| |
Abstract:
Previous research suggests that the resolution of structural
ambiguities is influenced by referential presuppositions (e.g.,
Altmann & Steedman, 1988; Britt, 1994; Spivey-Knowlton &
Sedivy, 1995; Tanenhaus et al., 1995). For example, the PP in (1)
will be initially associated with the VP when a referent for the
definite NP 'the box' can be uniquely identified without further
specification, i.e., when the referential domain contains a single
box. When multiple boxes are available, NP-attachment is preferred
because an unmodified NP would fail to individuate a unique
referent.
(1) Unlock the box with the... [key. (VP-attachment)] / [blue
lid. (NP-attachment)]
In the present study, we used a head-mounted eyetracking
methodology to examine whether referential domains are constrained
on-line by the pragmatic requirements of verbs. For example, does
the verb 'unlock' limit reference to only those objects relevant to
the task of unlocking? If so, the typical garden-path effect
observed for NP modifiers in one-referent contexts, and eliminated
in two-referent contexts, should be reinstated when there are two
potential referents, but only one is compatible with unlocking
(i.e., two boxes with locks, one unlocked, one not).
In one experiment, participants heard instructions such as (1).
The corresponding visual display contained a locked box with a blue
lid (the target referent), a key and an unrelated object. The
fourth item in the display was either a locked or unlocked
non-target box (with a green lid), or another unrelated item.
Analysis of eye movement data revealed that when both candidate
referents were compatible with the verb action (e.g., two locked
boxes), many looks were made to the non-target referent in the VP
attachment, compared to the NP attachment condition. This reflects
the 'standard' referential effect in which VP-attachment results in
referential indeterminacy. However, this pattern reversed when only
one of the two candidates was compatible, reflecting greater
difficulty with NP-attachment. Crucially, the pattern and timing of
eye movements in this condition were similar when there was only
one referent (e.g., the one-boxcondition) for sentences with both
VP and NP attachments.
In a second experiment, we used materials with linguistically
unambiguous counterparts, as in (2):
(2) Pour the egg (that's) in the bowl over the flour
mixture.
The corresponding display contained two eggs (one in a bowl, one
in another container), an empty bowl, a container holding a flour
mixture, and unrelated objects. Either one or both eggs were in
liquid form. Consistent with the previous result, an initial
VP-attachment bias (as reflected by looks to the empty bowl
following the first PP) resulted only when one of the two eggs
could be poured, despite the availability of two egg referents.
These results highlight the role of referential domains in
structural ambiguity resolution and are argued to be consistent
with a processing mechanism whose incremental hypotheses are
defined by the rapid integration of semantic-conceptual constraints
with situation-specific information. The referential domain that
influences syntactic ambiguity resolution is updated dynamically,
taking into account listeners' goals and task-relevant properties
of potential referents.
Altmann, G., & Steedman, M. (1988). Interaction with context
during human sentence processing. Cognition, 30, 198-238.
Britt, M.A. (1994). The interaction of referential ambiguity and
argument structure in the parsing of prepositional phrases. Journal
of Memory and Language, 33, 251-283.
Spivey-Knowlton, M., & Sedivy, J. (1995). Resolving
attachment ambiguities with multiple constraints. Cognition, 55,
227-267.
Tanenhaus, M., Spivey-Knowlton, M., Eberhard, K., & Sedivy,
J. (1995). Integration of visual and linguistic information in
spoken language comprehension. Science ,268, 1632-1634.
|
