| |
Abstract:
The relative clause attachment ambiguity in (1) is ultimately
resolved in different ways cross-linguistically. Speakers of
languages such as Spanish (including French, Italian, Dutch and
German) tend to attach the relative clause high, to NP1, while
speakers of languages such as English (including also Arabic,
Norwegian, Romanian and Swedish) tend to attach low, to NP2. This
observation has been confirmed in several off-line studies that use
slightly different methodologies and materials. (1) Someone shot
[the maid]N1 of [the actress]N2 who was on the balcony. However,
languages that attach high off-line do not always exhibit such a
preference in on-line studies (Italian and French: De Vincenzi
& Job, 1993, 1995, Baccino et al., in press; Spanish: Gilboy
& Sopena, 1993, 1996). Likewise, low attachment preferences
observed off-line have not always been replicated in on-line
experiments (English: Carreiras & Clifton, 1993, Henstra,
1996). Critical differences among the existing studies make
comparison of the on-line data practically impossible. Two aspects
are particularly prominent: segmentation and disambiguation.
Presenting the complex NP in two frames (De Vincenzi &
colleagues; Gilboy & Sopena) could bias attachment away from
NP1. Disambiguating the attachment of the relative by relying
exclusively on semantic/pragmatic information, without syntactic
reflex (Carreiras & Clifton, 1993, in press; Cuetos &
Mitchell, 1988), might, on the other hand, tap post-syntactic
strategies. The present study compares attachment preferences in
English and Spanish. The materials, created in parallel in the two
languages, are ambiguous in the off-line (questionnaire) task and
disambiguated syntactically by using number agreement in the
on-line (self-paced reading) task, where the critical sequence, _N1
of/de N2_, is presented in one frame. This design minimizes
contamination by post-syntactic reading strategies, and the
segmentation prevents biasing attachment away from NP1. The
questionnaire results replicate, with a new set of materials, the
previous finding of cross-linguistic differences. Spanish speakers
prefer to attach high while English speakers prefer to attach low.
Yet while the questionnaire data support the claim of
cross-linguistic differences, speakers of the two languages
resemble each other a great deal in their performance on-line.
Results from the self-paced reading task indicate an initial low
attachment preference, in both languages. The impact of
post-syntactic reading strategies is clearly evident in subjects'
error rates in answering a subsequent comprehension question
directed precisely at the attachment of the relative. The data here
indicate that subjects in both languages are more accurate with
questions about sentences where the relative clause attachment has
been forced high, syntactically. The on-line behavior of Spanish
and English speakers is remarkably _similar_, and appears to be
guided by the same general principle: attach low, locally, to NP2.
These findings are difficult to interpret under any model of
sentence processing that incorporates a language-specific component
in the parser. The phenomenon to be explained, therefore, is not
the initial preferences of the parser (which seem to be
cross-linguistically invariant) but the conditions under which the
initial attachment can be shifted away from NP2. We shall present
evidence that the implicit prosodic segmentation that readers
impose on the Input is one determining factor in the ultimate
attachment preferences.
|
