|
Abstract:
The attachment of a relative clause to an NP1-of-NP2
construction is the only documented instance of divergent
attachment preferences in different languages. Several explanations
have been proposed for this variability (Mitchell & Brysbaert,
1998, and references there). Some accounts propose tuning to local
usage frequencies, or setting an arbitrary parameter. Explanatory
approaches have ascribed low attachment to a Gricean principle of
ambiguity avoidance in languages with an alternative genitive form,
or to the lack of a prosodic break before the relative clause; high
attachment has been ascribed to anaphoric interpretation of the
relative pronoun, or to high activation of verbs in languages where
a verb and its arguments may be non-adjacent. None of these
proposals, however, is securely established.
We present here some off-line data from previously untested
languages which seemed likely, on the basis of informal intuitions,
to exhibit low attachment. Previously, the only languages shown to
prefer low attachment are English, Brazilian Portuguese and
possibly Italian. These have no obvious properties in common that
set them apart from Spanish, French, Dutch, German, Afrikaans and
other languages that exhibit a high attachment preference. Our data
do not solve this puzzle, but they may help to constrain the
possible hypotheses.
Twenty-four sentences with a relative clause in a complex NP
were presented with 28 unambiguous filler sentences, to be read
silently for comprehension and then aloud for recording. A
comprehension question then checked which noun the relative had
been attached to. The sentences were first composed in English and
then translated as closely as possible, preserving naturalness,
into the other languages.
Crucially, when these materials were tested on speakers of
English and Spanish, the results conformed to the standard
attachment preferences reported for these languages.
Some theoretically relevant properties of the newly tested
languages are listed below:
SWEDISH & NORWEGIAN: determiner final; alternative genitive
form; featureless complementizer, not relative pronoun; variable
prosodic breaks before relative clause; verb-object adjacency
common, but not required; usual preposition is
til(l)
(as in English "key to the door").
ROMANIAN: determiner final; no alternative genitive; case-marked
relative pronoun with agreement features; prosodic breaks before
relative clause sensitive to clause length; some adverbs may
separate verb and object; construction is NP1-NP2[+GEN].
While the facts for Swedish and Norwegian are consistent with a
Gricean account, this is not true of Romanian, for which a
syntactic or prosodic explanation appears more likely.
Reference
Mitchell, D. C., & Brysbaert, M. (1998). Challenges to
recent theories of cross linguistic variation in parsing: Evidence
from Dutch. In D. Hillert (Ed.),
Sentence Processing: A Cross-Linguistic Perspective.
New York, NY: Academic Press.
|
