
model take longer for the parser to adopt than those that do not.
Thus, if the preterite is more frequent than the participle, the
main verb structure becomes available first, and is integrated
quickly as it is consistent with the mental model. If the
participle is more frequent, the reduced-relative structure becomes
available first, but being inconsistent with the model integration
is delayed. While modifications to the model are underway, the
main-verb structure may become available to compete with the
reduced-relative. The results of parsing reflect both frequency
based ordering of availability of lexical items and time consumed
by adjustments to the mental model.
|
