|
Abstract:
Prior experiments on English using a head-mounted
eye-tracking system (Tanenhaus et al., 1996; Sedivy, 1998) have
provided evidence for immediate incremental processing of
adjectives and early integration of contextual information in
resolving temporary syntactic ambiguities. We report the results of
an experimental study of Russian scrambling constructions with
adjectives using this methodology, to monitor eye movements of 16
Russian-speaking adults who responded to spoken instructions. This
issue is especially interesting in Russian because it allows for
"Split Scrambling" in which the adjective can be separated from its
head noun by other intervening constituents.
We did this by presenting 24 experimental instructions
containing an adjective-plus-noun phrase scrambled together, as in
(1), or an adjective scrambled away from its head noun, as in (2).
Note that both adjectives and nouns are marked for gender and
case.
These sentences appeared in three different visual contexts
designed to alter the point of disambiguation in the speech. Each
visual context always contained a target object ("red car") and a
contrastive set member ("silver car"). A potential competitor
object was also present which was either the same color as the
target ("red swan") or a different color from it ("green swan"). In
addition, the competitor object had either a different gender from
the target ("swan+MASC") or the same ("squirrel+FEM").
In the Early Disambiguation display, there was no competitor to
the target in terms of color, and the point of disambiguation was
the onset of the adjective "red". In the Mid display, the
competitor differed from the target by the gender ending on the
adjective
-uju,
and in the Late, the target could not be identified until the head
noun "car".
The NonSplit/Split Type and Point-of-Disambiguation factors
resulted in 2 ( 3 design. Eye movements were scored from the onset
of the adjective. The analysis of looks to the competitor which
occurred anywhere in the trials where one was present showed that
subjects were likely to move their eye to it (mean 70%). The
analysis of the first look to the target (NonSplit: Early 86%, Mid
43%, Late 36%; Split: Early 100%, Mid 48%, Late 55%) showed a main
effect of the point of disambiguation, since subjects were more
likely to launch the first eye movement to the target in the Early
than Mid or Late conditions. The main effect of NonSplit/Split Type
showed that first looks were more frequent to the target in the
Split condition compared to the NonSplit one. At first, the
difference is surprising since the actual words are the same. The
effect can be explained if we assume that prosodic marking
naturally present on the adjective in the Split cases helps
listeners identify the target as a member of the only contrastive
set. Eye movement latencies showed a pattern similar to the one for
percentages of looks. Thus, this evidence suggests first that the
semantic content of the adjective has a large and immediate impact
on establishing reference of adjective+noun phrases in contrast to
gender. Second, an intonationally-marked contrast combined with the
pragmatically appropriate discourse model (contrastive set)
facilitates processing of complex Split Scrambling constructions in
Russian.
|
