| |
Abstract:
A growing number of studies has explored cross-linguistic
differences in the processing of sentences containing a complex NP
of the structure NP1-of-NP2 followed by an ambiguously attached
relative clause (RC). English, among others, tends to attach the RC
low, to NP2, while the preference for Spanish, Dutch, French,
German, etc, is to attach high, to NP1. Several explanations have
been proposed for this variability (Mitchell & Brysbaert 1998).
Low attachment has, among other things, been ascribed to a Gricean
principle of ambiguity avoidance in languages with an alternative
genitive form (Frazier & Clifton 1996), while high attachment
has been ascribed to anaphoric interpretation of the relative
pronoun (Hemforth et all 1998). Fodor (1998) has proposed that this
cross-linguistic variation may be attributable to differences in
prosodic segmentation patterns among languages. In particular, she
proposes that attachment preferences may partly be determined by
taking into consideration the prosodic weight of the attaching
constituent relative to that of the potential host constituent.
The present study examines RC-attachment preferences in
Croatian, a language for which no such data have been reported to
date. Croatian is interesting in two respects: 1) it has an
unambiguous alternative genitive form (eg.(2) vs (1)); and, 2) in
addition to the standard relative pronoun KOJI (with agreement
features for gender, number and case), it also uses the featureless
complementizer STO; the two are virtually interchangeable. Data
from Croatian may, thus, help to constrain the possible
explanations mentioned above, notably the Gricean and the
anaphoric-binding hypotheses.
The materials for Condition 1 were adapted from Ehrlich (1999)
and they consisted of 16 globally ambiguous sentences with a
KOJI-relative following a complex NP and 28 filler items, to be
read as part of an off-line questionnaire. The materials for
Condition 2 were identical to those in Condition 1, except that STO
-relatives were used instead of the KOJI ones. To test the
predictions of Fodors prosodic balance account, the target
sentences in both conditions were divided equally between those
with short RCs and those with long RCs.
Overall, Croatian exhibits a mild high attachment (HA)
preference (61% for Condition 1). Interestingly, the STO-relatives
(Condition 2), although attaching marginally lower than their KOJI
counterparts, also exhibit high attachment overall (57%). In both
conditions significant length effects were obtained: in Condition
1, 71% HA for long RCs vs. 53% for short RCs; in Condition 2, 71%
for long RCs vs. 42% for short RCs.
Results seem generally incompatible with a Gricean approach, as
this account wrongly predicts a low attachment preference for
Croatian. The attachment-binding dualism on the other hand, wrongly
predicts an NP2 preference for STO-relatives (Condition 2), as no
anaphoric binding is involved in the case of a featureless
complementizer such that it should override syntactic recency.
Fodors account, according to which prosodically heavier
constituents attach higher than their lighter counterparts, appears
to be the only approach that can account for the robust length
effects observed in both conditions, though it does not explain why
Croatian is a generally high-attaching language.
(1) Pokusali smo nazvati kcerku politicarke koja voli zapjevati
u drustvu. (AMBIGUOUS)
Tried(1.p.pl.) phone daughter(ACC)
politician(f.sg.GEN) RP(f.sg.NOM) likes sing in company
We tried to phone the daughter of the
politician who likes to sing when in company.
(2) Pokusali smo nazvati politicarkinu kcerku koja voli
zapjevati u drustvu. (UNAMBIGUOUS)
politician(f.sg.POSS) daughter(ACC)
RP(f.sg.NOM)
We tried to phone the politicians daughter who
likes to sing when in company
|
