| |
Abstract:
This study examines the time-course with which listeners use
verb semantic restrictions to constrain the domain of reference of
upcoming constituents. In a recent experiment by Altmann and Kamide
(in press), listeners saw an array of images (e.g., a boy, a cake,
and several distractor objects). Subjects heard sentences such as
"The boy will move/eat the cake." Eye movements to the cake were
faster for "eat" than "move", suggesting listeners use verb meaning
to restrict the referential domain of an upcoming direct object.
Although these results are highly informative, the design of the
experiment makes it difficult to determine the time-course of
competition between objects with similar affordances. Incremental
models predict temporary competition between objects of appropriate
affordances prior to hearing the noun.
To examine this issue, we conducted an eye movement experiment
which differed from Altmann and Kamide in two ways. First, we used
an act-out task that required subjects to manipulate objects placed
in front of them. Second, we varied both the restrictiveness of the
verb and the number of competing objects in the array. Thus,
subjects were asked either to "fold the napkin" (a strong verb) or
to "pick up the napkin" (a weak verb) when the array contained
either one or two foldable items (a napkin or a napkin and a
towel).
To assess the time course of consideration of objects with
different affordances, we examined the proportion of time subjects
spent fixating objects during two time slices of the speech: one
corresponding to 233 ms after the onset of the verb (e.g., "fold")
until 233 ms after the onset of the noun (e.g., "napkin") and a
second corresponding to the next 500 ms. Because it takes 200-250
ms for the eyes to respond to phonemic input in this paradigm
(Allopenna, Magnuson, & Tanenhaus, 1998), any differences
during Slice 1 are likely due to the perception of the verb and not
the noun, whereas differences during Slice 2 are due also to the
influence of the noun. During Slice 1, when the verb was strong,
subjects spent more time fixating objects with correct affordances
than any other objects in the array. Moreover, in the strong verb,
competitor absent condition (i.e., when the target was the only
object with the correct affordances), the target was fixated more
often than any other object. When a competitor was present, by
contrast, fixations were distributed evenly between the competitor
and target. This resulted in a reliable interaction between verb
type and competitor presence (F1(1,12) = 5.65; p<0.05). The
proportion of time spent fixating the competitor was greatest in
the strong verb, competitor present condition, again resulting in a
reliable interaction (F1(1,12)=7.97; p<0.05). As expected, after
hearing the noun (Slice 2), target fixations increased in all
conditions, resulting in no significant interaction (F<1).Our
results extend those of Altmann and Kamide in two important ways.
First, there were almost no looks to incompatible objects following
a strong verb, indicating that listeners rapidly eliminated such
objects from consideration. Second, the presence of a competitor
object reduced to looks to the target, even when the verb remained
constant. This suggests that early looks to the target in the
Altmann & Kamide experiment were not due simply to the
informativeness of the verb (e.g., to "eat" being more informative
about its direct object than "move"), but to the restrictiveness of
the verb relative to the visual scene.
Allopenna, P. D., Magnuson, J. S., & Tanenhaus, M. K.
(1998). Tracking the time course of spoken word recognition using
eye movements: Evidence for continuous mapping models. JML, 38,
419-439.
Altmann, G. T. M., & Kamide, Y. (in press). Incremental
interpretation at verbs: Restricting the domain of subsequent
reference. In press, Cognition.
|
