| |
Abstract:
It is now well established that reference to (visual world)
context can be achieved in an incremental, piecemeal manner, using
linguistic information to narrow down the set of available visual
referents as soon as that information first becomes available.
Thus, after a sequence such as 'the boy will eat the yellow cake',
eye movements will be initiated towards the target cake on 'cake'
if there are several yellow objects but just one is a cake; on
'yellow' if there are several cakes but just one is yellow
(Eberhard et al., 1995; Sedivy et al., 1999); and on 'eat' if there
are several objects but just one of them, the cake, is edible
(Altmann & Kamide, 1999). In this talk, we present recent data
using the same visual world paradigm (cf. Cooper, 1974) which
further address anticipatory processes in language comprehension.
The studies just mentioned demonstrate that the human sentence
parsing mechanism can project anticipated structures before those
structures are encountered, and that it evaluates those projected
structures against the (visual) context (cf. Altmann, 1999; Altmann
& Kamide, 1999). However, in these cases, the parser was in
each case anticipating a referring expression. In the studies we
report here, we ask whether the parsing mechanism can anticipate,
in effect, grammatical function. For example, according to the
referential hypothesis put forward by Crain, Steedman, and Altmann
(Crain & Steedman, 1985, Altmann & Steedman, 1988), if a
referring expression fails to identify a unique referent, the
parser will attempt to interpret subsequent material as a
restrictive modifier. Thus, on hearing 'the butterfly is next to
the man' in a context containing two men, the parser should
interpret what follows as providing restrictive information by
which to identify the intended man. In our studies, we ask whether
the parser does in fact anticipate such information at 'the
man'.
In Experiment 1, we presented participants with a visual scene
containing two men identical except for one distinguishing
characteristic (the colour of their hair, or of their briefcases).
A butterfly was next to one of these men. We recorded eye movements
as participants heard 'the butterfly is next to the man with the
brown hair' (participants had to perform a picture-sentence
verification task). We observed more movements to the
distinguishing characteristic than to any other part of the scene
before the offset of the noun 'man'. Of course, it could be argued
that the eyes will naturally be drawn to the distinguishing
characteristic because it is visually informative, in which case
these early eye movements to the distinguishing part of the scene
are neither anticipatory nor mediated by the language. However, in
a second condition, participants heard 'the butterfly is next to a
man'. In this case, the indefinite referring expression 'a man'
does not necessitate the identification of a unique referent. We
observed, at the offset of 'man', significantly fewer looks to the
distinguishing characteristic in comparison to the definite ('the
man') condition. Thus, looks to this part of the visual scene do
appear to have been mediated by the definiteness of the referring
expression, and the anticipation that reference would be made to
restrictive information.
In Experiment 2, we controlled for various potential confounds
identified in Experiment 1, which, for example, never included a
postnominal modifier in the indefinite condition. Thus, Experiment
2 contained, amongst other conditions, 'the butterfly is next to
the man with brown hair' and 'the butterfly is next to a man with
brown hair'.
|
