|
Abstract:
Atsu Inoue, Kantoo Gakuin University
Kamide and Mitchell (1997) (hereafter KM) investigated the
Japanese version of the familiar relative clause attachment (RC)
ambiguity with double NP heads. The materials were based on
Cuetos and Mitchell (1988). In Japanese the word order in the
complex NP is [RC + NP-Low+GEN + NP-High+CASE]. A questionnaire
study showed that readers prefer attaching RC to NP-High
off-line. An on-line self-paced reading experiment revealed that
RC initially attaches to NP-Low, followed by reanalysis to
NP-High attachment at the end of the sentence. We report two
further investigations addressing questions raised by KM's
study.
A. The initial low attachment preference.
The immediate NP-Low attachment preference might have arisen
because NP-Low was temporarily the only available attachment
site, especially since the genitive marker "no," indicating that
another noun will occur, appeared separately from NP-Low. If so,
this preference should disappear if "no" is presented together
with NP-Low. We conducted a self-paced reading experiment with
materials identical to KM's, but using [NP + case marker]
segmentation. In the High Attachment (HA) condition, RC was
pragmatically plausible with NP-High only; in the Low Attachment
(LA) condition, it was plausible only with NP-Low.
The results mirrored those of KM, though the effects were less
sharp. The early NP-Low attachment preference was apparent in the
numerical data but did not approach significance. The later
NP-High attachment preference replicated, but was significant
only in the analysis by subjects. A tentative conclusion is that
presentation mode matters, but the parser may have a mild
preference to associate a modifier to an NP in hand, even when
the existence of a later NP is signalled.
B. The final high attachment preference.
Both KM and this experiment make it clear that no high
attachment preference occurs until late in the sentence. The
source of this preference is of interest. Inoue and Fodor (1995)
reported native speaker intuitions suggesting that longer
modifiers have a greater tendency to attach high. To evaluate
this hypothesis, we re-examined KM's reading time data,
correlating relative clause length (in characters) with degree of
NP-High preference (LA minus HA).
Significant positive correlations, consistent with Inoue and
Fodor's observation, were observed at the NP-Low and NP-High
positions (where LA is overall preferred), and sentence-finally
(where HA is overall preferred). We consider two possible
explanations. First, a heavy adjunct might prefer high attachment
because the complex NP divides into two parts of roughly equal
weight, which is prosodically preferred (Kubozono, 1993).
Alternatively, it may be that discourse considerations (e.g.,
Relativized Relevance, Frazier, 1990; Predicate Proximity, Gibson
et al., 1996) are consulted early due to heavy processing load
for longer adjuncts.
|
