| |
Abstract:
The complex NP sequence [Adjunct + N1-no + N2] in Japanese is
ambiguous with respect to height of attachment of the adjunct: it
may attach low to the genitive-marked N1, or high to the head
noun N2. Kamide and Mitchell (1997) found that after initial
attachment on-line to N1, the preferred attachment of a relative
clause is to N2. Inoue and Fodor (1995) had noted on the basis of
native speaker intuitions that this is so for adjective phrases
(APs) also, if they are 'heavy' (adverb + adjective, e.g.,
*kyokutanni shinsetsuna* "extremely kind") but not if they are
light (a single adjective).
We present self-paced reading data, from an experiment modeled
on Kamide and Mitchell's, which support this judgment. Two-word
and one-word APs were presented in a sentential context where
they were either compatible semantically with both N1 and N2
(hence globally ambiguous), or compatible with N1 only (forcing
low attachment). For the two-word AP but not for the one-word AP,
reading times for the N2 segment (including its case marker) were
significantly higher in the forced low attachment condition than
in the global ambiguity condition. The interaction (AP length x
sentence ambiguity) was also significant.
There has been much recent debate (summarized by Mitchell and
Brysbaert, in press) on why relative clauses favor high
attachment in right-branching languages such as Spanish. An
effect of constituent length (or weight) on height of attachment
is not explained by most models of on-line phrase structure
assignment (or construal). But it is reminiscent of length
effects noted by Kimball (1973) and Frazier and Fodor (1978).
Inoue and Fodor (1995) and Fodor (in press) attribute these
length phenomena to prosodic phrasing (even in silent reading;
cf. Bader, in press), which is known to favor evenly balanced
phrases.
Kubozono (1993) studied the intonation contours for [Adjunct +
N1-no + N2] phrases, and found that when the adjunct is longer
than one word, a "uniformly left-branching structure is
intonationally neutralized with the symmetrically branching
structure." That is: even when the syntactic/semantic structure
is clearly left-branching, the intonation pattern is
characteristic of the structure [[Adjunct] + [N1-no + N2]]. This
contour would favor the high-attached-adjunct interpretation for
an ambiguous example. We propose that this is the source of the
high attachment preference for phrasal and clausal adjuncts in
Japanese (and possibly of comparable effects in other languages),
as well as or instead of the statistical, pragmatic, or
parametric influences discussed in the recent literature on Late
Closure.
|
