| |
Abstract:
The literature on adjunct parsing points to a general
preference for high attachment of relative clauses in unbiased,
null discourse contexts of the type NP1-Prep-NP2-RC. This
preference seems to hold for various, typologically diverse
languages, although its strength appears to vary (e.g., Cuetos
and Mitchell 1988). For Dutch, Brysbaert and Mitchell's (1996)
results indicate a moderately strong high attachment preference,
modulated by lexical factors. However, Brysbaert and Mitchell's
results are reliable only for constructions in which the
attachment is disambiguated late (i.e., near the end of the RC).
Morphological gender of the relative pronoun, a potentially early
disambiguator, apparently is not picked up on by readers.
To investigate the impact of the length of the ambiguous zone
on RC attachment, I set up a self-paced reading experiment with
stimuli of the following type:
NP1-prep-NP2-RelPro-Vemb-PP-Vmatrix-X.
Number agreement on the embedded verb disambiguated the
attachment. The order of the embedded verb and prepositional
phrase was varied, so as to contrast early (RelPro-V-PP) and late
(RelPro-PP-V) disambiguation. The reading times confirm the
(moderate) high attachment advantage reported by Brysbaert and
Mitchell, and moreover show that it occurs in both late and early
disambiguation contexts.
To account for cross-linguistic variations in RC attachment,
Gibson et al. (1996) proposed a two-factor processing cost model.
RC attachment is hypothesized to be affected by Recency
Preference, which is presumed to be universal, and a
parameterized principle, Predicate Proximity. The strength of the
latter is assumed to be related to word order strictness, which
correctly predicts a strong high attachment preference in free
word order languages (e.g. Spanish). The model predicts
non-monotonic attachment preference curves in three-site
(NP1-prep-NP2-prep-NP3-RC) constructions. The middle site, which
violates both recency preference and predicate proximity, would
be strongly dispreferred. The length of the construction would
favor recency preference, so that irrespective of the strength of
predicate proximity, low (NP3) attachment is preferred over high
(NP1) attachment. These predictions were confirmed for English
and Spanish. In an attempt to replicate these results, a
questionnaire study was performed with three-site complex noun
phrases corresponding to Gibson et al's template. The strong
dispreference for middle attachment was confirmed. However, high
(N1) attachment was strongly preferred over low (N3) attachment.
To determine whether this result is task-dependent, I am
currently running an on-line (self-paced reading) study with the
same materials.
Brysbaert, M. and Mitchell, D.C. (1996). "Modifier attachment in
sentence processing: Evidence from Dutch."
>Quarterly Journal of Experimental Psychology,
49A, 664-695.
Cuetos, F. and Mitchell, D.C. (1988). "Cross-linguistic
differences in parsing: Restrictions on the use of the Late
Closure strategy in Spanish."
Cognition,
30, 73-105.
Gibson, E., Pearlmutter, N., Canseco-Gonzalez, E. and Hickok, G.
(1996). "Cross-linguistic attachment preferences: Evidence from
English and Spanish."
Cognition,
59, 23-59.
|
